Add the matterfact connector to Claude to search investor podcasts and build and run analysis pipelines — without leaving the chat. Setup, tools, and FAQ.

TL;DR

• The matterfact connector is a remote MCP server that brings matterfact into Claude
• Two capabilities: search a corpus of investor podcasts (and pull transcripts), and author, validate, and run No-Code Builder analysis pipelines for any ticker
• Sign in once with your existing matterfact account — no new credentials, no copy-paste

What you can do

The matterfact connector gives Claude direct, governed access to matterfact:

• Podcast research — full-text search across an indexed library of investor podcasts and expert interviews, plus on-demand transcripts and key topics for any episode.
• No-Code Builder (NCP) pipelines — author, validate, save, and test-run analysis "pipelines": graphs of nodes that gather SEC filings, earnings calls, news, and web/deep research and produce a cited report for a given company or ticker.

You ask in plain English; Claude calls matterfact and brings back sourced answers, in the window you are already working in.

Requirements

• A matterfact account. You sign in with your existing login (the same one you use at app.matterfact.com) the first time you connect — there is no separate connector password.
• Access to the relevant matterfact products (podcasts, No-Code Builder). If a tool reports you lack access, contact your account team.

Add the connector

The server URL is https://mcp.matterfact.com. Authentication is OAuth 2.0 — Claude walks you through sign-in on first use.

Claude (web & desktop)

1. Open Settings → Connectors.
2. Choose Add custom connector.
3. Paste https://mcp.matterfact.com and confirm.
4. Click Connect and sign in with your matterfact account when prompted.

Claude Code

claude mcp add --transport http matterfact https://mcp.matterfact.com/mcp

Then run /mcp in a session to complete the sign-in flow.

Tools

Read-only (safe to call freely):

• search_podcasts, get_podcast_transcript, get_channel_episodes — find episodes and read transcripts.
• get_node_catalog, validate_pipeline — inspect the pipeline node types and dry-run a draft.
• list_pipelines, get_pipeline, get_run, get_run_metrics — review your pipelines and runs.

Write (create or run things in your account):

• save_pipeline — create or update a pipeline.
• test_run — run a pipeline and return the result.

FAQ

Why am I asked to sign in? The connector federates to your existing matterfact login, so your podcasts and pipelines stay tied to your account. You authenticate once via OAuth; Claude stores the connection, not your password.

I don't see podcast or pipeline results. Your matterfact account needs access to that product. Reach out and we will get you set up.

Is my data sent anywhere new? No. The connector reads and writes the same matterfact data you already use; Claude calls it on your behalf with your permissions.

Need help? Email support@matterfact.com.

Matterfact's podcast MCP server brings sentiment and whispers from a million investor podcasts into Claude and ChatGPT. Ask, and get sourced answers.

TL;DR

• Matterfact's podcast corpus is now available over MCP (Model Context Protocol)
• If you run research inside Claude or ChatGPT, you can pull sentiment and whispers about hundreds of tickers from millions of investor podcasts
• No new tab, no new login: ask in plain English and get sourced answers without leaving the Claude window you are in

Research podcasts from your favorite AI tool with Matterfact Podcast MCP

Most investors already work inside an AI tool. You probably draft your thesis in Claude and pressure-test it in ChatGPT, asking follow-ups until you are comfortable enough to pitch the idea. The limiting factor has always been what those tools know and the data they can access. They know the public internet up to a training cutoff. And even though they can scan the news, they still have no idea what a chip supplier let slip on a podcast three days ago, or how sentiment on GLP-1 manufacturers has turned over the past month.

That gap is full of alpha opportunities for investors who know how to harvest it, and it is the reason Matterfact exists. Today we are closing that gap inside the tools you already have open. We are releasing the Matterfact Podcast MCP server for Claude, ChatGPT, and any other client that speaks the protocol.

Chat with 120 million podcast episodes right inside Claude

A podcast MCP server is a governed connection between an AI tool and a podcast intelligence corpus. MCP, or Model Context Protocol, lets you set it up once so the model can call Matterfact directly. You ask a question in plain English, the tool reaches into our podcast corpus, and the answer comes back with the real quotes, the speaker, the date, and the sentiment behind them. There is no new tab to open, no new login, and no pasting raw transcripts into a prompt to summarize a single show. You can see every mention of your ticker or industry at once and get answers synthesized across thousands of episodes in real time.

One more interface to hone your edge

Investors are not going to consolidate work into a single app. In fact, the opposite is happening. Your research now spills across Claude, ChatGPT, a terminal, a few browser tabs, and a Slack channel, and the number of those touchpoints keeps climbing. Rather than fight that, we want Matterfact available at every one of them. The MCP server means our podcast intelligence is no longer a destination you have to visit. It is a tool your existing assistant can reach for the moment a question comes up, whether that is during a morning idea screen or halfway through writing a memo.

What you can ask from podcast MCP

A handful of queries that work from day one:

• What are the current whispers on NVIDIA's moat, and which operators are voicing them?
• How has sentiment toward GLP-1 weight-loss drugs shifted across healthcare podcasts over the last 90 days?
• Pull every mention of high-NA EUV adoption timelines from the past eight quarters of semiconductor podcasts.
• Find me the most contrarian take on Chipotle's unit economics.
• Summarize what a company's management has said in interviews that never made it into the filings.

Each answer is grounded in actual episodes with attribution, not the model's best guess from training data.

AI agents surface signal faster than transcripts alone

Plenty of tools will hand you a transcript, but the value is in the intelligence layer on top of it. Our podcast agent surfaces actionable signals across the entire corpus and sorts them by who said it and how the sentiment is shifting.

The server is live for Claude and ChatGPT now. Connect it, then ask a million podcasts the question you have been saving for an expert network. What would you ask first?

FAQ

What is the Matterfact Podcast MCP server?

It is a Model Context Protocol connection that lets AI tools like Claude and ChatGPT query Matterfact's podcast intelligence directly. You ask about sentiment, themes, or whispers in plain English and get answers grounded in real episodes, with speaker and date attribution.

Which AI tools does it work with?

We are starting with Claude, and any client that supports MCP, like ChatGPT, will be supported soon. More clients are expected as the protocol spreads across the AI ecosystem.

What kind of questions can I ask?

Sentiment shifts on a sector or ticker, contrarian takes, emerging whispers, management commentary from interviews, and thematic trends across thousands of episodes. Anything that lives in investor and industry podcasts.

How is this different from a podcast transcript tool?

Transcripts give you raw text. The MCP server gives you the analysis layer on top: who said it, when, how sentiment is moving, and which signals are early. It surfaces whispers before they reach the mainstream instead of leaving you to read for hours.

Is podcast data reliable enough for investment decisions?

Treat it as a leading indicator, not a terminal-grade dataset. It is the input you check before an expert call to shape your questions, not a replacement for verified fundamentals. Used as an early-signal layer, it is one of the cheapest edges available.

Gavin Baker wants a financial podcast agent. Here's what that means, why generic transcription falls short, and how portfolio-aware signal extraction changes research.

TL;DR

• Gavin Baker, CIO of Atreides Management, says the most useful AI agent he can imagine distills investment-relevant insights from podcasts
• A real financial podcast agent does far more than transcribe and summarize. It extracts signals relevant to your portfolio and cross-references them against your existing research
• The value is in surfacing the two minutes of a 90-minute conversation that could change your thesis

Gavin Baker just described the tool every PM needs

On his most recent appearance on Invest Like the Best with Patrick O'Shaughnessy ("Watts and Wafers," May 20, 2026), Gavin Baker made an offhand comment that caught the attention of every research team in the industry. He described his most useful AI agent as one that could distill potentially valuable insights from podcasts into a well-crafted summary. In other words, he was describing a financial podcast agent.

This is a $7 billion fund CIO disclosing, in public, the tool he wishes he had to generate alpha.

Gavin Baker on Invest Like the Best, "Watts and Wafers"

Gavin Baker, CIO of Atreides Management, on Invest Like the Best with Patrick O'Shaughnessy, May 20, 2026.

Baker went on to explain the math. There is roughly six hours of relevant podcast content produced every day in the financial world. From a professional standpoint, he feels he should consume all of it. Every time someone from OpenAI, xAI, Google, or a major allocator speaks publicly, he wants to hear what they said. But he simply does not have the time.

He is not alone. When we talk to research teams at funds of all sizes, this is the single most common pain point: there is an enormous amount of valuable, unstructured information locked inside financial podcasts, and no systematic way to get it out.

What is a financial podcast agent?

A financial podcast agent is an AI system that continuously monitors financial podcasts, extracts investment-relevant signals, and ties them to your portfolio and coverage universe. It goes well beyond transcription and summarization. It cross-references what guests say against earnings calls, filings, and your existing research, then surfaces only the moments that change your view. Think of it as an always-on research analyst that listens to everything so you do not have to.

The distinction matters. A transcription tool gives you text. A summarizer gives you a paragraph. A financial podcast agent gives you a prioritized, sourced brief tied to the names you actually trade.

Why transcription and summarization aren't enough

First, you cannot possibly monitor millions of podcast episodes yourself, so you need a system that looks across all the conversations happening at once. You may be tempted to just transcribe the podcasts and run them through an LLM for a summary. Several products already attempt this. The problem is that a simple summary of a financial podcast is next to useless to an analyst.

Consider a 90-minute conversation between a semiconductor analyst and a former TSMC engineer on a niche industry podcast. A generic summary would tell you they discussed capacity expansion, pricing trends, and competitive dynamics. That is what the episode description already says. It tells you nothing you did not know from reading the title.

What an analyst actually needs is the granularity and detail that translate into actionable insights. That is where the value is. They need to know: did the guest say anything about high-NA EUV yield rates that contradicts what ASML management said on their last earnings call? Did they mention a specific customer shifting orders to Samsung Foundry? Was there a throwaway comment about Intel's process timeline that has not shown up in any sell-side research?

That is the difference between summarization and signal extraction. Summarization compresses the content. Signal extraction connects the dots. A real podcast agent does not just tell you what was said, it tells you what matters, given what you already know and what you are working on.

The podcast as an alternative data source

Financial podcasts occupy a unique position in the information ecosystem. They sit between formal channels (earnings calls, SEC filings, sell-side research) and informal channels (social media, private conversations). Guests on financial podcasts routinely say things they would never put in a research report or an earnings transcript.

A former executive at a public company, freed from IR compliance, will speak candidly about competitive dynamics, internal culture, and strategic mistakes in a way that a 10-K never captures. A venture capitalist discussing their portfolio company's market will drop data points about customer adoption rates that will not appear in public filings for another two quarters. A macro strategist being interviewed casually will reveal their actual portfolio positioning rather than the sanitized version they present at conferences.

This is what makes podcast intelligence genuinely different from transcript search or document retrieval. The content itself is different. The candor level is higher. The signal quality, for specific use cases, is better than anything you would get from traditional sources.

We built Matterfact's podcast intelligence layer around this insight. The platform currently processes thousands of financial podcasts and lets analysts query the full corpus with natural language. You can ask, "What are the most contrarian views on GLP-1 medications from the last 30 days?" and get a structured response with source attribution, timestamps, and links to the exact moments in each episode. That is the podcast agent Gavin Baker is describing.

Why this niche is wide open and ripe for alpha

Search for "financial podcast agent" or "podcast AI for investing" and you will find almost nothing purpose-built for buy-side research. There are generic podcast summarization tools and broad financial AI platforms that include some podcast coverage. But no one has built a dedicated, portfolio-aware podcast intelligence agent and made it the centerpiece of their offering.

