Newsletter · 2026-09-28 · Ashutosh Agarwal

One AI Startup Goes From $200 Million to $10 Billion in a Week - The Raise - Week of September 28, 2026

The Raise for the week of September 21 to 28, 2026. Founder and VC podcast synthesis on investors paying almost anything for AI winners, from TypeSafe's jump from a $200 million valuation to $10 billion-plus talks in a week to Factory, OpenEvidence, Legora and Nscale's $35 billion IPO filing, with a founder story on Bending Spoons CEO Luca Ferrari turning a $40,000 leftover check into a company he values near $40 billion.

The Raise

Week of September 28, 2026: One AI Startup Goes From $200 Million to $10 Billion in a Week

A startup valued at $200 million last week is now being offered money at $10 billion or more. Around it, coding agents, legal AI and medical AI all got pricier, a bootstrapped French company sold its first shares at $100 million, a two-year-old GPU landlord filed to go public at $35 billion, and an Italian founder explained how a $40,000 leftover check became a $40 billion company.

Last week a startup called TypeSafe came out of stealth, meaning it went public about its existence for the first time. It said it had raised a $40 million seed round, which PitchBook put at a $200 million valuation.

This week, according to reporting read out on Tech Brew Ride Home, TypeSafe is talking to investors about raising more than $1 billion at a valuation above $10 billion. That is a 50x jump in about seven days, for a company whose product most people haven't touched.

The host couldn't quite believe it either:

"Again, let me repeat that. Last week, it was valued at $200 million. This week, possibly more than $10 billion. Look, folks, if this were to come to pass, this would without a doubt be the fastest fundraising glow-up of all time."

TypeSafe was the extreme case, but the rest of the week pointed the same way. Investors are paying almost anything to get into the few AI companies they think will win. On This Week in Startups, Jason Calacanis described the playbook behind these back-to-back rounds: find the famous investor you want, then use their name to push the price up again, sometimes within a few weeks. His guest's warning to founders and fans was the most useful line of the week:

"I'd say the audience shouldn't confuse a fast funding round with a real business or a signal around that. Just because there's a markup, it doesn't really tell you much about the business."

Keep that in mind as you read the list below.

