Newsletter · 2026-09-21 · Ashutosh Agarwal

The New Yardstick for AI Startups Is a $25 Billion Exit - The Raise - Week of September 21, 2026

The Raise for the week of September 14 to September 21, 2026. Founder and VC podcast synthesis on venture investors pricing AI startups against imagined $25 billion exits, from Instinct's $10 billion ask and Discovery Loop's reported $50 billion re-raise to Positron and Crusoe, Miro's sale at a 92% discount, Anthropic and OpenAI stepping back from IPOs, and the founder story of The INKEY List's Colette Laxton.

The Raise

Week of September 21, 2026: The New Yardstick for AI Startups Is a $25 Billion Exit

This week the venture world stopped pricing AI startups against their revenue and started pricing them against an imagined $25-billion-plus exit. That math sent one texting app to a $10B raise, poured billions into chips and power, quietly buried a 2021 SaaS darling at a 92% discount, and pushed the two biggest names in AI to hit pause on going public.

There was a moment on The Twenty Minute VC this week that explains the whole strange season we're in. The hosts were talking through why anyone would pay a fortune for an unproven AI company, and one of them laid out how a top firm now models the world: they assume there will be roughly 100 tech companies worth $25 billion or more in the AI era, and they build their whole strategy around owning a piece of as many of those as they can.

Once that's your worldview, the eye-watering prices start to look almost logical. If the winners are all worth $25B and up, then a $10B entry price is "only" a 2.5x away from a good outcome. As one host put it: "I wrote that the new decacorn is $25 billion. These instinct rounds make sense." ("Decacorn" is Silicon Valley slang for a startup worth $10B or more.)

You can see that logic playing out everywhere this week: in the rounds that got done, in the ones that fell apart, and in the SaaS star that finally got sold for scrap.

This Week's Rounds

Positron AI: $875M Series C at a ~$5B valuation. Positron builds custom hardware and software aimed squarely at AI inference, the running of already-trained AI models, as opposed to the training itself, which is Nvidia's ($NVDA) stronghold. Co-founder Thomas Sohmers came on to make the case that inference is where the real long-term compute bill lands, and that a specialized chip can undercut the general-purpose giants. Gavin Baker's Atreides Management is among the backers. (20VC, "'Anti-Data Centres is a Chinese Psyop' … With Thomas Sohmers, Co-Founder @ Positron," Sept 19.)
Crusoe: $3.9B at a $30.9B post-money valuation, co-led by Atreides, Valor Equity Partners and Abu Dhabi's Mubadala. Crusoe builds the physical guts of AI, giant data centers including the Abilene, Texas complex that OpenAI uses, and started life back in 2018 flaring-gas-to-Bitcoin before pivoting hard into AI. The new twist: it's moving toward smaller, factory-built modular data centers it calls "Spark," aimed at inference workloads closer to where they're needed. (We flagged this round as "finalizing" last week; it's now closed at a higher, confirmed number.) (Tech Brew Ride Home, "Catfished By AI!," Sept 17.)
Discovery Loop: $1B at a $10B valuation just weeks ago, and reportedly already raising again at ~$50B. This is the new company from Jeff Dean, Google's former chief scientist, one of the most respected engineers alive. The pitch is "automating discovery": running thousands of scientific and engineering experiments in parallel to speed up research itself. A 5x markup in a matter of weeks tells you how badly investors want exposure to a name like Dean's. (Tech Brew Ride Home, "The Great AI Slowdown?," Sept 14.)
Instinct: raising ~$1B at a ~$10B valuation. Remember Instinct? We covered it two weeks ago at $350M and a $2.5B valuation. It's the consumer AI companion app from Noah Shin (connected to Bret Taylor's orbit; Taylor chairs OpenAI's board), the one racking up thousands of texts per user. In roughly two months it has gone from a ~$500M valuation, to $2.5B, to a $10B ask. The 20VC hosts were openly torn: the consumer pull is real ("I did an Instagram reel on it. I had over a thousand DMs asking for invite codes. I've never had a thousand DMs"), but so is the danger. The app is "slow," which they read as a sign of brutal compute costs, and Meta ($META) just launched a rival personal-AI app called Muse. The bear case: burn through venture cash subsidizing users and end up like poolside, a company that ran out of runway and had to sell. The bull case rests almost entirely on Noah Shin being, in their words, "one of the most generational talents," and someone like OpenAI eventually buying it. As one host warned: "The way I was raised to invest in venture was don't take bets that 100% require an M&A outcome to be successful." (20VC, "Why 'Pacing the Frontier' is BS | Instinct Raising $1BN at $10BN …," Sept 17.)
Fortell: $163M Series B at a $740M valuation, led by Founders Fund, Thrive Capital, Valor Equity Partners and Antonio Gracias. Fortell is building AI-powered hearing aids using custom chips and spatial-audio processing, an unglamorous, giant market that a murderers' row of top-tier investors just decided is ripe for reinvention. (Equity, "Why Thrive, Founders Fund, and Antonio Gracias are betting on hearing aids," Sept 16.)
Tabapay: ~$155M growth round. The fintech quietly grew to $100M in annual recurring revenue on a single $2.5M seed raised back in 2017 (one investor, nine years, no additional dilution) before finally taking a big check to help capitalize a bank it's acquiring and expand beyond pure payment processing. A rare "we barely raised and it worked" story in a week of billion-dollar rounds. (The SaaS Podcast, "Inbound Marketing That Grew a Fintech SaaS to $100M," Sept 17.)
Cents: $140M Series C, led by Sumeru Capital ($110M in fresh primary capital plus a $30M employee tender, letting staff cash out some shares). Cents makes operations and laundry-management software for large apartment (multifamily) property owners, decidedly not an AI company, and doing about $60M a year in revenue. Proof that boring, sticky software can still command a nine-figure growth round. (Top Founders, "How Does Laundry Software Make $60M a Year?," Sept 17.)