The reason is that it is technically hard. Processing audio at scale requires speech-to-text pipelines that handle financial jargon, multiple accents, and poor recording quality. Extracting named entities (tickers, people, companies, products) from conversational speech is a different problem than extracting them from structured text. Cross-referencing podcast claims against financial data requires a multi-source knowledge graph that most AI startups do not have.

But the demand and the value are clearly there. This is exactly the kind of inefficiency that rewards investors who get serious about it before everyone else does.

Try it on a podcast you follow

Name a financial podcast you listen to regularly. We will build you a custom artifact: a research dashboard that monitors every episode, extracts signals relevant to your coverage universe, and delivers a prioritized brief you can use.

FAQ

What is a financial podcast agent?

A financial podcast agent is an AI system that continuously processes large volumes of podcasts and extracts investment-relevant signals. Unlike simple transcription or summarization tools, a podcast agent connects insights to your portfolio, cross-references claims against other data sources, and delivers structured, actionable output tailored to your research needs.

Why do hedge funds use financial podcast agents?

Podcasts now carry some of the most candid, forward-looking commentary in the market, often before it shows up in filings or sell-side research. Hedge funds use podcast agents to monitor that firehose at scale, extract signals tied to the names they cover, and turn hours of audio into a prioritized brief, all without adding headcount.

How is podcast intelligence different from earnings call transcripts?

Podcast guests speak more candidly than executives on earnings calls. Former employees, industry experts, and investors share opinions, data points, and competitive assessments that would never appear in a formal filing or prepared remarks. Podcast intelligence captures this informal layer of market intelligence that traditional tools miss entirely.

Can AI really extract useful signals from unstructured podcast conversations?

Yes, but only with purpose-built systems. Generic LLM summarization produces generic summaries that are not actionable to investors. A system designed for financial podcast analysis needs entity recognition trained on financial jargon, cross-referencing against structured financial data, and the ability to weight sources by expertise and relevance. Matterfact's platform was built specifically for this use case.

What did Gavin Baker say about podcast agents?

On the May 20, 2026 episode of Invest Like the Best ("Watts and Wafers"), Gavin Baker, CIO of Atreides Management ($7B AUM), described the most useful AI agent as one that could distill potentially valuable insights from the roughly six hours of daily financial podcast content into a focused, relevant summary.

Live tracker of all 30 Sohn 2026 Investment Conference stock picks. The Core Index beat the S&P 500 by 1.20% in 7 trading days. See who's winning and who's bleeding.

TL;DR

The 31st annual Sohn Investment Conference took place on May 12 at Jazz at Lincoln Center, raising funds for pediatric cancer research at MSK Kids. We built an equal-weight index of all 30 pitches and tracked them for seven trading days. The Core Index of 16 main-stage picks returned +0.74% versus the S&P 500's -0.46%, generating +1.20% of alpha. The biggest winner so far is Vista Energy (VIST), up 15%. The biggest loser is Carvana (CVNA), down over 15%.

Every May, some of the best minds in finance walk onto a stage at Lincoln Center and put their reputations on the line with a stock pitch or two. They do it for a good reason: the Sohn Investment Conference, now in its 31st year, raises money for MSK Kids, the pediatric cancer program at Memorial Sloan Kettering Cancer Center. The conference honors the memory of Ira Sohn, a Wall Street professional who died of cancer at 29. Since its founding in 1995, the Sohn Conference series has raised more than $150 million for children's health research.

Sohn 2026 brought the usual roster of hedge fund headliners (David Einhorn, Joyce Meng, Soren Aandahl, Kevin Salimian, and a deep Next Wave bench) onto a single stage at Jazz at Lincoln Center on May 12. For the rest of us watching from the cheap seats, the natural question is: how are those Sohn 2026 stock picks actually doing?

We tracked all 30 stock ideas presented at the 2026 conference and built a simple scorecard. Here is what the first seven trading days look like. (If you want to see how hedge fund hiring patterns are shifting alongside these pitch trends, our hedge fund hiring breakdown and skills survey are good companion reads.)

The Scoreboard: Core Index vs. the S&P 500

We constructed two equal-weight baskets. The Core Index includes the 16 main-stage pitches (14 longs and 2 shorts). The Extended Index captures all 30 ideas, including the Next Wave rising-manager presentations and the short-selling panel moderated by Jim Chanos.

Through May 20 (seven trading days from the conference date), the numbers:

Not bad for a week. Keep in mind these are equal-weight baskets, so a single name can drag the average, and short positions contribute the inverse of the stock return (a short that drops 10% adds 10% to the index). Both baskets jumped out to a strong lead by May 14, then gave back gains as the broader market softened. The Core Index held up better than the Extended basket, which tells you the main-stage presenters picked slightly better horses.

The full tracker is below. Filter by manager, side, or sector; the cumulative-return chart updates as you click.

Sohn 2026 pitch tracker

The Leaderboard: Winners and Losers

Three names lead the pack after seven days, and three are deep in the red.

Infineon Technologies (IFNNY), pitched long by Kevin Salimian of Voxel Capital at the Next Wave session, leads the pack at +22.83%. The AI power semiconductor re-rating continues, with Citi raising its target to EUR 80 and JPM to EUR 74. Auto destock headwinds are clearing, and the market is starting to price in Infineon's leverage to the next cycle of AI infrastructure buildout.

Nokia (NOK) from Philosophy Capital is the second-best performer at +17.46%, riding the renewed enthusiasm for telco infra capex tied to AI datacenter buildouts. TTM Technologies (TTMI), pitched long by Whale Rock, rounds out the top three at +16.26% on the same AI-infrastructure tailwind.

On the other side, Jumia Technologies (JMIA) from CHANGE Global's Thea Jamison is down 12.66%. The African e-commerce play has run into renewed skepticism about execution speed and the timeline to profitability. Acadia Healthcare (ACHC) from Greenlight is off 10.91% as longer-duration turnaround names came under pressure. Comfort Systems (FIX), the comparator in the Hiddenite presentation, fell 9.33% on rate-sensitivity concerns.

Other notable movers worth flagging from the broader basket: Vista Energy (VIST) is up double-digits on its Bandurria Sur and Bajo del Toro guidance raise and the Argentina re-rating; Rezolve AI (RZLV) continues to bleed for the short side after Joyce Meng's panel pitch; and Carvana (CVNA) remains weak as oil above $108 pressures the used-car consumer.

David Einhorn's Transition Basket: A Mixed Week

Einhorn's presentation was the marquee event, as it usually is. After years of pitching obscure European companies, he returned with five domestic "transition stories" where management repositioning could unlock value. His basket through seven days:

The average across all five is -2.05%. Versant Media, the recent Comcast spinoff, is the only name in the green. The thesis there centers on live news and event programming being more resilient to cord-cutting than the market assumes. The rest of the basket is underwater, with Acadia Healthcare and Victoria's Secret each down about 5%. Both are longer-duration turnaround stories that require patience measured in quarters, not days. Judging Einhorn's transition thesis after one week is like checking your sourdough starter after ten minutes. Give it time.

Fund-Level Standouts

A few managers are worth watching at the fund level:

Voxel Capital (Kevin Salimian) leads all funds at +11.94% on the strength of Infineon alone. Single-name conviction, and so far it is paying off.

FACT Capital (Joyce Meng) is right behind at +12.95% with her Rezolve AI short. The short-selling panel, moderated by Chanos, produced some of the strongest early returns in the index.

Blue Orca (Soren Aandahl) shorted OSI Systems (OSIS), which fell 7.94%, contributing +7.94% to the index. Two of the top five index contributors are shorts, which is unusual for a conference that historically skews long.

Maplelane Capital (Leon Shaulov) is averaging +2.32% across Datadog, Texas Instruments, and Lam Research. A balanced semi and software basket that is holding up nicely.

Balance Capital (Tariq Barma) pitched Perimeter Solutions long, up +6.23%. Another Next Wave manager outperforming the main stage.

General Equity (Andrew Bellas) is having the roughest week at -15.48% on Carvana alone.

What We're Watching Next

Seven trading days is just a snapshot. These things take time to play out. Some of the best Sohn pitches in history looked mediocre in the first week and crushed it over six to twelve months. Some looked great early and faded. The value of tracking them is in understanding the thesis, the catalysts, and whether the market is confirming or pushing back on the narrative.

A few things we are watching:

• The Einhorn basket needs a catalyst. Versant Media earnings and any Acadia Healthcare occupancy data will be the first real data points to test the transition thesis.
• Vista Energy's momentum depends on Brent staying above $80. If oil cracks, the Argentina re-rating unwinds quickly.
• The short sellers are winning early. Two of the top five contributors are short positions. If that holds, it says something about the current market environment and the quality of the short-selling talent Sohn is cultivating through the Next Wave program.

We will update this Sohn 2026 scorecard at the 30-day, 90-day, and one-year marks.

FAQ

What is the Sohn Investment Conference?

The Sohn Investment Conference is an annual hedge fund event in New York City where leading managers present their highest-conviction stock pitches. Founded in 1995 in memory of Ira Sohn, the conference has raised more than $150 million for pediatric cancer research at Memorial Sloan Kettering's MSK Kids program.

When was the 2026 Sohn Investment Conference?

The 31st annual Sohn Investment Conference took place on May 12, 2026 at Jazz at Lincoln Center in New York.

Which Sohn 2026 stock pick is performing best?

After seven trading days, Infineon Technologies (IFNNY), pitched long by Kevin Salimian of Voxel Capital at the Next Wave session, leads at +22.83%. Nokia (NOK) from Philosophy Capital and TTM Technologies (TTMI) from Whale Rock round out the top three.

Which Sohn 2026 pitch is performing worst?

Jumia Technologies (JMIA), pitched long by Thea Jamison of CHANGE Global, is the worst-performing main-stage long at -12.66% over the first seven trading days, followed by Acadia Healthcare (ACHC) at -10.91% and Comfort Systems (FIX) at -9.33%.

How did David Einhorn's Sohn 2026 picks perform?

Einhorn presented a five-name "transition basket" at Sohn 2026 of Versant Media (VSNT), Centene (CNC), Fluor (FLR), Acadia Healthcare (ACHC), and Victoria's Secret (VSCO). The basket is down 2.05% on average after seven trading days, with Versant Media the only name in the green at +4.98%.

How did the Sohn 2026 picks perform versus the S&P 500?

Our equal-weight Core Index of the 16 main-stage Sohn 2026 picks returned +0.74% over seven trading days versus the S&P 500's -0.46%, generating +1.20% of alpha. The Extended Index of all 30 ideas (including Next Wave and short-selling panel pitches) returned +0.07%, +0.53% above the benchmark.

Where can I track Sohn 2026 pitch performance over time?

We maintain a live tracker at app.matterfact.com/artifacts/sohn-2026-dashboard. It updates daily and lets you filter by manager, side (long/short), and sector.

We parsed 1,165 hedge fund job descriptions. Teamwork is up 31 points, tech skills are down.

TL;DR

We pulled the description text on 1,165 open hedge fund job postings and ran it through a 75-skill taxonomy. The three biggest moves between late 2025 and Q2 2026: Teamwork (+30.7pp), Communication (+19.7pp), and Ownership (+20.0pp). Meanwhile FPGA (-26.8pp), Machine Learning (-24.2pp), and low-latency systems (-21.5pp) have collapsed as required skills. The buy-side stereotype is less relevant in the age of AI. Today's funds want operators who can ship.

Last week we published the first cut of our hedge fund hiring dashboard built from a prompt in matterfact (Hedge Funds Are Hiring Like Tech Companies). The main finding was that 44% of every open role across 102 funds is an engineering role. Hedge funds are quietly becoming tech companies.

