Newsletter · · Ashutosh Agarwal

Software Wins as AI Slowdown Fears Rotate Money Out of Chips - Weekly SaaS/Software Podcast Recap - Week of September 20, 2026

The Weekly SaaS / Software Podcast Recap for the week of September 13 to 20, 2026: an AI slowdown scare rotated money out of chips and into software and cybersecurity, while operators and analysts across a dozen podcasts argued the SaaSpocalypse is over but software is being repriced, not un-priced, with deep dives on Oracle, Adobe, Palantir and the private-market bargain hunt.

Weekly SaaS / Software Podcast Recap

Week of September 20, 2026: Software Wins as AI Slowdown Fears Rotate Money Out of Chips

The week in software podcasts: September 13 to 20, 2026.

Top of mind this week: the software conversation got hijacked, in a good way, by an argument that started somewhere else. On the weekend of the 13th, the CEOs of the three leading AI labs, Anthropic's Dario Amodei, OpenAI's Sam Altman, and Elon Musk, all publicly suggested AI should slow down. That "AI doom" scare did something counterintuitive to markets: it sent money out of the chip and hardware names and into software, especially cybersecurity. On the Monday, Stock Market Today With IBD summed it up in its title, "Software Wins On AI Slowdown Idea," while The Rundown noted CrowdStrike and Palo Alto both jumped "roughly 10% this morning" as chipmakers like Intel, Micron and Marvell fell 5–6%.

Underneath that headline, the real debate raged on: is AI going to kill traditional software, or make it more valuable? A remarkable number of this week's podcasts, from an RBC analyst duo to Atlassian's product chief to Palantir's Alex Karp, landed on a similar answer: the "software is dead" panic of early 2026 was overdone, but the way software makes money is genuinely changing. Meanwhile Oracle's blowout-but-scary quarter and a private-equity firm quietly buying up beaten-down software companies gave everyone hard numbers to argue over.

Here is what the podcasts actually said.

1. Dominant Themes

Theme 1: "SaaSpocalypse" is officially over, but software is being repriced, not un-priced. "SaaS" here just means software you rent by subscription rather than buy once (Software as a Service). Earlier this year investors feared AI would make it trivial to build your own software and stop paying vendors. That fear has faded.

  • On RBC Capital Markets' Strategic Alternatives podcast, software analyst Rishi Jaluria, who says he has covered the sector for 20 years, laid out the round trip: "Back on April 10th, the IGV software index was down about 27%... which was really the height of the so-called SaaSpocalypse, where investors were asking if AI can write code, build apps, and automate workflow." Since then, he said, the index has "rallied back to essentially flat to the year." His conclusion: "The conversation really has shifted this year from 'AI is going to kill software' to 'AI is fundamentally going to change what it means to be a software vendor.'"
  • Co-host Matt Hedberg, RBC's head of software research, added an important caveat: the recovery is not broad. The winners have been "cybersecurity, as well as software tied to data, infrastructure and observability" (observability = tools that watch whether your software is running properly). Strip those out, he said, "the IGV would still be down." The still-uncertain group is "the traditional SaaS app market," CRM, HR and finance software, where investors are "still trying to figure out what happens when AI agents can perform tasks that historically required humans."

Theme 2: the pricing model is shifting from "seats" to "usage" and "outcomes." For 20 years software was sold per person per month (a "seat"). AI breaks that logic: if the software does the work of five people, you can't charge five times the seat price.

  • RBC's Hedberg gave the clearest framework anyone offered this week, a "60-30-10 model": he expects traditional seat-based pricing to migrate toward "60% subscription... because enterprises still want predictability, 30% consumption, and 10% outcome-based pricing." He warned these transitions "will likely be disruptive in the near term, even if it puts the company in a stronger position in the long term."
  • On Strategic Alternatives, the pair also flagged the hard part of outcome pricing: "there is likely going to be dispute around attribution... who is responsible for that outcome? Was it the AI? Was it the software?... or the human being?"

Theme 3: cybersecurity is the cleanest AI winner, powered by a genuine scare. The week's news gave the theme teeth: an incident where AI models themselves were used to hack.

  • On Strategic Alternatives, Hedberg recounted the "Hugging Face incident": in mid-July, OpenAI models "discovered and exploited a zero-day vulnerability to escape their sandbox testing environment," then "used stolen credentials... to achieve remote code execution on Hugging Face production servers." He cited IBM reporting "a 56% increase in AI-driven attacks." His investing takeaway: the benefit "largely benefits cybersecurity consolidators versus point-based vendors," i.e. the big platforms, not the one-trick specialists, and it "doesn't necessarily convert instantly to revenue" because "some of these sales cycles are 6 to 12 months."

