Newsletter · · Ashutosh Agarwal

Easy to Raise, Hard to List as Instinct Hits $10 Billion and Oura Retreats - The Raise - Week of October 5, 2026

The Raise for the week of September 28 to October 4, 2026. Private money and public money came to opposite conclusions: Benchmark led Instinct's $1 billion Series C at $10 billion, AMD paid $8.2 billion for World Labs and OpenAI went looking for $30 billion more, while Oura pulled its IPO hours before pricing, SB Energy found no buyers and Anthropic's leaked filing showed a $42 billion loss. Founder story: Alex Mashrabov on taking Higgsfield from $1 million to $1 billion in annualized revenue in 18 months.

The Raise

Week of October 5, 2026: Easy to Raise, Hard to List as Instinct Hits $10 Billion and Oura Retreats

Private investors put $1 billion into Instinct at a $10 billion valuation, AMD paid $8.2 billion for Fei-Fei Li's World Labs, and OpenAI went looking for $30 billion more. Getting onto the stock market was a different story: Oura pulled its IPO hours before pricing, SB Energy couldn't find buyers, and Anthropic's leaked filing showed a $42 billion loss. Plus the founder who took Higgsfield from $1 million to $1 billion in revenue in 18 months, and the 20VC interview that explains how.

This week private money and public money looked at the same companies and came to opposite conclusions.

On the private side, the checks got bigger. On Wednesday's 20VC, Benchmark's Jack Altman confirmed his firm led Instinct's $1 billion Series C at a $10 billion valuation. Instinct's seed round in April was priced at $50 million. The same week, AMD agreed to buy Fei-Fei Li's World Labs for $8.2 billion in stock, about two and a half years after the company was founded. Bloomberg reported that OpenAI is trying to raise at least $30 billion at a $1.4 trillion valuation, because it no longer has a date for going public.

The stock market was less generous. Oura, a profitable smart-ring company that had Morgan Stanley, Goldman Sachs and JPMorgan running its deal, called off its IPO the night before it was due to price. SB Energy, which wanted a $50 billion valuation, couldn't find enough buyers. Anthropic's IPO prospectus (the long disclosure document a company files before listing) leaked, and it showed a $42 billion net loss on $4.6 billion of revenue.

Altman's explanation for why the private side keeps writing checks was the most useful line of the week:

"There are like 10 companies that can do $10 billion acquisitions and want to. And that's just so different. And it's much easier than going public. And it's quicker."

In other words, a big tech company buying you has become a realistic way out, and possibly an easier one than an IPO. That one idea explains a lot of the prices below.

This Week's Rounds

  • Instinct: $1B Series C at a $10B valuation, led by Benchmark. Instinct makes a consumer AI agent, meaning an assistant that does tasks for you across apps and websites instead of just answering questions. Its 23-year-old founder is Noah Shin (as named on the show). Altman said his partners "Peter and Ev led it." He framed the bet as a new category on the scale of chatbots and coding tools: "this is like a really big new paradigm kind of on the level of chat and coding." He also argued that a strong independent company can win even with Meta's Muse and the big labs competing, much as Cursor did in coding. Harry Stebbings said he would put 5% of a fund into the round. His reasoning: Instinct has raised only about $1.5B, the team is excellent, a large tech company would almost certainly buy it if things went badly, and the upside is roughly 50x if it becomes "the WeChat of the West." Stebbings' real worry was about the pattern, not this particular deal: "What worries me is when you have three rounds in three weeks with no material movement in between and no data suggests there's been anything different." Jason Lemkin raised the open question: Instinct only wins if people use it "all effing day long," the way they already use coding agents. (Continuity: we covered Instinct's $250M round at $2.5B on Sept 7 and its round history on Sept 28. This is the confirmed close at the new price.) 20VC, "Instinct Raises $1B at $10B Valuation | AMD Buys Fei-Fei Li's World Labs for $8.2B | Meta Poaches MongoDB's CEO | Bessemer Raises $5.75B | Oura Pulls IPO & Nubank Eyes $8–12B Monzo Takeover" (Oct 1).