That post answered who the funds are hiring. This one answers a related question: what are the skillsets that funds value most? The answer, especially for engineering roles, was surprising.

We ran 1,165 of those postings, every one with real description text, through a curated taxonomy of about seventy-five skills, credentials, and competencies. Programming languages. AI and machine learning frameworks. Data and infrastructure. Quant math. Finance and markets. Soft skills. Education. Each posting gets counted once per skill it mentions, then we slice by time. The buckets are 2025 H2, 2026 Q1, and 2026 Q2 through May.

The result is a picture of how the buy-side skill stack is evolving. Almost none of it matches the traditional narrative.

Most-requested skills, competencies & credentials across hedge fund postings Share of 1,165 open hedge fund job postings mentioning each skill. Color encodes category: soft skills (pink), programming languages (blue), education / ML (purple), finance & markets (orange), infrastructure (cyan), quant & math (green).

The Lone Quant Is On His Way Out

The single most surprising number in the dataset is that three quarters of every hedge fund job description, 74% of the postings we parsed, now mentions teamwork or collaboration as a required attribute. A year ago it was 48%. Teamwork is now the most-cited skill across the entire industry, ahead of Python, even for engineers.

You can pair it with two other moves that tell the same story. Communication went from 26% of postings to 46%. Ownership and accountability went from 3% to 23%. These three competencies are the top three risers in the dataset.

Rising skills Biggest gainers 2025 H2 → 2026 Q2 (Apr–May). Bars show percentage-point change in share of postings. Color encodes category: competencies (pink), finance (orange), programming (blue), data/infra (cyan).

This finding is in contrast with the optics of the HF industry. The cultural image of the buy-side technologist is a head-down developer who knows more math than the people around him and is happiest with a Bloomberg terminal and his Jupyter Notebook. Funds today describe a different persona in their own postings: a teammate who communicates, takes ownership of an outcome, and ships work with other humans.

Multi-manager pod shops like Millennium, Citadel, Balyasny, and Schonfeld now dominate the AUM tables, and those structures only work if dozens of small teams coordinate on risk, data, and infrastructure. The platform is the alpha, and platforms are built by people who can talk to each other and work well with each other.

Soft Skills Up, Technical Skills Down

The biggest faller in the dataset is FPGA and hardware acceleration, down 26.8 points. Low-latency systems fell 21.5 points. Both were once a signal that a fund was working on its trading edge and both are cooling in favor of softer skills like project ownership.

Falling skills Biggest decliners 2025 H2 → 2026 Q2 (Apr–May). Bars show absolute percentage-point decline in share of postings. Color encodes category: data/infra (cyan), AI/ML & education (purple), finance (orange), competencies (pink), programming (blue), quant/math (green).

Low latency still matters at firms like Optiver, IMC, Tower, Citadel Securities, and Qube Research; they will keep paying for nanoseconds. But the marginal posting at the marginal fund is no longer about shaving 500 nanoseconds off a quote-to-trade path. That race has been won, the infrastructure exists, and the talent that built it is now being retained rather than freshly hired.

The same pattern shows up in the cloud numbers. AWS mentions are up 9.5 points. Kubernetes is up 9.4 points from a base of effectively zero. Five years ago the cloud was a punchline on a trading desk and today even latency-sensitive shops are running risk, research, and tooling on hyperscalers. The plumbing is changing.

The Machine Learning Paradox

Machine Learning as a required skill dropped 24.2 points and PhD requirements dropped 11.2 points. Deep learning, PyTorch, and TensorFlow each sit at 2 to 3% of postings.

A year ago "Machine Learning" was the marketing layer on a job description. Funds put it in because candidates wanted to see it and because investors wanted to see it. Today AI is plumbing, it is in the IDE, it is in the research workflow, it is in the data pipeline. Funds no longer need to hire "an ML engineer", or a lot less so than before. They still need a backend engineer who knows how to use foundation models, vector stores, and orchestration frameworks competently. But the job description doesn't say "ML." It says Python, AWS, and "ships fast."

That theory is supported by the rest of the data. Python rose 13.2 points and now appears in 60% of all hedge fund postings. NLP and LLM mentions are up to 8% from near zero. The infrastructure skills, including AWS, Kubernetes, CI/CD, and SQL, are all rising. Hedge funds are hiring the operators who will deploy AI inside the firm rather than the researchers who will invent new architectures. Most of the architectures they need have already been invented by Anthropic, OpenAI, Google, and Meta.

This also explains the PhD decline. When the model is downloadable, the credential matters less than the deployment skill. You can hire a master's-level engineer who has shipped three LLM-powered systems and get more leverage than a freshly minted PhD who has trained one from scratch. The funds know this and their job postings reflect it.

In terms of experience, the 3-5 year cohort is on the rise.

About this data

This data comes from an artifact that was built in matterfact from one prompt. The scraper was built on the fly and then the data was collated and presented as a dashboard. Specifically on the scraper, we run a curated keyword/regex taxonomy across each job's full description text for ~75 skills, competencies, and credentials grouped into seven buckets (Programming Languages, AI / Machine Learning, Data & Infrastructure, Quant & Mathematical, Finance & Markets, Soft Skills / Competencies, Education & Credentials). A skill is counted once per posting if any of its patterns match. Time comparison uses each posting's posted-by date bucketed into 2025 H2 (62 jobs), 2026 Q1 (274 jobs), and 2026 Q2 Apr–May (819 jobs).

Hedge fund skills and experience dashboard

What would you ask about HF job postings? Request access to matterfact and find out.

Top hedge funds are rebuilding their research process around AI. Here's the playbook for analysts who want to be indispensable.

TL;DR

• The hedge fund industry is splitting into two camps: firms that have embedded AI into every layer of their research process, and firms that are about to lose their best people to the ones that have
• If you're an analyst, the question isn't whether to use AI; it's whether you're using it well enough to stay ahead of the person who wants your seat
• The analysts pulling away are not just summarizing earnings calls. They are rebuilding their research workflows around AI
• This is the playbook for becoming indispensable

The hedge fund analyst job is changing, and it's clearly not immune from the disruption AI has caused in other areas.

You probably won't get fired because you were wrong on one stock. But you may eventually get replaced because you were too slow to adapt. If you're spending three days building a comp table that someone else could now build in ten minutes, it's time to take a serious look at AI.

Five years ago, being a strong analyst meant knowing your sector, building clean models, reading filings late at night, and having good instincts. All of that still matters, and yet the bar has moved.

The analysts who are starting to pull away are using AI every day. The use cases go well beyond an isolated tool or a chatbot to ask questions.

They are using it to move faster, cover more stocks and industries, go deeper on their thesis, research more sources, and spend more time thinking.

We work with a lot of hedge funds on AI strategy. The best analysts are not just summarizing earnings calls manually, they are rebuilding their research workflows around AI. This is the playbook to become indispensable and not worry about being replaced by someone who knows AI.

Start using AI for real hedge fund analyst workflows

Most hedge fund analysts say they are using AI but in reality they are barely scratching the surface.

They ask ChatGPT to rewrite an email, they use Copilot to fix a formula, or maybe they summarize a transcript once in a while.

That is a start, but it is not a workflow. Using AI chatbots as isolated tools is missing the bigger picture.

The analysts getting ahead of their peers are using AI on real research tasks like scanning dozens of companies at once and comparing sentiment and language across earnings calls. They are looking for early shifts in a sector and building first-pass models and dashboards from plain English prompts faster than their counterparts stuck in Excel and Tableau. They scan more sources and integrate that into their thesis. They look deeply at the bull and bear cases and get the essence of all the relevant sell-side research much faster than manually reading reports.

The best place to start is simple: use AI for one meaningful research task every day.

Use it for idea generation. Use it to sharpen a thesis. Use it to monitor risk factors. Use it to find something your PM has not seen yet. This practice will teach you not just what AI can do, but also its limitations.

Check everything

AI is useful but you still need to check sources.

LLM models tend to make up facts and even invent quotes. We wrote a piece about that, detailing exactly how to fail with AI. AI might give you analysis that sounds right and falls apart when you check the source and it's not really credible. (We have built ways to deal with these shortcomings of foundational models but still urge caution.)

A good analyst treats AI like a smart junior analyst or assistant. Helpful, fast, sometimes impressive, but not fully trusted without review.

This is also where the opportunity is and your ability to outshine others. The analyst who can use AI to move faster, then apply judgment, context, and skepticism, becomes much more valuable to the PM and the org as a whole. You may become the quality control layer as the organization adopts AI in more use cases.

That is a good place to be.

Use tools built for investment research

ChatGPT is amazing but it is not built for hedge fund research.

It does not know or understand your portfolio, your coverage universe, or have native access to the sources you care about. It cannot reliably connect filings, transcripts, podcasts, expert commentary, and alternative data into one research workflow.

That is why the best funds are moving beyond generic chatbots, and this is why we built matterfact.

Fund analysts want tools connected to real data, they want an audit trail with citations, and automated workflows that fit how they and their PMs actually work. They want output that is helpful at a Monday morning investment meeting.

That is where matterfact is different from most agentic models and off-the-shelf LLMs.

We built an agent that you can ask about what's changing in a sector, which management teams are shifting tone, where a competitor is hinting at pressure, or what experts are saying in places the Street is not watching closely.

Lean on specialty tools like that to supercharge your productivity with AI and you'll also realize how much time can be saved to think and invest.

Bring your PM something they did not think to ask for

Good analysts answer the question, but great analysts bring something nobody thought to ask for yet.

Say your fund owns Carvana and your PM asks for an update before earnings. The standard response is pretty obvious. You refresh the model, check consensus, update the deck, and maybe add a few notes from recent calls.

That is useful, but also what everyone else is doing.

The AI-native analyst goes a layer deeper. They search across podcasts, transcripts, expert commentary, and industry conversations to get at the sentiment trend. They find a competitor talking about pricing pressure in the Southeast. They catch a logistics expert discussing title processing delays. They pull every mention of Carvana, DriveTime, ADESA, used car financing, inventory turns, and regional demand from the last 90 days, then look for patterns.

Now the meeting is different than it would have been without you, and the PM will notice.

Instead of walking in with a refreshed earnings preview, you walk in with a sharper bull-bear case backed by primary audio and commentary most of the Street has not synthesized yet.

That is how you stand out in the age of AI. Blow their socks off.

Go beyond decks and build dashboards

For a long time, the default analyst artifact was the deck. It might run forty slides, take a week to update, and then get a few minutes of attention from the PM before the conversation moved on.

That workflow is already dated.

A better research artifact is something the PM can actually use in real time and on demand. It should be live, easy to explore, and organized around the questions that matter: what supports the bull case, what strengthens the bear case, which assumptions drive the model, where the comps sit, what the key risks are, and which sources support the conclusion.

This is where AI starts to change the workflow in a real way. An analyst can now build the first version of an equity research dashboard in minutes instead of spending days assembling the same pieces by hand. Peer comps, valuation work, scenario analysis, podcast-sourced thesis points, citations, and model drivers can all be pulled into one place from a natural language prompt.

The first version will not be perfect, but you can go from idea to finished dashboard in just a few minutes and tweak it later.

The analyst spends less time assembling the artifact and more time improving the research.

Catch red flags early

This may be the most valuable use case, as it can bring a ton of value and protect your fund from loss.

Every portfolio has risks that are not in the filings yet. It can be subtle: the CEO starts using different language describing demand trends. A supplier talks about capacity constraints on a niche podcast. A customer hints at slowing orders from a particular company. A former executive says something that changes how you think about the story.

