Newsletter · 2026-09-07 · Ashutosh Agarwal

Cognition Raises at 46 Billion as Agent Ready Software Draws the Biggest Checks - The Raise - Week of September 7, 2026

The Raise for the week of September 7, 2026: venture money poured into software that acts rather than answers, with Cognition priced at roughly 46 billion dollars, Clay at 7 billion and Linear and Instinct at 2.5 billion each, while Anthropic lined up a 15 billion dollar pre-IPO credit facility.

The Raise

Week of September 7, 2026: Cognition Raises at 46 Billion as Agent Ready Software Draws the Biggest Checks

The startups that AI agents themselves reach for drew this week's biggest checks: a coding company at $46 billion, a sales-data tool at $7 billion, an assistant you simply text at $2.5 billion, while Anthropic quietly loaded a $15 billion credit card ahead of what could be the largest tech IPO ever. Plus two founder stories: a woman who heard 70 no's before launching GoHenry, and five scientists who turned a hunch into an $838 million Merck deal.

After a quiet stretch, the venture chatter came roaring back this week. The through-line: investors are paying enormous prices for software that does things rather than just answers questions, and, increasingly, for software that other AI agents choose to plug into. The clearest sign of the mood was a single episode of The Twenty Minute VC where the hosts rattled off four separate rounds, Cognition at $46 billion, Clay at $7 billion, Linear at $2.5 billion, and an AI assistant called Instinct at $2.5 billion, almost as an afterthought. Here's what the podcasts covered in the week of August 31 to September 7.