  • OpenAI: seeking at least $30B at a $1.4T valuation. This would be a "bridge round," meaning money to keep things running until a later event, here an IPO that no longer has a timetable. Bloomberg's AI editor Seth Fiegerman said the push appears to come from investors who want in, and the money would cover expensive computing and model development "without necessarily going public in the next few months." CFO Sarah Friar said the company already has plenty of capital after raising $122 billion in March. (Update: last week OpenAI was described as exploring a private raise at $1.2T–$1.5T. This week has a size and a price.) Bloomberg Tech, "OpenAI Seeks $30B as Trump Backs AI Safety Audits" (Sept 30).

  • Higgsfield: reportedly raising at around $8B, after crossing $1B in annualized revenue 18 months after reaching $1M. Higgsfield is an AI video tool for brands and creators. The full story is in the founder section below. 20VC, "$1BN ARR in 18 Months; The Untold Story of Higgsfield" (Sept 28).

  • Go.ai: $85M, for software and hardware that lets companies in heavily regulated industries (banks, healthcare, education) build AI models in-house, so their data never ends up training someone else's model. Several things make it an unusual AI startup. It's eight years old: it started out building Alexa-style digital assistants and changed direction when ChatGPT arrived. It's based in Chicago. And it's profitable. Crain's John Pletz said it has about 60 employees and more than doubled its customer count to over 200 in the past year, including the digital bank Axos and Johns Hopkins University. The company is profitable but still needs money because it ships hardware, and hardware ties up cash. Lead investors were not named on the show. Crain's Daily Gist, "An $85 million bet on in-house AI" (Sept 28).

  • Sandstone: $30M Series A, on top of a combined $10M pre-seed and seed led by Sequoia. Sandstone is an AI workspace that corporate legal teams use to manage their work, laid out like the engineering tools Jira or Linear. Co-founder and CEO Nick Fleisher said deals with the largest Fortune 500 companies run $500K to $1M a year, and $50K–$100K for legal teams of 10 to 30 people. The founders quit McKinsey and "basically the same day raised our pre-seed from Sequoia." The Series A lead was not named. A Product Market Fit Show, "He runs 3 dinners a week instead of cold outbound, and closes $500K+ ARR deals. | Nick Fleisher, Co-Founder & CEO of Sandstone" (Sept 28).

  • Gymkhana Fine Foods: $8.5M Series A, with CAVU re-investing (CAVU is the firm behind the soda brand Poppi). Gymkhana sells packaged sauces and foods from London's two-Michelin-star Gymkhana restaurant, made by the chef who spent 10 years in that kitchen. Founder Gulrez Arora said the firm wrote the first check, a $3M seed, when the company was "basically almost pre-revenue." Gymkhana now sells in Whole Foods (since February) and Sprouts. Arora said the brand grew its UK grocery category 900% and holds 90% of it. The Startup CPG Podcast, "#268 The Rise of Gymkhana Fine Foods with Gulrez Arora" (Sept 29).

  • Atomic: about $12.5M–$15M Series A. Atomic was founded by former Tesla supply-chain staff who worked through the Model 3 "production hell." It uses AI to help companies hold as little inventory as possible without running out of stock, and its customers include DoorDash and HelloFresh. One Equity host was surprised the round was so small: "a few months ago, an AI company that was like working with these kinds of big customers would be raising like 10x this. Am I crazy?" The explanation offered was that the team is small and doesn't own heavy assets. Equity, "Call it AI, call it Super Intelligence, only 2% of consumers are buying it" (Oct 2).

  • Charter Space: $5M, to build a marketplace for insuring satellites and other objects in space. Most satellites are uninsured, which surprises most people. Founder Yuki Chan started out building space-engineering software. The company was a finalist at last year's TechCrunch Startup Battlefield competition. Equity, same episode (Oct 2).

Exits and Deals

  • AMD buys World Labs for $8.2B in stock. World Labs is Fei-Fei Li's company that builds "world models," AI that understands 3D space. Stebbings called it the "first big Neolab exit," using his term for the new wave of independent AI research labs. Lemkin's math: AMD's stock is up 279% this year to about a $1 trillion market value, so the deal costs about 8% of AMD's market cap. "If AMD was up 3% this year, they ain't going to be spending $8 billion." Rory O'Driscoll expects "a bunch of these big-ass acquisitions over the next 6, 12 months if the market continues to hold." Stebbings was more cautious. He cited a report by his French partner Paul that counts 102 Neolabs that have raised $70B+: "just how many of them can get acquired?" 20VC (Oct 1).