Most of these signals are out there before they become headlines, but the problem is scale. No analyst can listen to everything, read everything, and connect every dot, all the time.

AI changes that.

You can monitor thousands of sources, track sentiment over time, and get alerted when something shifts.

The analyst who walks into the morning meeting and says, "Hey, I flagged something overnight: our supplier's CEO was on a podcast yesterday, and his language around Q3 capacity changed materially from April," is not getting replaced.

That analyst is hard to live without, because people who get how to use AI are in extremely high demand.

Become an analyst rock star with AI

AI will not remove the need for analysts who can think, but it will put pressure on analysts who only gather, format, and summarize like it's 1999. Those who resist AI might eventually find themselves on the way out.

The best hedge funds are going to pair human judgment with AI-powered research. The best analysts will become the bridge between those two things.

They will move faster, cover more stocks, and bring better questions to the table.

The PM will know that AI helped you but won't care. In fact, they might ask you to train your colleagues to do the same.

What is an AI investment research platform? A buyer's guide covering the 5 core capabilities, evaluation criteria, and how it differs from terminals and chatbots.

TL;DR

• An AI investment research platform is not a chatbot, not a terminal, and not a smarter Ctrl+F
• It is a synthesis and workflow layer purpose-built for institutional public-markets teams
• This guide defines the category, walks through the five capabilities that matter, and gives you a framework for evaluating vendors before you sign anything

What Is an AI Investment Research Platform?

An AI investment research platform is software that combines large-scale data synthesis and workflow execution into a single environment designed for institutional equity and credit investment teams.

Every vendor with an LLM inside a wrapper and a financial data feed now claims to be an AI-native tool. But there is a meaningful difference between a general-purpose chatbot that can answer a few questions on stocks, a niche tool that performs an isolated workflow with an AI layer on top, and a full-fledged system that plugs into all the workflows an investor runs through on any given day. Only the latter is a true AI platform. A platform should ingest thousands of earnings transcripts, broker reports, and expert interviews, synthesize them, weigh them against a specific investment thesis, cite its sources, and then build a live dashboard tracking the variables that matter to your position.

The distinction is the same one that separates a search engine from a Bloomberg terminal: both give you information, but only one lets you run end-to-end workflows.

If you are a CIO evaluating tools, a head of research building a tech stack, or a senior analyst trying to figure out whether this category is real or just repackaged chatbot marketing, this guide is for you.

What an AI Investment Research Platform Is Not

Before we get into what it does, it helps to define what it isn't. The confusion in this market is not accidental as the proliferation of AI tools blurs the lines.

First, a platform is not a chatbot. A chatbot answers questions one at a time. It does not understand your portfolio context, does not track your coverage universe, does not build persistent analytical outputs, and does not monitor for changes alerting you when you need it most. If the tool forgets important context the moment you close the tab, it is not much more than a chatbot.

It is not a terminal. Bloomberg and FactSet are data delivery systems. They are extraordinary at what they do, which is structured data retrieval, charting, and portfolio analytics. But they are not synthesis engines. They will not read 200 earnings calls and tell you that management sentiment on capital expenditure has shifted from cautious to aggressive over the past three quarters, because that requires a different architecture entirely.

It is not just document search. AlphaSense and similar platforms made "smart Ctrl+F" a category. You can search across SEC filings, transcripts, and broker research with natural-language queries. That is useful, but it is still retrieval, not really synthesis. You get a list of passages that match your query but still have to read them, compare them, and connect the dots yourself.

It is not a copilot. Microsoft and Google sell general-purpose AI assistants that sit alongside your existing tools. They are good at summarizing an email or reformatting a spreadsheet. They are not good at telling you whether AMD's strategic pivot into hyperscaler partnerships creates a credible threat to Nvidia's GPU moat, because they do not have the domain-specific data architecture to support that kind of reasoning.

An AI investment research platform sits in the gap between all four. It combines the data awareness of a terminal, the search capability of a document platform, the conversational interface of a chatbot, and the workflow integration of a copilot, all built on top of investment-grade data sources and calibrated for the way institutional analysts actually work.

The Five Capabilities That Define the Category

If a vendor claims to have an AI investment research platform, these are the five capabilities you should look for. Miss any one and you are looking at a partial solution that will fall short of covering most workflows. While no one solution is perfect, investors strive for maximal coverage of the work that is done manually in their org. Here are the five things they look for in an AI investment research platform.

1. Multi-Source Synthesis

The platform should be able to pull from basically any source like earnings transcripts, broker research, SEC filings, news, social media, and alternative data sources simultaneously, and synthesize them into a single coherent output, either a live dashboard or a report. This is the core of the category. An analyst covering semiconductor capital equipment should be able to ask about high-NA EUV adoption timelines and get a synthesized answer drawing from management commentary, supplier disclosures, and competitor transcripts. The output should be a synthesis with citations (covered next) so the analyst can double-check the source and build trust in the system.

2. Source Citation and Audit Trail

Every claim the platform makes should be traceable back to a specific source, with date, speaker, and context. This is non-negotiable for institutional use. Analysts need to be able to verify what the platform tells them. Portfolio managers need to trust the inputs before they size a position. If the platform gives you an answer but cannot show you exactly where that answer came from, it is not ready for institutional deployment.

3. Workflow Execution

The platform should do more than answer questions. It should execute research workflows: tracking a coverage universe over time, monitoring management sentiment shifts, flagging when new information contradicts your existing thesis, and scheduling recurring analyses. The difference between a tool you use when you have a question and a platform that works for you continuously is the difference between a flashlight and an electrical grid: both produce light, but only one works while you sleep.

4. Dynamic Dashboards and Artifacts

Institutional research produces artifacts: models, charts, comparison tables, scenario analyses. The platform should be able to generate these on the fly from conversational queries. Ask about Chipotle's unit economics by metro area and get a live dashboard, not a paragraph of text. Ask for a bull/bear framework on a biotech name and get a structured output you can put in front of your PM without reformatting it in PowerPoint first.

5. Source from Unstructured Intelligence like Podcasts

This is the newest layer and the one most underserved by legacy tools. There are over four million active podcasts globally, and a growing share feature executives, industry practitioners, and domain experts discussing exactly the kind of forward-looking, qualitative information that institutional investors need. An AI investment research platform should be able to ingest and synthesize this unstructured content alongside traditional financial data. The insights buried in a two-hour podcast interview with a supply chain executive at a medical device company are often more valuable than anything in the 10-K. It is like learning from thousands of expert calls on demand. The problem is nobody has time to listen to two hours of audio to find the three minutes that matter, so you need AI to help.

How to Evaluate Vendors: A Framework for Buyers

If you are at the point of evaluating specific platforms, here is a practical framework. These criteria are listed roughly in order of importance for a typical institutional buyer.

Data coverage and freshness. What sources does the platform ingest? How current is the data? A platform with a 48-hour lag on earnings transcripts is significantly less useful than one with same-day coverage. Ask about the cadence and whether the platform can ingest proprietary data sources specific to your firm.

Synthesis quality. Run the same complex query across multiple platforms and compare the outputs. Does the platform connect information across sources, or does it just summarize them individually? Does it surface contradictions? Does it identify when a company's public commentary diverges from what channel checks or industry experts are saying?

Citation reliability. Verify the citations. Open the cited source and check whether the platform accurately represented what was said. Do this at least ten times. Citation hallucination (plausible-sounding but fabricated or misattributed citations) is the single fastest way to lose analyst trust, and it is more common than vendors want to admit.

Workflow persistence. Can you set up recurring analyses, scheduled reports, and monitoring dashboards? Or does every interaction start from zero? Institutional workflows are ongoing. Your tools should be too.

Security and compliance. Where is the data stored? Who can access your queries? Is the platform SOC 2 compliant? Can it be deployed in a way that satisfies your firm's information security requirements? This is often the deciding factor for larger allocators, and it should be high on the list for everyone.

Total cost of ownership. Consider not just the subscription price but the analyst time required for onboarding, the cost of maintaining integrations, and the opportunity cost of choosing a platform that does 60% of what you need versus one that does 90%. The cheapest seat license is rarely the cheapest solution.

Why This Category Exists Now

Three things converged to create this category.

First, language models got good enough to do this for investors. The transformer architectures that power modern LLMs can finally handle decent reasoning, summarization, and synthesis tasks that investment research demands. Two years ago, they could not do this reliably. Now they can, when properly constrained and calibrated against domain-specific data.

Second, the data landscape exploded. The volume of unstructured information relevant to investment decisions (podcasts, social media, satellite imagery, web traffic data, patent filings) has grown faster than any human team can process. The analysts who figure out how to systematically extract signal from this noise will have an edge. The ones who do not will spend their time reading the same broker notes everyone else reads.

Third, the economics of research teams changed. Headcount at buy-side firms is flat or declining while the complexity of coverage universes is increasing. Firms need leverage, and AI investment research platforms provide it by giving each analyst the effective throughput of a much larger team.

FAQ

What is an AI investment research platform?

It is software that combines data synthesis, source citation, and workflow execution into a single environment built for institutional investment research teams. It sits between a terminal (structured data retrieval), a document search platform (passage retrieval), and a chatbot (conversational Q&A), combining elements of each with investment-domain-specific architecture.

How is an AI investment research platform different from a Bloomberg terminal?

Bloomberg excels at structured data delivery, charting, and portfolio analytics. An AI investment research platform focuses on synthesizing unstructured information (transcripts, reports, podcasts, news) into analytical outputs. They are complementary, not competitive. Most institutional teams will use both.

How is it different from AlphaSense?

AlphaSense is primarily a document search platform. It retrieves relevant passages across a large corpus of financial documents. An AI investment research platform goes further by synthesizing information across sources, generating analytical artifacts, and executing persistent research workflows.

Can an AI investment research platform replace expert networks?

For many use cases, yes. Practitioners discuss forward-looking insights, industry dynamics, and operational details on podcasts and in public forums. An AI platform that can synthesize millions of hours of this content gives you much of what you would get from an expert call, without the compliance overhead, scheduling friction, or $1,500-per-hour price tag.

Is it safe for institutional use?

That depends entirely on the vendor. Evaluate SOC 2 compliance, data residency, access controls, and whether the platform can be configured to meet your firm's specific information security requirements. Ask for the security documentation before you start a trial.

What does an AI investment research platform cost?

Pricing varies widely across the category. Some vendors charge per seat per year, others by usage. Expect institutional-grade platforms to cost between $5,000 and $50,000 per seat annually, depending on the scope of data access and the level of customization.

How long does onboarding take?

Most platforms can be deployed in days, not months. The real onboarding cost is analyst adoption: getting your team to build the platform into their daily workflows rather than treating it as a novelty they use once and forget.

Can I integrate it with my existing tools?

Look for platforms that offer API access, data export, and integration with your existing data warehouse or portfolio management system. The best platforms fit into your stack. They do not ask you to replace it.

Will AI investment research platforms replace analysts?

No. They amplify analysts. The bottleneck in institutional research was never "not enough smart people." It was "too much information for any team to process." These platforms address the throughput problem, not the judgment problem.

How do I measure ROI?

Track three things: analyst time saved on information gathering (most teams see 30-50% reduction), speed to insight on new positions, and the number of differentiated insights surfaced that would not have been found through traditional workflows.

The Bottom Line

The category is real, but the noise is high. An AI investment research platform is not a chatbot with a financial-data feed, and it is not a smarter search box. It is a synthesis and workflow layer that unifies your data, encodes your sector knowledge, monitors your theses, cites every claim, and slots into the workflows your analysts already run.

Want to test drive a true AI investment research platform? Try it at matterfact.com.