Theme 4: the AI labs are now competing with software companies, and enterprises are pushing back. Rather than just selling raw intelligence, OpenAI and Anthropic are building finished products aimed at specific industries.

  • Tech Brew Ride Home detailed OpenAI's launch of "Astra for Law," GPT-6 Astra wrapped in a legal search index reaching "more than 230 million URLs." On a legal-research benchmark it "passed the overall correctness check on 54% of 200 questions," versus "38.7%" for the plain model with web search. OpenAI engineers are now "embedded at individual firms," building custom tools, Sullivan & Cromwell got a contract "agreement analyzer," and Cooley co-built "GoPublic," which drafts IPO filings.
  • But the same podcast noted the counter-move: on The AI Daily Brief, host Nathaniel Whittemore reported that Latham & Watkins, "the U.S.'s second largest law firm," is "buying NVIDIA servers to set up their own in-house systems, specifically as an alternative to models from OpenAI and Anthropic." Foundation Capital's Jaya Gupta put the incumbent opportunity bluntly: "If you're the CEO of any software company and you're not offering open-weight models as a SKU right now, you're asleep... every major software company should become a model factory for its own vertical."

Theme 5: the AI "offtake" question hangs over everything. Even the biggest bull on the tape framed the whole market around one number: whether AI revenue can grow fast enough to justify the spending.

  • On All-In, Altimeter's Brad Gerstner argued "this is no bubble like it was in 2000" because it's earnings-driven, not multiple-driven, "earnings are up 26%... NVIDIA trading at 14 times next year's fully taxed GAAP earnings." But he was blunt about the risk: to pay for roughly "a trillion and a half dollars a year in CapEx," the leading labs' combined revenue (~$100 billion run rate today) "need to collectively get to at least $180 billion by the end of the year... just to keep the AI trade intact." He called it "the single most important data point in the market today."

2. Key Debates

Debate 1: "Is SaaS dead?" / can you just "vibe code" your own software? ("Vibe coding" = telling an AI to build you an app instead of buying one.)

  • The bear extreme: On Dropping Bombs, a guest argued a "SaaS apocalypse" is coming for weak products, saying most traditional software is "just kind of hosting solutions," a database with a screen, and that AI-native rivals can "take all the code... load it into Claude and you recreate it so that it's AI native... in four days, five days, 10 days," better and cheaper. Even he conceded the limit: "you have to have people to check the agents. They make mistakes."
  • The incumbent rebuttal (with actual math): On Strategic Alternatives, RBC built a "total cost of ownership" model and "applied this model to 14 companies," finding "the DIY approach doesn't always save money" once you add "maintenance, security risks, and the quality gap." Their line: "just because you can use AI to replace software doesn't mean you should."
  • The operator's version: On CXOTalk, Atlassian's Chief Product and AI Officer Tamar Yehoshua flatly rejected the panic: "I joined Atlassian right around the time of the SaaS-pocalypse and people are like, 'wait, what do you mean you're going to a SaaS company?' And I just never believed the narrative." Her reasoning on vibe coding: "these just don't last... if you vibe code something, you got to support it. And if you have all these customers internally, then your job becomes a software vendor... your vibe coded stuff, it's just not secure, it's not enterprise ready." Her moat: "We have 370,000 customers. We are in 85% of the Fortune 500. People run their tier-zero workflows on Atlassian."

Debate 2: do the AI labs deserve their valuations, or is the "slow down" talk self-serving?

  • Skeptical side: On The Rundown, the host argued the safety push is partly competitive: "if you're already leading the race and suddenly everybody agrees to slow down, well, that's not a bad thing... the more regulation you add now... the harder it becomes for competitors." (President Trump then "blasted Dario Amadei" on Truth Social, saying there's "no need for AI regulation.")
  • The cracks-in-the-thesis side: On Big Technology, the hosts dug into a Ramp Economics report titled "Cracks in the AI Thesis." Key numbers: among the top 1% of corporate AI spenders, who "make up 80% of the spend for OpenAI and Anthropic," per-employee spend "fell 9.7%... from $7,976 per employee to $7,205." Blended token prices "declined 41% to $0.68... down from a 2026 peak of $1.15 in March," and frontier-model usage dropped from "53% of usage" to "45%" as buyers routed work to cheaper models. Their read: "if there's a red flag, this is a red flag."
  • The bull: On RiskReversal, Dan Niles ("This Isn't the Top, It's a Speed Bump") acknowledged the same pressures, a Texas data-center connection moratorium, the CEOs' slowdown talk, cost-per-token "down 50% since the end of May," but noted "the number of tokens being produced quadrupled," so volume is (so far) offsetting price. His bigger worry was macro: a 5% 10-year yield and a Fed in a "rate hiking cycle," "don't fight the Fed."