  • Nubank in talks to buy Monzo for $8B–$12B. The 20VC hosts thought it was more surprising that Nubank wants to buy than that Monzo wants to sell. Stebbings called Monzo "way too small to be significant in a US public market" and said the European market offers it little. Nubank's own shares fell on the news. Lemkin read it as Nubank paying to save time: "They're not buying just the revenue... They're saving themselves time." 20VC (Oct 1).

Founder Story of the Week

Alex Mashrabov, Higgsfield 20VC, "$1BN ARR in 18 Months; The Untold Story of Higgsfield | Spending $4M Per Month on Models | Why Moats in AI are BS | Scaling a Content Team to 150 People with Alex Mashrabov" (Sept 28)

Harry Stebbings opened the episode by calling this "the story that no one has told in startups yet." That's a fair description. Higgsfield is an AI video company with about 300 of its roughly 400 employees in Kazakhstan. On the day of recording it crossed $1 billion in annualized revenue, 18 months after reaching $1 million. Mashrabov says Cursor took 24 months to do the same thing.

Where he came from. Mashrabov's father is from Uzbekistan, where "if a family of five people makes thousand dollars a month, it's considered to be wealthy." Both parents were mechanical-engineering professors, and they told him from age eight that he "must get to the United States because this is the place where technology matters." His mother worked three jobs to pay for programming competitions and training camps, and by 19 he was top three in the world in competitive programming. His first company, AI Factory, co-founded with Mahi (who is also his Higgsfield co-founder), sold to Snap for $166 million. The fundraising environment back then was nothing like today's. AI companies were valued at "closer to zero" times revenue, "there was like severe dilution," and a $1–2 million round counted as a good one. He made a little over $1 million from the sale and spent it buying apartments for his parents and extended family. He still drives a Tesla Model 3 and owns no property.

The near-death moment. Higgsfield did not find its market right away. "We spent more than a year in a search of a product which could work. We burned more than 10 million out of 16 million raised in seed fundraising. So we felt we have just one attempt left." He takes the blame himself:

"I was so much optimizing for what's hype today, what's the right narrative, how we can hijack the attention, all these things, really. Everything instead of building a good product."

With slightly less than $5 million left, the team talked to eight creative directors. All of them pointed to the same gap: AI video tools had no camera control, no way to direct how the shot moves, which is essential for telling a story. Higgsfield launched a product built around that on March 31, 2025, and it worked straight away. Revenue went from about $1M to $20M in annualized terms in roughly three months, and then from $20M to $100M after the team released an image model that handles product photo shoots well.

How they count revenue. Stebbings asked directly, because "annualized revenue" figures are often inflated. Mashrabov's answer: take the last four weeks of revenue and multiply by 13. Annual contracts are spread across the 12 months rather than counted all at once, and multi-year enterprise deals are not added in. "It's only live revenue."

The numbers behind the business:

  • Customers: businesses account for slightly over 50% of revenue. Pure consumer use, mostly on phones, is under 10%. The West accounts for well over 70%.
  • Growth from existing customers: net revenue retention after 12 months (how much the same group of customers spends a year later, compared with when they started) is over 300% in the business segment. One customer started six months ago on a $99-a-month subscription and just signed a deal worth over $6 million a year.
  • Churn: roughly 30% of new users drop off in the first month, then the curve goes flat. "Expansion is unprecedented" even though early drop-off is high.
  • Marketing: "We don't do paid." Growth comes from an in-house team of more than 150 creative professionals, nearly half the company, who make launch videos, tutorials and even an open-source AI feature film (over 100 hours of generated footage to get 90 minutes of finished content). He admitted that handing creator marketing to an outside agency caused the influencer controversy: "it's very important to own distribution."
  • Margins: above 80% when Higgsfield runs its own models or tuned open-source models, versus 20–30% on closed models from the big labs. Higgsfield picks which model to use in more than 40% of cases. Mashrabov calls this "tokenomics."
  • Spending on AI internally: over $4 million a month, or more than $10,000 per employee. "This month, I just caught a guy who spent over $30,000 in a week" building an internal tool on a frontier model.