A tutorial on Playbooks: 217 pre-engineered research workflows across 12 categories, from morning briefs to sector-specific KPI math. The prompt does the thinking before you arrive.

TL;DR

• 217 pre-engineered prompts with structured inputs and visible logic
• Organized into 12 phases of research, from morning briefs to sector-specific KPI math
• The prompt is already smart by the time you arrive; you're not prompting, you're configuring
• The point is not that the AI is smart; the point is that the workflow is
• A guided tour through how the system is built and where the leverage actually lives

Sometimes the blank prompt is a problem

Most analysts know that AI is transforming industries and changing the way people work and live. And yet, many who have tried an AI research assistant have had mixed results. At times, they stare at a blank box not really knowing what to ask and how to ask it. When they prompt something like "find me five interesting names in semis" what comes back is almost never useful and rather disappointing. Nothing you can't find with a simple Bloomberg screen. They might be tempted to think that the technology is not yet there for real-life investment research.

In reality, the tech is fine. The prompt is the problem.

An investment analyst does not have the time and patience to think and interact with AI like a seasoned prompt engineer. They vaguely know what they want but are not ready to expend the energy to provide all the context the model needs to really nail that reply.

Fund managers and analysts who are using AI successfully usually lean on prompts that have already been engineered for the job. This skips a big step, takes the onus off the analyst, and yet still gives pristine results.

Analysts don't have time to prompt engineer. That's why we built Playbooks.

That is what Playbooks are. Pre-engineered prompts, with structured inputs and customizable context, organized by the phase of research you are in. We have shipped 217 of them across 12 categories, and they are shortening the time to value with AI. This post is a guided tour through how the system is built and how to use it for real work.

The shape of the system

Open the Playbooks tab in the matterfact terminal and you will see the entire workflow of an institutional analyst broken into twelve numbered phases:

youtube url: https://youtu.be/1cqTDTodd8A title: A quick tour of matterfact Playbooks caption: A quick walkthrough of the Playbooks library inside the matterfact terminal.

## The anatomy of a Playbook

Click any card and you get a modal with three things: a short description, a set of structured input fields, and a Prompt Preview that shows you exactly what gets sent to the model. The Prompt Preview teaches you how to correctly ask AI for what you need in real time.

request-access
heading: Configure a sector framework for your team.
description: 53 sector-specific Playbooks covering the math generic LLMs miss. Hotels RevPAR, QSR unit economics, semis capex, REIT NAV, and more.
buttonText: Request access

Operationalizing podcast intelligence

We have written before about why podcasts are now a primary research input. The Playbooks under Podcast Insights are how that insight becomes a workflow.

Topic Expert Consensus is the one I would point a new analyst to first. You give it a topic, say, "AI infrastructure capex cycle and hyperscaler ROI" and an optional time window. The Prompt Preview shows what a great prompt looks like:

Sweep podcast transcripts for expert commentary on: AI infrastructure capex cycle and hyperscaler ROI. Time window: last 90 days. Generate 4-6 diverse search angles to maximize recall (topic synonyms, adjacent terms, specific company / person names tied to the topic). For each distinct expert who has spoken substantively, capture their view, the underlying argument, and the timestamp / episode. Then synthesize: (1) the consensus view, (2) dissenting / contrarian takes, (3) variant perception (where the buy-side debate hasn't caught up to the experts), and (4) the strongest single argument on each side. Cite episodes by name and date. Schedule sending me a weekly expert-consensus digest on this topic to my email at: weekly Sunday at 8:00 AM ET.

First, the recall step of generating diverse search angles before retrieving addresses the "feature" of the model to give you what matches your established assumptions. Second, the synthesis step explicitly separates consensus from variant perception, which is where the alpha actually lives. Third, citations to specific episodes mean you can verify or escalate to a longer listen on anything that matters.

Run it on Sunday morning, get a digest in your inbox by 8:00 AM ET, walk into Monday already several hundred episodes ahead.

There are several other categories modeled after real investment and finance workflows and engineered to both save time and improve the results you get.

The point of Playbooks is that the mental load to structure the perfect prompt is done before you arrive. What is left is the thing only you can do in seconds: read, decide, and run it.

https://app.matterfact.com/playbooks

Analysts love Playbooks, and if you haven't tried them yet, we are sure you will love them too. Find the perfect Playbook for you.

44% of all open hedge fund roles are for engineers. We scraped 1,555 job postings across 102 funds to show how the buy side is quietly becoming a tech industry.

TL;DR

• 1,555 open job postings analyzed across 102 hedge funds managing nearly $5 trillion
• 44% of all open roles are engineering positions, more than traders, analysts, and PMs combined
• Senior engineers now command $450K–$700K in the US, £300K–£500K in London
• London leads with 315 open eng roles; India, Singapore, and Bulgaria emerging as cost hubs
• Hedge funds aren't using technology, they're becoming technology companies

You'd Be Forgiven for Thinking This Was a Tech Company

Across 102 hedge funds collectively managing $4.9 trillion, 44% of every open job posting is for an engineer.

Engineers are hired more than traders, analysts, portfolio managers, operators.

Engineers.

Hedge fund open roles by category Share of 1,555 open job postings across 102 hedge funds, by role category.

We built this dashboard with one prompt on matterfact. It built a scraper on the fly to look across 102 funds and found 1,555 open postings. Only 35 of those funds had any public job listings at all. Another 58 hire exclusively through executive search and on-cycle recruiters. Nine funds had their portals walled off entirely. They happen to be some of the larger funds and pod shops.

Hedge fund hiring dashboard

In other words, the real number of engineering hires happening across the industry is almost certainly much higher than what the public data shows.

What Kind of Engineers Are They Hiring?

Hedge fund open roles by category All 1,555 open job postings across 102 hedge funds, broken down by function.

Let's break it down. Of the 679 engineering roles we tracked, the largest bucket is general software engineering, at 205 roles. Then infrastructure and DevOps at 101. After that comes a category you won't find in a typical tech company's job board: quant developers, at 62. Security engineers (42), low-latency specialists (29), data engineers (28), and trading systems engineers (25) round out the picture.

Engineering function mix Sub-bucket breakdown of the 679 open engineering roles at hedge funds.

Looks more like a software company building a consumer app. This is an industry that needs engineers who understand microsecond latency, real-time data pipelines, and execution infrastructure that can handle billions of dollars in daily trading volume without flinching.

Goldman Sachs made headlines a few years ago when David Solomon revealed that 25% of the bank's entire workforce were engineers. Today, hedge funds are looking to blow past that number. At firms like Qube Research & Technologies (249 open roles), Millennium (227), and Optiver (165), engineering headcount isn't a support function; it's the core of the business. And the fact that AI can write code is not replacing engineers but creating more room for them.

How Does This Compare to the Rest of the Economy?

Let's put 44% in context.

At a typical technology company, engineering roles make up roughly 25–35% of total headcount, depending on stage and product type. At major banks like JPMorgan, which budgeted $18 billion for technology in 2025, engineers represent a significant but still minority share of the workforce. JPMorgan has over 1,000 tech openings out of 7,000+ total openings at any given time, roughly 15%.

Manufacturing and industrial companies have engineering roles typically sit at 8–12% of workforce, focused on product design, process optimization, and increasingly on automation. Traditional financial services (asset managers, insurance companies, wealth advisors) hover around 10–15% for technology and engineering combined.

Hedge funds at 44% are way above all those. The only organizations that consistently hit similar engineering density are pure-play software companies and AI labs. The caveat here is these are new roles, not the current makeup, which is surely less heavy with engineers.

The Geography Tells a Story Too

London leads with 315 open engineering and technology roles, followed by a significant concentration across the US (553 total roles, with New York and Chicago as the primary hubs). But what's really interesting is where the growth is happening at the margins.

Open hedge fund roles by city Top 15 cities by number of open engineering and technology roles.

India accounts for 126 roles. Singapore has 91. Bulgaria (yes, Bulgaria) shows up in the data. Hedge funds are doing what tech companies did a decade ago: building distributed engineering hubs in lower-cost markets to scale capacity without breaking the compensation budget.

Open hedge fund roles by country Top 15 countries by location of posting, not fund HQ.

Senior engineers at top US hedge funds now command $450,000 to $700,000 in total compensation, matching or exceeding Big Tech. In London, the range sits at £300,000 to £500,000 for infrastructure and trading systems engineers. At those price points, firms are inevitably looking at Bangalore, Sofia, and Warsaw to build out the next layer of engineering depth.

Why This Matters If You're an Investor

If you're allocating capital, this data is a leading indicator. The funds that are investing most aggressively in engineering infrastructure are the ones positioning for the next cycle of performance, not just through better stock picks, but through better systems that lead to a better process.

The shift from speculative AI hiring toward foundational software engineering is especially telling. Several large systematic funds have rebalanced their technology teams away from experimental ML roles and back toward backend systems, execution pipelines, and real-time data processing. They're not chasing the AI hype cycle. They're building the plumbing that makes performance repeatable and scalable.

Think of it like this: a fund that hires 100 engineers isn't just building a trading system. It's building an operating system, one that can ingest data faster, execute with lower latency, manage risk in real time, and adapt to new markets without rebuilding from scratch. That's a structural advantage that compounds over time, much like the technology moats that define the best software companies.

The Talent War Is the Real Story

The uncomfortable truth is that the hedge fund industry is now directly competing with Google, Meta, and OpenAI for the same engineering talent. And in many cases, they're winning, not on mission or culture, but on compensation and the proximity to direct business impact.

An engineer at a hedge fund isn't building a feature that might move a metric by 0.3%. They're building systems where a microsecond improvement can translate into millions of dollars. That feedback loop, code to P&L, is something rather unique in hedge funds.

For anyone tracking where the smart money is flowing, follow the engineering headcount. It's the most honest signal in the market.

The data in this analysis was compiled from public websites across 102 hedge funds, last refreshed May 6, 2026. For a deeper look at what's driving these hiring patterns and what it signals about the future of institutional investing, explore matterfact.

I asked Matterfact to build a full Chipotle research dashboard with peer comps, podcast-sourced bull/bear cases, valuation analysis, and scenario modeling. It took 4 minutes.

TL;DR

• Most equity research dashboards take weeks to build, days to update, and are a pain to share
• One natural-language prompt produced a complete Chipotle research artifact
• Includes price action vs peers, bull/bear cases with podcast citations, full QSR comp table, scenario analysis
• Four minutes, zero code, no BI tools, no engineers

How most research dashboards get built at investment firms

If you have ever exceeded the capacity of Excel with your financial models, we know how you feel. The next logical step may have been to ask your data or engineering team for a custom dashboard. After some time they come around to it, scope the work, and create a ticket. An engineer needs to pull the data or write a scraper, clean it, store it, model it, and push it into Tableau or Looker. Weeks (or months) after your initial request you get something close to what you asked for. Unfortunately the market has moved on and the opportunity is no longer there, you are now looking at a different name.

These workflows are a relic of the pre-AI era. They are a remnant of a world where data engineering was expensive and compute was limited. In 2026, neither of those things is true and you can go from idea to reality in seconds. Sometimes it feels like magic.

I wanted to test whether matterfact could compress the entire cycle of idea → product into minutes with something we call "Artifacts". I wanted to build an equity research dashboard for a public company using only natural language. I knew I needed to see performance, peer comps, most recent bull and bear cases with real source citations. I needed some fundamental analysis to see if there was value relative to other companies and I also was curious to see some scenario modeling.

I chose Chipotle ($CMG) because the name is hotly debated in hedge fund circles. There are strong opinions on both sides of CMG right now and the data to support either case is scattered across earnings calls, sell-side notes, and hundreds of podcast conversations most analysts will never hear.