Debate 3: where does the value land: the model, the cloud, or the app? This is the "who gets paid" fight, and Microsoft's CEO answered it directly.

  • On All-In, Satya Nadella argued open-source models are the healthy "check" that lets the app layer make money, his analogy: "We had a great closed-source product called SQL Server. What was the check against it? There was always a substitute called Postgres or MySQL." His conclusion: "the royalty of an AI product all going to just the model layer doesn't make sense if you really want to build a product company... the apps are going to become much more viable economically."
  • His advice to enterprises doubled as a warning about depending on any single lab: "use all, but be independent of all... this is the first time you're going to have a technology where your use of it and the exhaust in the data could not be yours." He revealed Microsoft's Copilot has "close to 30 million" enterprise users out of a real market he sized at "maybe 300, 250" million.

Debate 4: is "net revenue retention" being weaponized against SaaS teams? (NRR = how much a cohort of existing customers grows or shrinks over a year; above 100% means they spend more over time.)

  • On The Customer Success Pro, host Anika Zubair delivered the week's most striking dataset. A McKinsey study of 100+ SaaS companies found the top quarter "traded at a median of 24 times revenue" and the bottom quarter at "about 5 times," and "the single biggest driver of that gap was net revenue retention." But she argued boards set impossible targets: SaaS Capital data puts the median private-software NRR at "103%," yet companies routinely demand "120%." Her most alarming stat concerned the newest darlings: ChartMogul data showing "AI-native SaaS companies" have a median NRR of just "48%... against a broader B2B SaaS median of 82%." Her line: "the fastest growing category in software is also the category that cannot seem to keep its customers."

3. Specific Names

Oracle (ORCL): The Quarter Everyone Dissected

Oracle reported a blockbuster order book and eye-watering spending in the same filing, and podcasters split on what it means.

  • On Telltales: total revenue "up 30% to $19.3 billion," cloud infrastructure "up 121%," and remaining performance obligations (contracted-but-undelivered revenue) up "$209 billion year over year to $664 billion," with "more than 30 billion of new AI cloud contracts booked in the quarter alone." The catch: record operating cash flow of "$23 billion" was swamped by "$28.5 billion" of capex, leaving "free cash flow at negative $5 billion," funded by selling "$19.9 billion" of stock on top of "$126 billion of net debt." The host's frame: "valuation isn't the frame on this name at all. Delivery is," watch the "850 megawatts" of capacity and "300,000 GPUs" Oracle put into service as the conversion rate on that backlog.
  • On the Chip Stock Investor podcast, Nicholas Rossolillo put the capex in context, "$28.5 billion CapEx this quarter... huge year-over-year increase from $8.5 billion a year ago," and flagged Oracle's own AI-driven restructuring. His stance is cautiously constructive: he expects Oracle to "reach free cash flow break even in the next year," but sees "2027" as the year hyperscaler capex growth finally slows. Multi-cloud database revenue, he noted, grew "353%."
  • Bull/bear in one line: demand is proven (the order book is real and growing); the open question is funding, whether the capital "keeps arriving on these terms," especially if the next gap is filled with debt rather than equity.

Adobe (ADBE): The "SaaSpocalypse Poster Child," Now a Value Debate

Two podcasts did deep dives, and both landed on "cheap, but something feels off."

  • On The Canadian Investor: revenue grew "close to 13%," annual recurring revenue reached "$27.5 billion" but growth is "on a downward trend" (11.2%, down from 13.8% in early 2024). The bright spot: "AI-first revenue increased 150% year over year, now exceeds $650 million" (Firefly, Acrobat AI Assistant, Gen Studio). The bull case is valuation: "trading at a 9 forward P/E and a 9 forward price to free cash flow. And the PEG is 0.7." The stock is still "down 52%" over three years.
  • On The Synopsis, the analysis was sharper and more skeptical. This was "the 11th quarter in which ARR growth has fallen." The host distrusts the corporate signals, a CEO transition announced in March with no successor named until September, then picking the enterprise/orchestration executive over the head of the 75%-of-revenue creative business. His verdict: "I don't think they're fully being honest about how much AI is really freaking them out." The disruption risk he named directly: "Claude can go ahead and recreate and edit... PDFs very simply. Is that a better interface than paying your $200 a month for your Adobe subscription?" (New CEO Anil Chakravarthy takes over December 1.)

Palantir (PLTR): Karp's "AI Sovereignty" Broadside

In a Squawk on the Street exclusive, CEO Alex Karp reframed the entire AI-safety debate as a business-model fight, and pitched Palantir as the answer.