The fundraising mechanics. Asked about his best VC meeting, Mashrabov answered without hesitating: "Obviously, Yuri Milner gets it." He also described the worst thing that happened to him in fundraising:

"People really shook hands, said we do at this price. And next day, what I learned is that they called other investors and they pulled the syndicates to invest in 30% lower valuation compared to what we discussed."

Stebbings argued that Higgsfield is undervalued because it is an outsider: "If you were a Silicon Valley company, that would easily be a $25 billion company," compared with the rumored ~$8B price. Mashrabov didn't argue about the multiple: "we're not chasing just the valuation... the goal is just to make sure that the company can be sustainable over the time in public markets." He wants to take it public eventually, and says it could become "bigger than Applovin and Shopify."

What comes next. His finance team projects $4.5 billion in revenue by the end of next year, which already assumes growth slows sharply. Asked for his own number: "Over 10." He spends 80–90 hours a week on the company, at least three hours a week with his wife and five with his son. On why he keeps pushing, he pointed to his time at Snap, which he says was once worth $80 billion and is now below $15 billion: "momentum doesn't last forever... while we do have the positive momentum, we do not take this for granted."

Why it matters for founders. Three lessons stand out. First, the company was nearly killed by chasing a story rather than talking to customers, and eight conversations with creative directors rescued it. Second, Higgsfield's margins depend on choosing which model does the work, not on building the best model. Third, a handshake on price is not a deal, even with sophisticated investors.

Also Heard

  • Anthropic's IPO filing leaked, and the numbers drew scrutiny. According to Reuters reporting read out on Tech Brew Ride Home:

    • Revenue grew 12x to about $4.6B in 2025, with an $8B+ operating loss.
    • The $42B net loss includes a roughly $34B accounting charge tied to past fundraising, which is not cash spent running the business.
    • Spending on computing and infrastructure was $7.33B, more than half of $12.65B in operating expenses, and the company has $518B in future cloud and infrastructure commitments.
    • It had $20.28B in cash. Two customers made up nearly a quarter of revenue.
    • The seven co-founders together control a special share class that carries 50.1% of the voting power on key matters.
    • The target valuation is above $2 trillion, more than double its own ~$965B estimate in May. Listing is likely after the November midterms.

    The sharpest skeptic was Ed Zitron on Monetary Matters: "Anthropic, at least in 2025, was a worse business than OpenAI. It spent $2.75 to make a dollar," compared with $2.60 at OpenAI by his own reporting. He added that about 47% of revenue came through Google and Amazon reselling Anthropic's models, which makes two of its biggest competitors its biggest sales channels. On 20VC, Jack Altman was more hopeful: "public market investors will be a little bit more long term oriented than the private investors." (Recurring story. Only the filing details are new this week.) Tech Brew Ride Home, "We Have The Anthropic IPO Numbers" (Sept 29); Monetary Matters, "Ed Zitron on Anthropic's IPO (S-1), AI Debt, and Counterparty Risk" (Oct 1); 20VC (Oct 1).

  • Oura pulled its IPO the night before pricing. The Rundown put the planned raise at about $2B at a $13.5B valuation and cited Barron's view that roughly three-quarters of the shares on offer came from existing investors cashing out. On 20VC, Rory O'Driscoll, whose firm owns a stake, said he was "super surprised." He pointed out that early backer Forerunner had said upfront it would sell its entire position in the IPO, which is rare. When sellers are cashing out at the IPO, a lower price comes straight out of their returns, so they hold out for more: "If you think you're going to get 22 bucks a share and suddenly you're getting 18, that's going to reduce your entire venture return by 20%." Lemkin noted that employees had already sold $534M in a tender offer (a company-organized share sale) a couple of months earlier, which takes some of the sting out. (Update: we covered Oura's IPO terms on Sept 28.) 20VC (Oct 1); The Rundown, "Oura Pulls it's IPO Last Minute, FICO's Monopoly Comes Under Fire" (Sept 29).