What I asked for

I typed this prompt into the matterfact platform:

Build an interactive artifact for CMG. First, show me recent price action relative to the market and companies similar to Chipotle in the fast food restaurant space. I want to see how CMG was performing against its peer group.

Second, give me the best bull and bear case, and tell me who has made each argument and in which podcast it was mentioned.

Third, build a general dashboard so I can understand Chipotle's financials with a comp analysis across its peer group. Focus on valuation metrics, operating KPIs, the things that fundamental analysts look at when building a position.

Fourth, give me the best drivers of forward-looking returns and some scenario analysis and what-if modeling.

That was it. One prompt. The agent went to work and what it produced was impressive.

Price action and peer comparison

The first section of the dashboard gave me CMG's recent stock performance plotted against the broader market and its QSR peer group, which it chose automatically for me. You can immediately see how Chipotle was trading relative to McDonald's, Yum Brands, Restaurant Brands International, Wingstop, Cava, Sweetgreen, and the other names that institutional analysts track in the space, and it gave me the ability to turn off some names to focus on the group I care about.

This would take a while in MS Excel or even Bloomberg, not because the data is hard to find, but because pulling the peer set, normalizing the time series, and formatting it into something presentable takes time.

Bull and bear case with podcast citations

This is where it gets interesting.

The agent didn't produce generic bull and bear arguments from a language model. It went into matterfact's podcast intelligence engine, scanned hundreds of relevant episodes, and pulled the strongest recent arguments from each side. All cited and sourced. You can see exactly who made each argument, on which podcast, and when.

On the bull side, you see operators and analysts talking about menu pricing power, unit economics durability, digital penetration, and international expansion optionality. On the bear side, you hear arguments about BOGO-driven brand dilution, commodity headwinds from beef inflation running 7-20% across the sector, and the question of whether same-store sales growth can sustain at current valuation multiples.

This is the kind of research that used to cost thousands in expert network calls and two weeks of scheduling. The AI agent assembled it in seconds, and every claim is traceable to a real human expert speaking on a real podcast. These are all real people making specific arguments in actual conversations recorded live.

For analysts preparing a pitch to their portfolio manager, this is the section that can really help prep you. You can walk into the investment committee meeting with the counterarguments already in your back pocket, ready to defend.

Fundamental comp analysis

The third section is the core of equity research: a full comp table across the QSR and fast casual peer group, organized around the metrics fundamental analysts rely on.

Valuation multiples like EV/EBITDA, P/E, and P/FCF across the peer set. Operating KPIs: same-store sales growth, restaurant-level margins, average unit volumes, new unit economics, digital mix. Balance sheet metrics: net debt/EBITDA, ROIC, free cash flow conversion.

This is the table every analyst covering restaurants has in their model, and it is the one they spend the most time maintaining. Every earnings season they have to update three dozen cells by hand. The dashboard builds it from scratch in minutes, across the full peer group in one view, and pulls in fresh data when it's out.

Scenario analysis and forward-looking drivers

The dashboard also includes scenario modeling and what-if analysis for Chipotle's forward-looking returns.

What happens to the stock if same-store sales growth decelerates by 200 basis points? What if beef inflation persists through the back half of 2026? What if digital mix continues to expand and drives operating leverage? What if the international opportunity in Europe and the Middle East starts to contribute meaningfully to unit growth?

Beyond Chipotle

This is one dashboard for one company built in a few minutes from a single natural language prompt. The same approach works for any public company in any sector, or anything an analyst could imagine.

Give matterfact a ticker and a set of questions and it will build you a research artifact that combines market data, peer analysis, podcast-sourced intelligence, and scenario modeling. You can build custom screeners, risk dashboards, macro analysis, news aggregators. You are no longer limited by tech, only your imagination.

See the Chipotle dashboard for yourself below, or open it in a new tab.

Chipotle research dashboard

What company would you build a dashboard for? Tell us at matterfact.com and we will build one for you. Or, give it a spin yourself.

Built with matterfact

This dashboard was built live inside the matterfact platform using a single natural language prompt. No code, no BI tools, no engineering support. The platform combined market data, peer analysis, podcast intelligence from hundreds of episodes, and scenario modeling into a complete equity research artifact in under four minutes.

In 1986, Charlie Munger told Harvard graduates how to guarantee a life of misery. Inspired by this framework we show you exactly how to guarantee disastrous results with AI in your investment process. Invert, always invert.

"Pick 5 Best Stocks, Don't Make a Mistake"

In 1986, Charlie Munger stood in front of the graduating class at Harvard and did something no one expected but all will remember. Instead of telling the graduates how to live their life, he told them how to guarantee a miserable one. He borrowed the idea from Johnny Carson, expanded it with his own characteristic Munger wisdom, and delivered one of the most memorable commencement speeches ever.

We are going to do the same thing here, humbly, except our subject is not life, but something almost as important. Your AI-powered investment research. If you want to guarantee that every AI tool you touch produces absolute garbage and wastes your time, if you want to ensure that your research assistant becomes a useless yes-man who encourages your worst ideas and destroys your ability to reason objectively, and embarrasses you in front of colleagues, follow these instructions closely.

Step 1: Never Give the Model Context

This is the single most important step in producing useless AI slop. We have seen it happen with our own models at matterfact. When you sit down at your terminal and open your AI research tool, type something like this:

"Find me 5 best stocks. Don't make a mistake"

Do not tell it what sector you cover. Do not tell it your fund's mandate, your market cap range, or your investment universe and horizon. Do not mention whether you run a long/short book or a long-only portfolio and how you think about risk, factors, and position sizing. Do not specify geography, models are great at assuming you are looking for investments in Asia. Do not mention valuation frameworks you care about, as that might produce a more coherent result.

The less context you provide, the better. This useless output lets you say with confidence to your PM: "I tried AI. It doesn't work." Don't worry, this is exactly what every PM wants to hear so they can confidently be left in the dust while their competitors figure out how to use AI for their research.

Step 2: Be as Vague as Possible

Specificity is your enemy. Never tell the model whether you are looking for long or short ideas, high or low vol stocks and never limit the market caps. Never specify a time horizon either, do not mention if you are looking for a catalyst trade in three weeks or a core position you want to hold for two years. The model does not know and you should not tell it, it is much more fun to let it assume something random from the depths of its digital brain.

Never mention whether you want consensus ideas or contrarian ones. Never say whether you are looking for momentum or mean reversion plays. Never reference a factor exposure you are trying to hedge or monetize.

The fund managers we work with get extraordinary results from AI but you should do the opposite. They write prompts that read like mini investment memos and include their thesis, assumptions, constraints, and the specific output format they want. They tell the model what kind of investor they are. But you should NEVER do any of that.

Step 3: Never Ask for the Other Side of the Trade

Charlie Munger admired Charles Darwin for one specific reason: Darwin always gave priority attention to evidence that would disconfirm his most cherished theories. Darwin actively hunted for reasons he was wrong. What a waste of time!

If you want to fail with AI, you must do exactly the opposite.

Never ask the model for the bear case if you are bullish. Never ask for the bull case if you are short. Never ask for opposing views, for instance never say, "What are the three strongest arguments against this thesis?" And never ask the model to poke holes in your reasoning.

One of the craziest things you can get an AI research tool to do is stress-test your conviction. It has access to thousands of expert conversations, earnings transcripts, interviews, and news sources. It can find the dissenting voice you have never heard and that can surface the risk you have not considered. Total disaster if you want to make sure you fail with AI.

If you only ask it to validate what you already believe, it will happily do that because foundational large language models are agreeable by nature - that's how they evolved to get high rankings on leaderboards. If you push them toward confirmation, they will confirm and that is a great way to fail with AI.

Step 4: Never Check the Sources

Look, AI never makes mistakes so you must trust everything the model tells you and take it at face value. Never ask it where it got a specific data point, waste of tokens. Never cross-reference a claim or ask it to cite the specific podcast episode and the time stamp, the earnings call, or the filing where it found that number.

AI models do not hallucinate. They would never present plausible-sounding statements that are partially or entirely fabricated. It would be terribly unwise to use AI as a research accelerator, only to then verify critical claims against primary sources.

You should definitely quote an AI-generated statistic in your investment committee meeting. When someone asks for the source, shrug and say "the AI said it, so it must be true." This is a terrific way to destroy your credibility in ninety seconds and make sure you are next in line to get fired.

Step 5: Never Listen to What the Experts Say, They Don't Know Anything

Ignore the best way AI can help you by scanning petabytes of data in seconds. Thousands of hours of podcast conversations happen every month across finance, technology, healthcare, and consumer sectors, and you should never use AI to scan them for signal.

If you do, you risk getting better insight than the street and eventually risk becoming a better investor.

If you want to fail, ignore all of this. Stick to the sell-side notes that every other analyst on the street is reading, attend the same conferences, and read the 10-K. Stay in the consensus lane.

Step 6: Never Invert

If you want to fail, do not follow Munger's advice: "Invert, Always Invert". Never challenge assumptions and never step out of your comfort zone.

Munger closed his Harvard speech with a backward toast: "May each of you rise high by spending each day of a long life aiming low."

We will close with our own: May each of you produce extraordinary research by first understanding exactly how to produce terrible research.

Then, do the opposite!

So, take our advice, do not use AI. You are already perfect as an investor and there is no room for improvement, especially with some latest technology fad.

Do not reach out to us, we do not want to hear from you at matterfact.

We scanned 635 podcast episodes and pulled the 10 highest-signal conversations for analysts covering QSR, fast casual, casual dining, and coffee. Here are the episodes worth your time and the 10 stocks driving the debate.

TL;DR

• Restaurant sector squeezed in 2026: 122 tariffs in effect, ICE-driven labor disruption
• Beef inflation running 7-20%+ across the sector
• K-shaped consumer is hollowing out the middle class; the Street keeps blaming GLP-1s
• 635 podcast episodes scanned, top 10 highest-signal conversations pulled
• Covers QSR, fast casual, casual dining, and coffee, with read-throughs to the most-debated stocks

The restaurant sector is more complex than your model

If you cover restaurants, you know investors are focused on the main risk drivers: value wars, GLP-1 headwinds, labor challenges. Every sell-side note offers some version of the same narrative.

To add more nuance, there are still 122 tariffs in effect, including a baseline 10% on imports that flows directly into food costs. Beef inflation hit 20%+ at Burger King in 2025, ran 7%+ at Texas Roadhouse, and is compressing Chipotle's margins into 2H 2026 before lapping. ICE enforcement has materially reduced the workforce in food manufacturing and restaurant operations, and 16% of the U.S. population has pulled back on out-of-home dining as a result. Dye reformulation, anti-plastics legislation, and bans on brominated flour are adding cost across the supply chain. These are real P&L headwinds that sell-side notes are barely touching.

Then there is the GLP-1 narrative unfolding. The Street loves it as an explanation for softening demand but GLP-1s are a "rich person's drug". The actual change in consumer behavior may be a COVID hangover that the industry still has not worked through. Operators loaded up on deals, increased inventory, and got caught when consumers slowed spending. Private label did extremely well, the middle class got hollowed out. The GLP-1 reasoning is convenient because it sounds macro and inevitable but the reality is way more complicated.

What the data shows is a K-shaped consumer. High-income transactions are going strong while lower-income traffic is down high single digits. Fine dining is the unexpected winner: Capital Grille and Eddie V's are seeing 8-12% foot traffic increases, and OpenTable data shows experiential dining bookings up 46%. On the value end, Taco Bell and McDonald's are winning the fast food war. Wendy's, Wingstop, and Cracker Barrel are the problem children. Brinker/Chili's is another share-gain story in the sector with twenty straight positive comp quarters, accelerating against the industry by 560 basis points in April.