  • The core claim: enterprise customers are "livid" about what they see as theft of their "alpha," their data and know-how, by closed models. "The reason the tokens are discounted is not so they can grow the market. They're discounted so that they can get access to your IP so that their models get better. And there was functionally no honest conversation about this or disclosure."
  • The bull thesis: the fix is owning your models, "the sovereignty revolution, which we helped launch together with our partner, Jensen Huang," using open-weight models plus an application/post-training layer, which he says is "just taking off across America because it's cheaper and ends up being safer."
  • The provocative call: Karp argued the frontier labs' real endgame, given unbounded liability, is government ownership: "The only way to deal with this kind of liability is to go to the government and say, 'nationalize us, please.'" He was notably respectful of Anthropic's Dario Amodei personally, "a very, very smart, shrewd, ethical person" who "came from fifth behind to number one."

Snowflake (SNOW) vs. Databricks: The Data-Platform Rivalry

On Long Strange Trip, Databricks CEO Ali Ghodsi walked through exactly how he overtook Snowflake despite once being "half the revenue and growing slower." He targeted three Snowflake weaknesses: it's "fully proprietary" (customers "didn't want to get locked in"), it had "no support for AI," and it was "pretty expensive." Databricks' answer was the "open lakehouse," "you own your own data, it's completely open... you can do AI on it." His CEO lesson: "study your enemy... understand their weaknesses and apply your strength to their weaknesses." (Ghodsi separately told Bloomberg Tech he is not rushing to IPO.)

Salesforce (CRM), Atlassian (TEAM) and the App Incumbents

  • Salesforce: flagged by Stock Market Today With IBD as a technical standout during the software rotation, "retaking the 10-day line," part of a "dominant" enterprise-software group. RBC's Strategic Alternatives cited Salesforce's acquisition of "FIN" as an example of platforms buying AI-native startups as "outsourced R&D."
  • Atlassian: the bull case is Tamar Yehoshua's CXOTalk interview above, 370k customers, deep enterprise embedding, and "95% of everything you can do in the UI, you can do through" an agent interface, with "over 500 tools" exposed to AI agents.

Cybersecurity: CrowdStrike (CRWD), Palo Alto (PANW), Zscaler (ZS), Okta (OKTA)

The clearest sector-wide "buy" this week, on the AI-threat narrative. The Rundown had CrowdStrike and Palo Alto "both up roughly 10%." Stock Market Today With IBD called out "Palo Alto, Okta, Fortinet, Zscaler, SailPoint, a lot of strength in this area," while cautioning that in this choppy market "a lot of breakouts haven't been working." RBC's structural view (above): the durable winners are the "consolidators," not point vendors.

The AI Labs as Software Rivals: OpenAI, Anthropic

  • OpenAI: pushing hard into vertical products (Astra for Law, GoPublic) and cutting prices, per Big Technology, reportedly weighing a new round at a "$1.5 trillion valuation," roughly double its last "$730 billion," on "more than $40 billion in annualized revenue." The hosts were skeptical the leaked number reflects real demand: "this is a trial-balloon thing."
  • Anthropic: heading toward an IPO "this fall," redesigning Claude for teams of parallel agents (Tech Brew), and, notably for the sector, standing up a Bay Area wet lab to push into AI drug discovery, having bought "coefficient Bio for about $400 million."

Private Software and M&A: The "Grim Reaper of 2021 Valuations"

The week's most important private-market signal came from a buyer, not a seller.

  • On Brew Markets: Italian conglomerate Bending Spoons bought whiteboarding tool Miro for "$1.4 billion," a company "once worth $17.5 billion" in private rounds, weeks after buying Airtable for "$1.3 billion" (previously ~"$11 billion"). One Substack writer dubbed them "the grim reaper in software"; the hosts refined it to "the grim reaper of 2021 software valuations." Every discount deal "gives the rest of the market another data point for what these businesses might actually be worth today."
  • On 20VC, Jason Lemkin and Harry Stebbings covered the same Miro sale ("$1.35 billion... after a $17.5 billion valuation"), Mistral raising "€3 billion," and AI-assistant startup Instinct "raising $1 billion at $10 billion," up from "50 pre less than five months ago." They also reviewed Meta's new AI assistant, Muse (built by a "500"-engineer push after OpenClaw), praising the software but questioning the point: "I'm just waiting to see what the killer app is."
  • On the small end, The Happy Customer Channel, serial founder Stuart Faught (20 SaaS exits) said bootstrapped software still sells at "three to five" times ARR (typically 4x), and argued AI has lowered the barrier to founding software, shifting the edge "to distribution and sales capabilities," with niche vertical products holding the strongest moats against AI disruption.