This Week's Rounds

Cognition, around $1 billion at roughly a $46–47 billion valuation. This is the maker of the AI software "engineer" Devin, and the number is the eye-popper of the week: its valuation has jumped from about $26 billion in May. What's driving it is real revenue. The company is currently doing an estimated $800–900 million in annualized revenue (annualized means the latest run-rate stretched out over a full year) and is expected to finish the year at about $1.6 billion. The striking part, as the 20VC hosts noted, is that Cognition isn't even the clear number one in AI coding, since Anthropic's Claude Code and Cursor are ahead, and it's still this big. Jason Calacanis called it "the Postmates of the category": not the leader, but riding a market so large that being number three is enough. His math on why the hosts underestimated it: "people are literally building 100X more software than we were 18 months ago," so the total market for coding tools is far bigger than anyone modeled. (The Twenty Minute VC, "NVIDIA Crushes Quarter and Buys Hugging Face | ... Cognition Raises at $46BN, Linear $2.5BN and Clay $7BN," Sep 3, 2026; Bloomberg Tech, "Nvidia Nears $14B Hugging Face Deal," Sep 2, 2026.)
Clay, raising at a $7 billion valuation. Clay is a sales-and-marketing data tool (it helps companies find and enrich lists of potential customers). For scale, the hosts pointed out that the incumbent it's chasing, ZoomInfo, is worth about $1 billion, so Clay is being priced at seven times its established rival. The bull case is Calacanis's "agent-friendly" thesis: increasingly it's not humans picking software, it's AI agents choosing which tools to build with, and they pick whatever plugs in most cleanly. As he put it, "you can't take [an] agent out to a steak dinner to get him to buy your product. It just has to be better." (The Twenty Minute VC, "NVIDIA Crushes Quarter and Buys Hugging Face | ... Cognition Raises at $46BN, Linear $2.5BN and Clay $7BN," Sep 3, 2026.)
Linear, a tender at a $2.5 billion valuation, having just crossed $100 million in annual revenue, growing 100% a year. (A "tender" is a deal that lets existing shareholders and employees cash out some shares rather than the company raising fresh money.) Linear makes project-management and issue-tracking software for engineering teams. The thesis for paying up: if every developer is now shipping vastly more code with the help of AI agents, teams need a new "system of record" to keep track of it all, and Linear is winning that race while the old guard (the hosts singled out Asana) fades. (The Twenty Minute VC, "NVIDIA Crushes Quarter and Buys Hugging Face | ... Cognition Raises at $46BN, Linear $2.5BN and Clay $7BN," Sep 3, 2026.)
Instinct, a $350 million round at a $2.5 billion valuation, roughly five times the ~$500 million it was worth just weeks earlier. Instinct is one of the buzziest consumer launches of the year, and the company is only four or five months old. The product is deceptively simple: you don't download an app, you just text or call a phone number, and the AI on the other end actually does the errand, books the restaurant, finds your buried frequent-flyer number, calls your payroll company on your behalf using a synthesized voice. Founder Noah Shin has early backing from Bret Taylor (the former Salesforce co-CEO and current OpenAI chairman). The "magical moment" is real, one user reportedly sent 10,693 texts to the AI in 11 days, and another used it to buy a house, but so are the risks: the terms of service appoint Instinct as your legal agent, meaning actions it takes are "binding as if entered into directly by the user." The TBPN hosts were both dazzled and wary: "back in my day, when a hot consumer tech company caught a little bit of attention, they would get a 25 [million] on 150 [million] from Benchmark... it's just crazy to see this quickly a company get the $2.5 billion round." A B2B rival for the same "AI that acts" space, Town, has been funded at around $1 billion, co-led by Index Ventures and Benchmark. (Elon Musk Podcast, "Texting an AI to run your life," Aug 31, 2026; TBPN, "WE'RE BACK, Meta Addiction, Tim Cook's Last Day," Aug 31, 2026; The Twenty Minute VC, "NVIDIA Crushes Quarter and Buys Hugging Face...," Sep 3, 2026.)
HiBob, $166 million led by Salesforce, lifting its valuation to $3.2 billion. HiBob makes HR software (payroll, org charts, headcount data) for mid-sized companies; the round brings its total funding to about $700 million, and the company said it will push back its planned IPO to at least 2027. The logic behind Salesforce writing the check is the interesting part: as companies deploy AI "agents" internally, those agents need to know who reports to whom, who's paid what, and what each person's job actually is, so the workforce data HiBob sits on becomes fuel for Salesforce's Agentforce ambitions. As the Chad & Cheese hosts put it, HR "used to basically be an org chart... because of AI, it becomes an intricate part of the entire thing." One host's read on why Salesforce chose HiBob over pricier targets like Deel or Rippling: it may be "the Dr Pepper of this whole thing," not the obvious number one, but a sound bet at a better price. (The Chad & Cheese Podcast, "HiBob Beast Mode & DOL Dismantles Worker Protections," Sep 4, 2026.)
Regent, a $240 million Series B. Regent, based in Rhode Island, builds "sea gliders," all-electric craft that skim just above the water, sitting somewhere between a boat and a plane. Founder Billy Thalheimer walked through what makes the raise unusual for a company that is still pre-revenue: real orders are already in the door. Regent has about $10 billion in commercial backlog, and crucially those are firm orders with over $10 million in non-refundable deposits already banked, customers putting money down years ahead of delivery. On the defense side it's up to about $20 million in orders, mostly from the Marine Corps. The company is now ~120 people and is filling out a quarter-million-square-foot factory to build its larger "Viceroy" and smaller "Squire" gliders. Batteries give a few hundred miles of range (fine for commercial coastal hops), so for military missions across the Indo-Pacific, "we need to go from Guam to the first island chain," Regent is developing a hybrid powertrain. (TBPN, "WE'RE BACK, Meta Addiction, Tim Cook's Last Day," Aug 31, 2026.)
Infinimmune, a $75 million Series A, co-led by existing backer Playground Global and Regeneron Ventures, with RA Capital and Merck's venture arm joining. The biotech uses the human immune system itself as the starting point for discovering antibody drugs (most rivals start with mice or synthetic libraries). The round takes its first two eczema programs toward clinical trials in 2027 and brings total funding to roughly $90 million. It's also this week's second founder story, see below. (Raising Biotech, "S4, E8: Infinimmune, From a Scientific Hunch to an $838M Merck Partnership, with CEO and co-founder Wyatt McDonell," Sep 3, 2026.)

Founder Story of the Week

Geraldine Lionnet, founder of GoHenry, she started at 47 and collected roughly 70 no's before the money came (Screw It Just DO It with Alex Chisnall, "She Started GoHenry at 47. Then Came 70 Rejections," Sep 1, 2026.)

Most of this week's founder talk was about AI. The most human story was about teaching kids to handle money, and about how long it can take to hear "yes."