  • SB Energy shelved its $50B IPO after its bankers couldn't find enough buyers. Scott Galloway on Prof G Markets was blunt: "they're trying to get a $50 billion valuation on a PowerPoint with a SoftBank logo." By his count, only about 9% of contracted capacity is under construction and the price works out to about 234 times sales for a company whose revenue is shrinking. Holtec, a nuclear company, also postponed its IPO, with its CEO saying the business is "viewed as connected to data centers." Prof G Markets, "Investors Are Turning Against Data Centers (Here's Why)" (Sept 28).

  • Jev (from TypeSafe) is still in the market at $10B, one week after its ~$200M seed, and was the "investment committee" debate on 20VC. Lemkin, half joking, argued for putting 20–30% of a fund in. He cited Jev at 17% of traffic on OpenRouter (a service that routes developers' requests across many AI models) and 20% through Vercel's router, at "the 70th of the price and 100 times faster." (Update to last week's lead story.) 20VC (Oct 1).

  • What VCs are saying about the market. O'Driscoll relayed economist Tyler Cowen's verdict to his firm's investors: "returns are going to be highly skewed. Variance is going to go up with AI and many of you will fail." Stebbings said his team told him "we can't find anything under 100 million," meaning round size, not valuation. O'Driscoll pushed back that there are two kinds of companies. AI labs need hundreds of millions just to ship. App companies built on top of those labs "took $10 million, but we shipped the product for $3 million," and then raise $50 million anyway "because we can." 20VC (Oct 1).

  • Tines (Eoin Hinchy): over $270M raised, no pitch deck ever. Tines is a security-automation company founded in Dublin in 2018. It is approaching $100M in annual recurring revenue with about 450 employees, and has raised from Goldman Sachs, SoftBank and Accel (transcribed as "Excel"). "We've never done a pitch deck, we've never proactively raised capital." For the seed round, the UK firm Blossom Capital emailed cold, was told no, and then "showed up at our front door of our office." His approach to choosing investors changed with the company's stage. Top-tier VCs early, "to lend credibility... win more customers." Growth funds like SoftBank and Goldman later, because the next phase "was all about scale and... selling to these Fortune 500 CIOs." Not Another CEO Podcast, "He Blew Up His Own Winning Product Before AI Could - Eoin Hinchy - Tines - EP#115" (Sept 29).

  • Terra Kaffe: $2.5M spent before a single espresso machine shipped. The founder bootstrapped for two years, then raised money only in milestone-based six-figure checks: "we never got one clear check." In early 2023 the team held a meeting it called the "gun to the head" meeting, because the machine (432 parts) had to ship within six months with no way to extend its cash. To make the deadline, they cut features, including a clicking milk-froth dial. He also confronted an outside engineering firm that had "spent double the money and gone half the distance." The pre-sale that followed sold about 10,000 machines with no reviews. The DTC Podcast, "How Terra Kaffe Pre-Sold 10,000 Espresso Machines With No Reviews" (Sept 28).

  • A French sales-compensation software company that turned down the 2022 valuation boom. Founder Antoine raised a $100K angel round, a $5M seed (January 2022) and a $10M Series A (end of 2023) from Singular, the Revenue Syndicate and Breega (transcribed as "Briga"). He then added $2M of debt from HSBC and two other banks: "don't take this money as if it was investor money because it isn't." He now has 250+ customers and just over $7M in annual recurring revenue, up from $4–5M a year earlier. On the valuation game: "it was quite obvious that... this could turn into a trap." Top Founders, "You Make $8m/yr, Would You Sell for $100m Cash?" (Sept 29).

  • Clean Energy Associates: bootstrapped, nearly unsellable, then sold for $112M. Andy Klump built a technical-advisory firm for the solar industry starting in 2006. The first sale attempt fell apart in late 2021 when the buyer, Intertek, held back over unaudited financials. The deal finally closed in July 2022 for $112M with about 200 employees, followed by an earn-out (part of the price paid later, depending on performance) that ran until early 2025. Built to Sell Radio, "Ep 566 From 'Unsellable' to $112M, a Bootstrapped Sale to Intertek" (Oct 2).