Coffee has the most attractive sub-sector setup. Starbucks turnaround is inflecting. Dutch Bros is still compounding. Arabica is down ~25% year over year, and that margin tailwind starts rolling through in 2H 2026 as the 2-3 quarter inventory lag catches up.

Of course, little of this story shows up in a single earnings call or even a sell-side note. It does show up in podcasts. Podcasts are becoming a great source of investor intelligence because you find CEOs explaining pricing psychology in plain English and operators breaking down why bone-in chicken drives higher repeat rates than tenders. These guests talk about what they see in their stores, their supply chains, and their customer data. And because the format is long and informal, they say more than they would in any other channel. Unfortunately, it's impossible for a human to listen to all of them, that's why you need AI.

Our AI agent (matterfact Podcast tool) ran a scan across 635 recent podcast episodes covering QSR, fast casual, casual dining, and coffee. We narrowed it down to the 10 highest-signal conversations. The goal is to help you generate new ideas, sharpen your thesis on existing positions, and get a better read on what is actually happening at the unit level.

1. McDonald's CEO on Going Viral, the Big Arch and the Fast-Food Value War

Bold Names, April 8, 2026

This is the episode to start with. Chris Kempczinski lays out MCD's barbell strategy in granular detail. Sub-$3 value menu on one end, $8-10 Big Arch premium burger on the other. He walks through the chicken category pivot, explains why MCD's addressable market in chicken is 2x beef, and drops the line that franchisee cash flow rose in 2025 despite aggressive discounting. If you model MCD, this is required listening. The CEO's claim that MCD is gaining share with both low-income and upper-income consumers simultaneously is a direct challenge to the consensus view that value wars are margin-destructive.

2. McDonald's "Value" Beat — Morning Brief

Yahoo Finance Morning Brief, February 12, 2026

CFO Ian Borden right after Q4 2025 earnings. The critical data point: average franchisee cash flow increased year over year despite 15% EVM discounts and corporate co-investment. That is the single most important unit economics signal in the value war debate. Also covers management's GLP-1 risk assessment (dismissive but adapting menu) and the beverage platform launch timeline for late 2026. Host Brian Sazi pushes back on consumer staples broadly, which sets up an interesting tension. If the Street is underweight staples on GLP-1 fears, MCD may be mispriced relative to its defensive positioning.

3. The Cisco Skid, McDonald's "Value" Beat, Trump Tariff Setback

Squawk on the Street, February 12, 2026

Multi-analyst reaction to MCD's +6.8% US comps. UBS, Citi, Bernstein, and Barclays all weigh in. UBS calls it the best guest count gap relative to peers in recent history. But the real insight here is about Starbucks. Kate Rogers points out that SBUX saw traffic rise while maintaining premium pricing and avoiding discounts entirely. Both value and premium working simultaneously means this is a stock-picking market in restaurants, not a blanket sector call. Also contains a JPMorgan note on MCD's Google Edge AI partnership for voice ordering.

4. Economic Outlook for Restaurants

Restaurant Masterminds Podcast, January 30, 2026

If you only listen to one episode for sector-wide positioning, make it this one. Bloomberg Intelligence analyst Michael Halen covers MCD (bullish on E. coli comp laps), YUM/Taco Bell (bullish on new management), CMG (contrarian cautious on 2027 BOGO hangover), EAT/Chili's (bullish on ops plus marketing), and SG/Sweetgreen (bearish, calls it "absolutely insanity"). He also walks through AI kitchen monitoring (Dave's Hot Chicken case study) and loyalty analytics (Bikky platform, 18-day optimal messaging cadence). The contrarian CMG call alone is worth the listen. His argument: BOGOs are training consumers that the food is not worth menu price, and the brand damage shows up in 2027.

5. AI Valuation Fears, McDonald's Rises Despite Q3 Miss

Squawk on the Street, November 5, 2025

Jim Cramer's real-time reaction to MCD's Q3 miss. Contains the foundational consumer bifurcation framework that persists through 2026. The CEO quote: lower-income QSR traffic declining nearly double digits, higher-income traffic increasing nearly double digits. The $8 psychological ceiling thesis. And the underappreciated Hispanic spending collapse driven by immigration enforcement, affecting 16% of the population. If you cover CMG, CAVA, or YUM/Taco Bell, the Hispanic spending headwind is a variable most models are not capturing.

6. Office Market Meltdown Hits Small Business and Restaurants

Fast Casual Nation, April 28, 2026

The most current episode in the set and the most forward-looking. Three themes: AI drive-thru adoption accelerating (Presto AI claims 90% order accuracy at Dairy Queen, Carl's Jr., Taco Bell, White Castle), MCD entering the dirty soda/energy drink category to compete with SBUX and BROS at lower price points, and the office vacancy crisis creating a structural headwind for urban-heavy fast-casual (CAVA, SHAK, SG). Also covers the SBUX ChatGPT integration. Panel consensus: marketing gimmick, not moat. And a sharp bearish take on Jersey Mike's IPO at $12B valuation.

7. Bank of America's Investor Day, McDonald's Earnings Report

Bloomberg Intelligence, November 5, 2025

Bloomberg Intelligence analyst Red Brown provides the clearest articulation of the long MCD / short CAVA-CMG pair trade thesis. The argument: fast-casual $20 bowls are structurally pressured as higher-income consumers trade down into MCD. This directly contradicts JPMorgan's overweight and TD Cowen's buy on CAVA. If income bifurcation is accelerating, fast-casual names may be structural shorts, not buy-the-dip opportunities. November 2025 vintage but the thesis remains structurally relevant.

8. Why Chicken and Beverages are Surging — QSR Trends with Datassential

QSR Uncut, November 13, 2025

Datassential's Huy Do brings the receipts. Chicken segment growing 2x the rest of QSR. Specialty beverage market can support 20,000 additional U.S. coffee outlets. Plant-based protein in consistent and gradual decline. And the strongest contrarian CMG call in the set: the host argues CMG is actually a better deal than Chick-fil-A for family meals at $38, and the negative value perception is perception, not pricing reality. The "mobile app is the new dollar menu" insight reframes how to think about value delivery across the entire QSR sector.

9. Beverage Trends, Menu Moves, and the Research Powering Restaurants

QSR Uncut, February 12, 2026

Tomás Gilbert (Curion, ex-Jack in the Box) goes deep on qualitative trends. Hispanic consumer loyalty as a structural QSR moat. Signature sauce platforms as higher-ROI innovation versus new proteins. Mocktail premiumization beyond Gen Z. And the under-discussed driver behind Chili's turnaround: menu simplification (40% SKU reduction) was as important as the viral marketing but gets far less credit. If you are long EAT, the operational durability argument here strengthens conviction beyond the TikTok hype cycle. Also worth noting: beer in "tremendous decline" is a headwind for beer-heavy casual dining names.

10. Jollibee is Ready for its U.S. Growth Push

QSR Uncut, January 22, 2026

Private company focus limits direct actionability, but the data here has strong public market read-throughs. Jollibee's Peter Wright (ex-Starbucks/Panera) reveals $4.6M AUV on 107 units, which is 2.5x the median chicken QSR. Confirms trading-down from full-service to QSR is active. And the bone-in chicken contrarian thesis: bone-in drives differentiation and repeat rates versus the industry shift to tenders. If WING's bone-in focus is an underappreciated moat, this episode provides the supporting evidence.

We scanned 538 podcast episodes and pulled the 10 highest-signal conversations for TMT analysts covering AI infrastructure, semis, hyperscalers, and enterprise software. Here are the episodes worth your time and the 20 stocks driving the debate.

TL;DR

• Best TMT signal is showing up in long-form audio before it hits sell-side notes
• 538 podcast episodes scanned, top 10 highest-signal conversations ranked
• Coverage spans AI infrastructure, semiconductors, and enterprise software
• Plus bull and bear cases on the 20 most-debated stocks

TMT is built different

If you cover general equities, your research workflow has an established structure to it. Trailing cash flows fit into a DCF and standard multiples like EV/EBITDA give you a starting point. GAAP profitability tells you most of what you need to know about bottom-line trends.

If you are a TMT analyst, you don't have that luxury.

You are in a sector where traditional accounting punishes the best business models. A generalist might reject a company for negative EPS, but your job is to figure out whether massive capex and heavy R&D spend are building a 130% net revenue retention machine or simply burning capital. You are constantly translating unit economics, LTV:CAC ratios, and intangible assets into future operating leverage.

Your biggest challenge is not building the model but getting the right inputs before everybody else does.

TMT analysts value better signal, and there is a new source they should be scanning.

Why podcasts

Podcasts are quietly becoming one of the highest-signal research inputs in TMT. The best episodes can surface things that matter before they appear in sell-side research.

You hear CEOs talk through spending plans before they file the 10-Q and investors explain what they are buying and avoiding. You will also hear operators describe bottlenecks that are not in a model yet — power, memory, data centers, custom chips, software pricing, agent adoption, cloud share, free cash flow.

Our AI agent (matterfact Podcast) ran a scan across 538 recent podcast episodes and narrowed it down to the 10 highest-signal conversations for analysts focused on AI infrastructure, semiconductors, hyperscalers, and enterprise software. The goal is simple: help you generate ideas, improve your existing thesis on a current portfolio position, and get a better read on where the market may be too excited or too dismissive.

1. Catalyst with Shayle Kann: Inside Google's Massive AI CapEx

Start here. Shayle Kann's conversation with Amin Vahdat, Google's chief technologist for AI infrastructure, is one of the cleanest windows into how a hyperscaler actually thinks about the AI buildout. This is not generic AI talk. It goes into chips, power, labor, reliability, and the real tradeoffs behind data center scale.

If you cover Alphabet, Nvidia, Broadcom, Marvell, Vertiv, utilities, or power-exposed industrials, this is required listening. The takeaway: the AI race is an infrastructure race. The winner may not be the company with the best model. It may be the company that can source power, optimize silicon, manage reliability, and deploy capacity faster than anyone else.

2. RiskReversal: Peter Boockvar on the AI Semi Trade

Every long book needs a good bear case. Peter Boockvar makes one here. His argument: the semiconductor rally has become stretched, driven by front-loaded orders rather than durable demand. He walks through the AI beneficiary trade, data center hardware, hyperscaler free cash flow pressure, and the risk of a reversal if capital spending guidance changes.

Most TMT teams already know the bull case for Nvidia and semis. The harder work is understanding where it breaks. The key question: what happens if hyperscalers keep spending, but investors stop rewarding the spend?

3. Cheeky Pint: Sundar Pichai on the Future of AI at Google

This is the CEO-level version of the Google thesis. Pichai talks through Google's AI strategy, the capex budget, memory and power constraints, and the cultural shift inside the company. 2026, he argues, is a supply-constrained year.

For Alphabet analysts, the market still tends to reduce Google to one question: will AI hurt Search? That framing is too narrow. The better question is whether Google's full-stack advantage matters more as AI shifts from experiments to high-volume production. Google has chips, models, products, distribution, and cloud. That does not mean the stock is easy. It means the debate deserves more than a chatbot share chart.

4. No Priors: How Capital Is Powering the AI Infrastructure Buildout

Sarah Guo's conversation with Neil Tiwari of Magnetar is one of the better episodes on the money behind AI infrastructure. GPUs, data centers, long-term contracts, private credit, sovereign buyers, take-or-pay structures — these are now part of the core TMT research process.