Geraldine Lionnet spent a long career in e-commerce and operations, "the day-to-day running of the website, the pick, pack and dispatch, warehousing, logistics," before starting GoHenry, a service that gives children their own debit cards while letting parents set limits and teach them to save. The idea came from a flat "no": people in financial services kept telling her that children simply couldn't have debit cards. Her instinct, honed over years of operations work, was to ask for the actual rule: "it doesn't say children can't have debit cards. Technically, the parent is the owner of the account. And on that basis, we can issue debit cards and they can be used by the children." She frames it as a lesson that applies to any industry: when someone says "you can't because the regulation says so," ask to see the exact clause, because "it very often doesn't say that at all. It's just it has become the norm to interpret it in a particular way."

The first raise took about 18 months and a brutal amount of rejection. Asked how many no's she'd racked up over the whole 14-year journey, she didn't hesitate: "Probably 60, 70." One round alone took 103 separate pitches. And the worst moment came when she thought she was over the line: a family office had agreed to put in "a fairly substantial amount," and she travelled to London believing she was going to sign the paperwork. Then one of the family's sons, flown in from abroad, "didn't like it. And they pulled the plug." Roughly nine of those 18 months were spent going back out to replace the couple of hundred thousand pounds that vanished that day. "It was not a good journey home that day," she said, on the train home.

She had set a target of £750,000 just to launch. The company needed to plug into banking systems and survive a few months past launch before going back for more, and, as she put it, "none of it went to plan, of course. We ran out sooner than we thought we would." There were later stretches she called, memorably, "squeaky bum time," the nights of wondering "are we going to be able to pay salaries this month?" She tells a lovely story about a school-gates acquaintance who asked how her "little business" was going; Geraldine mentioned they'd just secured 12 months of runway, and the woman said, "oh my God, you poor thing... what on earth are you going to do?", not realizing that 12 months of runway was "more runway than we have had for the last 18 months."

The thing nobody warned her about was how much of a founder's life fundraising eats. "The amount of time we were going to have to spend on fundraising... I had no concept." Early on she'd tell the whole company about every investor meeting, until she realized the steady drip of no's was "bruising the morale of the company." Now GoHenry spins up a small, ring-fenced "project team" for each raise so the rest of the business can keep building. One organizational habit she's especially proud of: a recurring internal campaign she named "Single Points of Failure," where every team hunts for any task, login, or client relationship only one person controls, and eliminates it. When they first ran it, she admits, "my name was on it more than anybody else's."

Fourteen years in, the mission got bigger than the product. GoHenry's own data convinced her that teaching kids about money changes their lives; she cites figures suggesting children taught about money are far more likely to start businesses and end up with materially larger pensions. That pushed her from running a subscription business to lobbying government, handing the UK Treasury anonymized data on the economic impact. The payoff: in November 2025, the government announced that financial education will become mandatory in every school in England starting September 2028. Her "screw it, just do it" lesson, told through the launch of a junior investment account: stop over-researching, put a "fake front door" on the website to test what customers actually want, and always anchor the decision to what's right for the customer, even when the other option makes more money.

Second Founder Feature

Wyatt McDonell, co-founder and CEO of Infinimmune, from a scientific hunch to an $838 million Merck deal in under four years (Raising Biotech, "S4, E8: Infinimmune, From a Scientific Hunch to an $838M Merck Partnership, with CEO and co-founder Wyatt McDonell," Sep 3, 2026.)

If GoHenry is a story about patience, Infinimmune is a story about speed, and it's the cleanest "how a raise actually works" walk-through of the week.

In 2022, five scientists who had worked together at 10x Genomics kept noticing something in their data: new single-cell sequencing tools were revealing antibody biology no one had been able to see before. They didn't have a business plan. They rented, in McDonell's words, "a shoebox office in San Ramon" and spent weeks pressure-testing whether there was even a company there. When they decided there was, McDonell pushed to skip the usual small government grant and go straight for a real seed round. His co-founders, he recalls, "just sort of chuckled and said, yeah, pat me on the head like, okay, Wyatt. And I'm like, yeah, let's go raise a seed round. And then it happened."