Founder Story of the Week

Colette Laxton, The INKEY List (Ladies Who Launch, "The Story Behind Global Skincare Brand: The INKEY List," Sept 16)

If you want to see how a tiny, thesis-driven bet turns into a brand that has now sold 90 million products worldwide, listen to Colette Laxton walk through the founding of The INKEY List.

She spent a decade at Boots head office, starting on £13,000 a year, then a year building a nail brand while, in her words, "living on her sofa." The idea for INKEY hit her on a 2018 flight to Canada. She'd noticed two things at once: beauty brands were, as she puts it, "absolutely bullshitting consumers," slapping pretty marketing over products with zero ingredient transparency; and meanwhile Reddit was "bubbling" with ordinary people asking each other what was actually in their skincare. Her thesis was simply to close that gap and be the honest, plain-speaking expert friend:

"Imagine … being a brand that truly could support somebody. They didn't have to turn to each other. They could turn to us and us be that expert friend."

She launched with a £300,000 budget, split roughly £150,000 into stock (15 products), £50,000 into marketing, and £100,000 into operations and staff. Most of that marketing money went to one bet: the skincare influencer Caroline Hirons. She hired a PR agency on pure contingency (no monthly retainer) and tasked it with landing a Daily Mail headline and key editor relationships. Day one: 19 press pieces, including Vogue. The brand launched through the UK retailer Feel Unique, and within three months Cult Beauty (now Sephora UK) called wanting to buy her entire warehouse.

Then came the moment that forced a real fundraise. Six weeks later, in a Paris meeting, Sephora North America's Artemis Patrick looked at her and asked, point blank, about a US rollout, a plan Colette did not have, for a country she had never even visited:

"She was like, no, seriously. Are you ready for this? … Do you have a field team to help in store? Do you have 500 reviews per product to launch? Do you have a full marketing plan?"

Colette's answer, essentially, was to jump and build the parachute on the way down: "I'm going to make it happen." She had six months to stand up a US field team, hit 500 reviews across 15 products, set up an American warehouse and secure manufacturing, all from the UK, with no US network.

The money to do it came from an inbound call. A placement in The Guardian ("The New Frontiers of Beauty") caught the eye of Unilever Ventures, the venture arm of the consumer-goods giant, which committed "over a million" pounds to fund the warehouse, manufacturing, review-seeding and that all-important field team. (No valuation or formal round name was disclosed; she kept the details close.)

The near-death moments were operational, not financial. The agency she hired for the US field team sent a rep into Sephora in "a dirty suit too big for him," having come straight off a nightclub shift; she fired the agency on the spot and rebuilt the team by poaching independent contractors from competitors. Her US stock, planned for six months, sold out in four, forcing painful air-freight-versus-sea-freight margin calls. COVID then poured fuel on demand. And years later, 90-hour weeks caught up with her: she burned out completely, collapsed, and spent two weeks bedridden, a forced reset on what "success" was even supposed to feel like.