The AI buildout is no longer just a technology story. It is a financing story. Analysts who do not understand how AI infrastructure gets financed are missing a large part of the picture.

5. Morgan Stanley: AI's $3 Trillion Question

Morgan Stanley's Thoughts on the Market puts the whole AI infrastructure cycle into a capital markets frame. The point: AI spending is so large that balance sheets matter again. That has implications for hyperscalers, software, REITs, private credit, utilities, chip suppliers, and industrial infrastructure.

If AI capex becomes a multi-year financing cycle, the winners may include companies far outside the obvious Nvidia trade.

6. 20VC: Anthropic, the Pentagon, Stock Picks, and the Data Center Arms Race

This one is opinionated and better for it. Jason Lemkin, Rory O'Driscoll, and Harry Stebbings debate stock picks, public software, data centers, and where AI is actually changing the market.

For software analysts, this is one of the strongest episodes in the set. The core question: can software companies reaccelerate with AI, or does AI compress seats, margins, and product value? That debate matters for CrowdStrike, Salesforce, Cloudflare, ServiceNow, Atlassian, Monday.com, Wix, Klaviyo, and many others.

7. The Circuit: CAPEXXXXXX and the Trillion Dollar Datacenter Race

Ben Bajarin and Jay Goldberg are very strong on semis and infrastructure. This episode connects cloud share, chip allocation, custom silicon, and capex plans in a way that most single-topic pods do not.

Amazon's AWS position matters. Google's TPU story matters. Nvidia's Vera Rubin allocation matters. Broadcom and Marvell matter because custom silicon is not a side story anymore — it is one of the main ways hyperscalers try to control cost.

8. Motley Fool Money: AI Capex Is Off the Charts

Less institutional than some of the others, but it states the simple risks clearly. The valuable part is the skepticism around companies using heavy debt to chase AI infrastructure demand. CoreWeave and Oracle are the obvious debates.

The question is not whether demand exists. The question is who carries the risk if demand is slower, financing costs stay high, or one major customer becomes too important.

9. Hedgeye: AI Capex, Back to Infrastructure

Hedgeye's Protect the Pile episode puts AI capex into a macro and portfolio risk frame. It widens the lens in a way most TMT pods do not: power, grid equipment, commodities, construction, cooling, data center real estate, industrial capacity.

Analysts who only look at chips will miss second-order winners. Analysts who only look at software will miss the stress building under some software business models.

10. Chip Stock Investor: Beyond AI Data Centers

This episode is about the second-order semiconductor names. Every TMT pod already covers Nvidia. The harder work is finding the suppliers that benefit from higher power density, more complex racks, timing, analog, power management, networking, and industrial recovery.

Names like Monolithic Power, NXP, Texas Instruments, Microchip, and Littelfuse may not have headline power, but they may tell us more about how broad the AI cycle is actually becoming.

The 20 stocks driving the debate

Across these ten episodes, the same names keep showing up — but the bull and bear cases sit on top of each other. Below is the large-cap version of the debate sheet, distilled. The SMID-cap counterpart is available as a download.

Large-cap: where the debate is loudest

Read the corpus, not the episode

Ten episodes, twenty stocks, one debate that is moving faster than sell-side notes can keep up with. Each episode is a useful hour. Read together, they do what no single hour does: they let you weight the views — Vahdat against Boockvar, Pichai against the chatbot-share narrative, Goldberg against Jensen — against each other, and against the actions of the people writing the largest cheques.

That kind of synthesis is what TMT analysts keep asking matterfact to do for them. Not "summarise this episode" — "tell me what 538 episodes of serious AI infrastructure coverage add up to, with every source one click away."

Top funds treat podcasts as a primary research input. 200M hours of expert conversation, often more candid than compliance-monitored calls, should be an input into your own investment research.

TL;DR

• Mining podcasts is becoming a critical input into the fundamental investment research process
• AI makes it possible to mine 200 million hours of expert conversation
• Aggregated into actionable insights on demand

Why podcasts and why now

Podcasts have exploded in popularity in recent years, but you may not have heard how elite investors and money managers are using them to mine for investment insights at scale.

Here is just one example out of thousands that we unearthed using matterfact's Podcast tool. It is AI agent that sits on top of millions of podcasts and acts like a research assistant who can build high-quality, polished, and ready to use research reports.

Consider this: Between November 2025 and April 2026, CoreWeave ($CRWV) CEO Michael Intrator shared the same point across several podcast and CNBC appearances. But each time, he added a little more detail.

In November, he said the company had started the year with 80% of its backlog tied to a single customer. By early January, that number had dropped, and no customer represented more than 30% of the book. Later in January, he put the figure at 32%. By April 10, he said it was under 35%, and publicly named Meta, Anthropic, NVIDIA, and OpenAI as anchor customers.

These granular details matter to investors. Over 23% of the outstanding float has been sold short (According to NASDAQ and CapIQ) and for short sellers, customer concentration presents a risk. It also matters if you were long NVDA and trying to understand how durable demand from neocloud players might really be.

What is interesting beyond the content is how and where the story unfolded.

Few investors could follow it fully because it did not show up in one filing or an earnings call soundbite. It came out over time, across several interviews, podcasts, and appearances that most investors would simply not have the time to listen to or read.

The funds that picked up on it did not necessarily have access to better people via an expert network, but they did have a better process for listening. At scale.

Until now, podcasts were treated as entertainment, and optional listening for money managers. Interesting, sometimes useful, but not central to the research process at any serious hedge fund. Well, that is now changing and mining for insights from podcasts at scale may become as ubiquitous as using expert networks. Some of the most candid, specific, and forward-looking commentary in the market now shows up in long-form audio before it appears anywhere else.

The problem is, if you are not systematically data mining that resource, there is a good chance you are missing critical information and competing against people who are.

How podcasts went from entertainment to investment tool

First, the quality of the format improved. Podcasters are "monetizing" this content and are after higher quality, longer and more in-depth interviews that attract more views and sponsorships & ad revenues. The guests are incentivized to provide quality content for a different reason. For many executives, founders, and specialists, podcast appearances are now part of how they build their own reputation and personal brand. They prepare diligently, explain things clearly and in detail, and they bring specifics to back up their claims and appear highly competent. And (this one is critical) because the format is longer and less formal than an expert network call, they often say more than they would in traditional investor communication channels or on a one-to-one call with an investor.

Second, the quality of guests is extremely high. If you scroll through investing podcasts, for instance, you will see a mix of practitioners, operators, subject matter experts (SMEs), and hedge fund managers that you would have never seen even five years ago. They were known for keeping strategy close to the vest. Many of these people are difficult to access directly, but surprisingly willing to spend an hour speaking in public on a pod. This is not only true of investing but every other industry as well. Noam Brown (From OpenAI) went on the Redpoint podcast to explain reasoning as a new paradigm of scaling. World-renown economists are commenting on the war in Iran and leading experts in cloud computing are discussing market-moving innovations.

Third, the specialist layer got much deeper. If you are looking to unearth insights that are not common knowledge on Wall Street, the real opportunity is often in niche podcasts focused on specific industries like semiconductors, freight, biotech, industrial automation, healthcare IT, energy infrastructure, specialty chemicals, and dozens of other verticals. That is where you start hearing informed people talk about things that truly matter for companies and whole sectors.

To ground that in data: across the episodes we index, Information Technology and Real Estate dominate the conversation, but every GICS sector has meaningful coverage — hover any bar to see the exact share.

Episodes by GICS Sector Share of indexed podcast episodes that mention a listed company, grouped by the company's GICS sector.

So the question is no longer whether useful signals exist in podcasts, it clearly does. The issue is volume and the sheer scale of content out there.

The scale problem with using podcasts for investment research

There is an enormous amount of audio in the world. On top of the millions of podcasts out there with hundreds of millions of hours of recorded conversation, thousands of new episodes are published every day.

No analyst (or even a team of analysts) can keep up with that volume manually. The bottleneck is the ability to search, filter, compare, and retrieve the insights that matter to you and your org and it's akin to finding a needle in a million digital haystacks.

Thankfully we have AI. Now, instead of listening at 2x speed for hours in the hope of finding one useful detail on a name in their portfolio, an analyst can simply ask a question and query the full corpus. It is an AI agent that can pull the relevant data, compare discussions across hundreds of sources, and let the analyst spend their time doing what actually matters: interpreting the signal.

We are seeing managers use this data in two distinct ways. Idea generation and building a better investment thesis. We are sure more exist, like red flag detection and risk management, but will focus on the two common use cases for now.

Use case one: idea generation

Every investor works with some version of an investable universe and a certain concentration threshold. They wake up every day thinking about the next stock to buy, either to replace a name that is on its way out, or to get exposure to a new theme.

Podcasts can be a surprisingly good resource for coming up with new investment ideas.

There are "whispers" about NVIDIA's AI strategy and the cost of new GPU clusters. There is talk about how legacy tech companies are becoming exceptionally cheap relative to the new commers. A former product lead at a big AI company explained why a platform shift will likely pressure margins across an entire category. One podcast is interesting but millions begin to formulate a theme, and an investment opportunity.

These things happen constantly on podcasts and most of the new ideas go unnoticed by wall street, save a few savvy investors who know how to mine these insights.

Here is another example. On the All-In podcast in March 2026, Perplexity CEO Aravind Srinivas cited OpenAI CFO Sarah Fryer on a point that has implications for how investors think about the economics of AI in general: the cost of a million output tokens had fallen from roughly $32 at the GPT-3 launch to roughly $0.09 today. That is a dramatic decline especially given that much of the reduction came from software rather than hardware alone.

Just imagine the questions you could ask of this vast body of data to help you identify your next idea. The question itself becomes your value. What would you ask? Send us your question and we will (privately) reply with a full report.

Use case two: thesis enrichment

Idea generation is valuable, but building a bullet-proof thesis may be the more important use case, especially for analysts pitching a new idea to the PM.

Analysts already know how to build a DCF model and form a view and know exactly how to present it to their portfolio manager. Where things get harder is substantiating that view with real world evidence or even having someone poke holes in it before they make the pitch to the PM.

Expert networks were one of the main ways funds tried to bridge that gap; and they still have value. But they are not without limitations.

Experts networks are not cheap and experts are warned about saying too much. On a formal, compliance-approved call, people tend to be careful. On a podcast, especially a friendly one, they are often looser, more opinionated, and more willing to explain what they actually think is going on. Also, investors will often need to speak to multiple experts about the same theme - this takes time.

Mining podcasts for alpha is much cheaper and provides instant value, drawing from a large pool of experts. It can give you a substantiated bull or bear case on almost any public company.

Take the current debate around GPU depreciation and asset life in AI infrastructure. One version of the bearish case says that GPUs are likely to become obsolete quickly, which means today's aggressive capex could age badly and lead to painful write-downs later. That is an ongoing debate with experts arguing both sides, and investors need to consider both sides of it when crafting their thesis.

Podcasts are still underused as an investment research tool

What makes this especially interesting is that the market has not taken advantage of all the insights hidden in podcasts yet.

A handful of sophisticated firms have, but most small and mid-sized funds still do not have a proper workflow for it. They are not systematically indexing, querying, comparing, and extracting signal from the medium and that creates an opportunity.

We have seen something similar with credit card data and other alternative data sets. A new source of alpha rewards investors who get serious about it before everyone else does. First it was satellite data, then transaction data, app data, web traffic, or earnings call sentiment parsing. Today, one of the most underappreciated sources of differentiated public signal is long-form spoken content.

Because it is too large to manage manually and still not widely integrated into research workflows. That is exactly the kind of inefficiency that tends to persist and offer ample alpha for early adopters.