The seed round mechanics are a masterclass. McDonell's tactic: take today's well-funded startups, trace them back to their seed rounds, and find "who were the first checks in the door." That led him to Josh Elkington at Axial VC, whose scheduled half-hour "walk and talk" turned into a three-and-a-half-hour walk across Berkeley. Axial introduced others, and Playground Global's Ben Kim and Jory Bell ended up leading a $12 million seed round that closed in October 2022, just three months after the company started. His read on what made investors comfortable writing the first, riskiest check: you have to show a real syndicate is forming, because "nobody likes feeling alone... People are thinking about risk-free returns, not return-free risk. And if you're the first money in the door, it can be hard to overcome that fear of return-free risk." Investors were so surprised that five senior people had all quit 10x for this that they'd get "all five of us... on Zoom" just to check the team was real.

Then they beat their own plan. They'd budgeted two years to hit their first two technical milestones; they hit them in three and four months. "We didn't budget for success, which is an interesting problem to have." That let them start courting big pharma early. The key insight on landing a giant like Merck: it's earned over many meetings by simply doing what you say. "They told us these data... will be available in three months' time. Let's check in with them in three months' time and see if that's true. So once you have a couple of those meetings in a row, then it's like, okay, this is a team that... does what they say they're going to do." Merck's real question, in McDonell's telling: "are you capable of building a quality asset? Because if you aren't... why on earth would we give you high-priority targets from our therapeutic pipeline?"

The payoff came in March 2026: a multi-target partnership with Merck worth up to $838 million in milestone payments. Then last month, the $75 million Series A (co-led by Playground and Regeneron Ventures). McDonell's own reflection on the pace: if he'd told his co-founders at the start where they'd be in four years, "rightfully my co-founders would have looked at me and said, like, that's a great vision for somebody else to work on. Have fun, bud."