The most recent chapter is the one no plan predicted: an unplanned explosion on TikTok Shop brought 33,000 affiliates in three months and made one INKEY body product the number-one body product on the platform, despite the brand never intending to be a body-care line. Her hard-won lesson, repeated throughout: "No one ever knows what's going to sell." Today INKEY moves roughly 1,000 units a day through Boots alone, has 7 million users on its ingredient "recipe builder," and employs about 100 people.

Also Heard

The 2021 dream finally got buried. Miro, the online-whiteboard company that raised at a $17.5B valuation in 2021, sold to Italy's serial-acquirer Bending Spoons for about $1.35B, roughly a 92% haircut. The 20VC panel called it "inevitable": Miro was, by their tracking, "the largest, utterly stale valuation from that period," a productivity tool priced miles ahead of its actual market while newer AI names blew past it. Early investors and founders made money; the late-stage backers who paid the 2021 price got close to their money back and no more. As one host put it, quoting a Sequoia partner's meme: "Death comes for us all in SaaSland." (20VC, "…Miro Sells for $1.35BN After a $17.5BN Valuation…," Sept 17.)
The giants hit pause. In a striking reversal, both AI leaders backed away from the public markets this week. Anthropic delayed its planned IPO, the one floated at a $2 trillion valuation, explicitly citing AI-safety concerns, even as it reportedly hit around $100B in annualized revenue and lined up Nvidia ($NVDA) as an anchor investor for up to $10B (a deal one podcast framed less as validation and more as Nvidia locking in its biggest chip customer). OpenAI, meanwhile, ruled out a 2026 IPO and is instead reportedly exploring a private raise at a $1.2 trillion to $1.5 trillion valuation, with Sam Altman arguing the company needs the "structural freedom" to prioritize safety over shareholders. Underneath the trillion-dollar headlines, Big Technology flagged a wobble worth watching: AI token prices have fallen 41% from their March peak, and frontier-model usage slipped from 53% to 45% of the total, a hint the growth story isn't as vertical as the valuations assume. (Recurring storyline; new this week is the delay and the private-raise pricing. Elon Musk Podcast, "Anthropic delays $2 trillion IPO for safety," Sept 20; Big Technology Podcast, "AI Doom Backlash Arrives, Anthropic & OpenAI IPO Outlook…," Sept 19; TBPN, "…OpenAI Explores $1.2T Valuation," Sept 17.)
A tale of two exits, and a case for bootstrapping. Serial founder Ankur Nagpal broke down the cold mechanics of selling a company. He sold Teachable for $250M in 2020 (right before COVID doubled its revenue from $25M to $50M in three months), then sold his fintech Carry to AngelList for about $70M, but the real lesson was what happened next. AngelList only wanted the team and the promise, not the underlying platform, so he sold that separately for just $6M, even though it generated $4M a year. His takeaway: "70 million for basically talent and the promise for what could be, and 6 million for the actual asset," the difference between a company being bought versus sold. He also made a pointed argument that venture money is overrated status: "I think bootstrapping should be higher status. I think VC means you kind of fail to run a business and grow with profits." Bonus for founders: he detailed the QSBS tax break: sell C-corp shares held 5+ years and you can pay zero tax on $10–15M in gains, per shareholder, multipliable across family members. (The Nathan Barry Show, "$300M+ Founder: What Nobody Tells You Before Selling Your Business | Ankur Nagpal," Sept 17.)
The bootstrapped popsicle. JonnyPops, the better-for-you ice-pop brand, reportedly landed around $175M in new backing. The origin story is the fun part: founded roughly a decade ago by college friends Erik Brust and Connor Wright, who met in a physics class, each chipped in $2,000, and made their first popsicles in a free church kitchen before breaking out at the Minnesota State Fair. Their growth hacks (eight-color rainbow "firecracker" pops built for Instagram, and kind messages printed on the sticks that turned into a collectible kids trade Pokémon-style) helped them win freezer-aisle space by bringing brand-new buyers into the category. (The investor and valuation behind the $175M figure were not named.) (The Best One Yet, "'Get Rich Quiet' … JonnyPops' $175M popsicle," Sept 15.)