Also Heard

Nvidia is buying Hugging Face for about $12.9 billion, one of the largest acquisitions in Nvidia's history. Hugging Face is the "GitHub of AI," the main platform where developers share open AI models. The strategic logic is that it becomes a distribution layer for Nvidia as open models and "physical AI" (robotics) spread. The price is striking: Hugging Face had raised only about $400 million in its life (a Series C at a $2 billion valuation in 2022, a Series D at $4.5 billion in 2023), and it reportedly turned down a $500 million investment from Nvidia at a $7 billion valuation late last year, before agreeing to be bought outright for nearly double that. One podcast pegged the price at roughly 86 times sales. CEO Clement Delangue reportedly approached Nvidia over the summer. (TBPN, "Model Mayhem, GPT-6 Astra, Why Nvidia Bought Hugging Face | Diet TBPN," Sep 3, 2026; Tech Brew Ride Home, "Nvidia Buys Hugging Face For A Rabbit?," Sep 3, 2026; Morning Brew Daily, "Nvidia Buys Hugging Face for $12.9B...," Sep 4, 2026.)
Anthropic is loading a $15 billion credit card ahead of its IPO. The company is finalizing a $15 billion revolving credit facility, a pre-approved line it can draw on as needed, expanded from an earlier $10 billion target, and larger than the $5 billion facility SpaceX arranged before its own listing. As one banker on Bloomberg put it, "it's just a giant corporate credit card... It removes the IPO timing risk. So none of the potential investors would have leverage over them because they don't need the capital." Morgan Stanley is leading, with Barclays, Wells Fargo, Goldman Sachs and JP Morgan in the syndicate, and where a bank sits on the loan usually mirrors where it sits on the IPO. The listing is expected to price in the last days of September or early October, targeting "SpaceX or higher," roughly $75–82 billion raised before the greenshoe, possibly more. One note of caution: the revenue figures thrown around for Anthropic vary wildly across podcasts, from a $45 billion run-rate to $65 billion to one banker's claim of "$100 billion of ARR," so treat the specific number with skepticism; what's not in doubt is that it's growing extraordinarily fast. Separately, Anthropic has been stacking up enormous multi-year compute deals to feed that growth, reportedly around $35 billion with Lambda, $45 billion with Nscale, $50 billion with Fluid Stack, and $45 billion with SpaceX. (Bloomberg Tech, "Anthropic Builds Its War Chest Ahead of IPO," Sep 4, 2026; Bloomberg Intelligence, "Anthropic Finalizing $15 Billion Pre-IPO Credit Facility," Sep 4, 2026; Bloomberg Tech, "Anthropic's Compute Bet, Musk on AI, Apple's New CEO," Sep 1, 2026.)
The Stripe and OpenRouter deal closed at $7.5 billion, and here's the full arc. We flagged this acquisition in prior weeks as a rumored ">$7 billion" deal; the podcasts this week laid out the whole three-year sprint. OpenRouter, which routes AI requests across 400-plus models, raised a seed at a $60 million valuation (a16z and Sequoia, late 2024 into 2025), a Series A at $500 million (Menlo Ventures, June 2025), and a Series B at a $1.3 billion valuation in May 2026 (CapitalG, Nvidia's venture arm, Databricks, Snowflake, MongoDB, plus a16z and Menlo), then sold to Stripe for $7.5 billion in August, all within roughly three years. The three co-founders, Alex Atala, Chris Clark and Louis Vichy, each walked away with an estimated $1.5 billion. The company was doubling its usage (token throughput) every 11 weeks. As the hosts noted, the art of the deal was timing: sell with enough proof of "breakout hypersonic growth" but before any hint of a slowdown could trigger "a significant haircut." (Market Maker, "Stripe's $7.5 Billion AI Bet: Why It Just Bought OpenRouter," Aug 31, 2026; AI to ROI, "Stripe and OpenRouter Combination Wants to be #1 in Routing your AI Workloads," Sep 1, 2026.)
Shein's IPO landed with a thud. The Chinese fast-fashion giant went public in Hong Kong and closed its first day roughly flat after dropping as much as 10% early, now worth around $25–26 billion, down from over $100 billion in private markets back in 2022. A blunt reminder that a rich private valuation is a promise, not a price, until the public market votes. (The Rundown, "Amazon Sued Over Ad Prices, Shein's IPO Falls Flat," Sep 1, 2026; The Best One Yet, "'Spritz School'...," Sep 2, 2026.)
A founder's counter to all the bubble talk. Aaron Katz, CEO of the database company ClickHouse, walked through his own fundraising history on 20VC, and it's a useful frame for this week's sky-high numbers. He raised a $2 billion Series B (co-led by Code 2 and Altimeter) when the company had "no revenue... no product... no customers... 15 employees" and just controlled an open-source project. Today ClickHouse is valued at over $15 billion, with revenue that went "zero, 12, 50, 200" and will "finish this year north of 500" million, net revenue retention above 200%, and, his answer to the fear that AI companies are dangerously concentrated, the entire basket of AI customers (Anthropic, OpenAI, Harvey, Sierra and others) is "less than 12% of revenue." His explanation for why these valuations aren't crazy: "you're not getting priced for where you are... You're getting priced for where you're gonna be in 12 to 18 months." (The Twenty Minute VC, "The AI Bubble Is Wrong | ... with Aaron Katz, ClickHouse," Aug 31, 2026.)
Not everything was AI. Barilla, the 150-year-old Italian pasta company with about $5 billion in annual sales, acquired Goodles, a premium mac-and-cheese startup founded during the pandemic, for an undisclosed sum. Goodles had reached about $88 million in annual revenue, was profitable, and had quietly captured all six percentage points of market share that Kraft Heinz lost since 2022. (The Best One Yet, "'Flagrant', LA Clippers financial foul...Goodles' $100M mac-&-cheese," Sep 4, 2026.)
A charming fundraising origin story. The "physical AI" theme kept running, including StarCloud's orbital-data-center round (which we covered last time). The fresh detail worth passing on: investor Caitlin, who backed StarCloud early, met founder Philip not in a pitch meeting but seated next to him at a Founders Forum dinner in the Cotswolds, both of them, by her account, "wildly and inappropriately overdressed" and battling social anxiety. They talked all night; by dessert she asked what he did, he said orbital data centers, and she said she was a space-tech investor: "he was like, oh, my God, I should have given you a much better pitch." She got in at around a $150 million valuation; the company is now worth $2.3 billion. (This Week in Startups, "VC experts on why Physical AI funding is heating up | E2333," Sep 2, 2026.)