This Week's Rounds

TypeSafe: in talks to raise $1B+ at a $10B+ valuation, one week after a $40M seed at ~$200M led by DCVC. TypeSafe is building a new kind of AI model called Jev that doesn't use the same design as today's large language models (the technology behind ChatGPT and Claude). The co-founders are former OpenAI researcher Diogo Amieda, former Meta research engineer Sasha Sheng and entrepreneur Eric Gaffney. At The Information's AI Agenda conference, Nvidia's Dion Harris called the model "incredibly fast," while also pointing out its limits. The report adds a telling detail: all the buzz may be driving up TypeSafe's computing bills, which makes a giant round more appealing. That is the same pattern that played out with Instinct. Not everyone is sold. The Elon Musk Podcast spent a whole episode on Jev and told listeners to "look past the founder praise and the viral launch numbers to examine the actual math of what the system is doing." Tech Brew Ride Home, "Is Copilot Microsoft's Everything App?" (Sept 25); Elon Musk Podcast, "Why Jev Refuses to Write Sentences" (Sept 27).
Core Automation: $630M raised since January, most recently at a $3.5B valuation (a figure that includes the new money). Core Automation was founded earlier this year by former OpenAI senior researcher Jerry Tworek. It is building AI models that keep learning after they've been trained, while today's models are frozen once training ends. Tworek's reason for leaving a big lab is a sign of the times: he decided to start his own company partly because he thought he could get more computing power for his experimental research on his own than inside a larger company. Tech Brew Ride Home, "Is Copilot Microsoft's Everything App?" (Sept 25).
OpenEvidence: $250M at a $15B valuation, up from $12B in January. OpenEvidence is an AI search engine for doctors. According to CNBC figures cited on the show, more than two-thirds of U.S. doctors now use it for diagnosis and treatment advice. The Miami company has raised over $1 billion in the past year from Thrive Capital, DST, GV, Kleiner Perkins and Sequoia, and Nvidia is also an investor. Business Insider's report adds that it could be open to selling itself. OpenAI and Anthropic are both building their own healthcare tools, so a company that doctors already use and trust would be a fast way in for either of them. Tech Brew Ride Home, "Is Copilot Microsoft's Everything App?" (Sept 25).
Factory: $200M at a $5B valuation, triple its last price. Factory sells AI coding agents (it calls them "Droids") to large companies. On 20VC, Jason Lemkin argued it was worth the price, based on what he heard from executives at Salesforce's Dreamforce conference. Big companies, he said, don't trust the AI labs with their most sensitive data, and they want a coding tool from someone who isn't also selling them the model: "I think people in the next 12 months are going to be like, whether it's my data for my drug or just my code, I don't want my core code polluted in Anthropic and OpenAI where they're going to train on it." Rory O'Driscoll agreed and summed up the panel's view: "Coding is the mother lode. It's that simple." (Host Harry Stebbings is a Factory investor, and the panel said so on air.) 20VC, "Meta's Muse Hits No. 1 … Factory Triples Its Valuation to $5 Billion …" (Sept 24).
Legora: next round reportedly at an $11B valuation, after announcing $200 million in annual recurring revenue (ARR, the yearly value of its subscriptions). Legora makes AI software for lawyers and competes head-on with Harvey. The same 20VC panel was more careful here. Lemkin pointed to a report from The Information that Harvey's gross margins (revenue minus the direct cost of delivering the product, mostly AI computing costs) had fallen to minus 50%. He said that if margins were "spiraling down rather than V-shape," he would be "a hint nervous" and wouldn't lead the next round, even though he likes both companies. 20VC, "Meta's Muse Hits No. 1 …" (Sept 24).
Taktile: $120M round led by Goldman Sachs. Taktile sells AI software that banks and insurers use for credit decisions, fraud and transaction monitoring, and more and more for insurance claims. The CEO says one of the world's biggest insurers now runs a large share of its claims on Taktile. The more useful part for founders was how he runs a raise (see "Also Heard"). Fintech Leaders, "Taktile's CEO Raised $190M Betting Most Bank Jobs Won't Exist in 10 Years" (Sept 22).
Numeral: new round of about $100M. Numeral automates sales tax and VAT filing, the "low-level grunt work that gets done by armies of tax workers throughout the globe," in the words of co-founder and CEO Sam Ross. It already files in more than 80 countries, and he said the new money will go mainly into R&D (building product). TBPN, "Jensen vs NYT, New Model Reactions …" (Sept 23).
Emma (EMA): $77M. This India-based startup sells teams of AI agents that handle work across HR, IT and finance together, rather than one task at a time. Google and Microsoft are early customers, which stood out to the Equity hosts because both companies make their own AI. They also noted that more AI startups are coming out of India with backing from Indian funds. Equity, "Meta's Muse just stole the spotlight from OpenAI and Anthropic" (Sept 25).
Fleet.co: first outside money after seven bootstrapped years, at a $100M valuation. This Paris company rents laptops and phones to businesses (about €50 a month instead of buying a €2,000 computer) and manages them. It grew to $40 million in revenue across 20 countries with close to $10 million a year in EBITDA (a rough measure of operating profit) and never needed to raise. The deal was a leveraged buyout, or LBO: investors bought existing shares from the founders, partly with borrowed money, and no new money went into the company. Co-founder Seyvan Marjan said the total cash-out was almost $40M, a bit under $30M of equity plus roughly $15M of debt, with investors ending up with about 25% of the company. They set up a new 5% employee stock pool and are aiming for a $300–500M valuation within five years. More below. Top Founders, "How He Makes $40M/yr Renting Laptops for $50/Mo" (Sept 24).
Instinct: fuller details on the ~$1B-at-$10B raise. (We covered this round last week; the new part is the round-by-round history.) According to Harmonic data shown on This Week in Startups, the iMessage-based AI assistant raised a $25M seed at a $50M post-money valuation in April 2026 ("post-money" means the value including the new cash). That was followed by $75M at $500M and then $250M at $2.5B with Benchmark and Index Ventures, both in August. Sequoia and Benchmark are reportedly circling the new round, with about $350M of the $1B already committed. Calacanis said it gave him "a Clubhouse vibe." On 20VC, Rory O'Driscoll put the risk in terms of how much room for error you're buying: "your margin of safety at 50 pre is infinite … Your margin of safety at 10 billion, yeah, maybe you get a 1x, but there's risk." This Week in Startups, "VCs Would Bet on Open-Source AI Over OpenAI and Anthropic | E2341" (Sept 23); 20VC, "Meta's Muse Hits No. 1 …" (Sept 24).

Founder Story of the Week

Luca Ferrari, Bending Spoons All-In, "Luca Ferrari, Bending Spoons CEO: The $40K Origin Story, Buying Product-Market Fit & Why Private Equity Can't Compete" (Sept 23)

Most founder stories start with an idea. This one starts with a failure and a small check the investor didn't bother to take back.

In 2010, Luca Ferrari and his co-founders started an AI company. "Very early, too early, clearly," as he put it. Three years later it had "crashed and burned," and the only thing left was about $40,000 of the money they'd raised from a venture capitalist. The investor didn't want to pay lawyers to shut the company down properly, so they made an unusual offer:

"They told us, you guys keep it. We'll sell our shares to you for $1, like nominal value, and you go and get a nice vacation. We're clearly a little bit sick in the head, and so we took the money and enthusiastically turned it into seed financing for Bending Spoons."

The idea: buy product-market fit instead of searching for it. ("Product-market fit" is startup shorthand for having a product that people clearly want.) The failure had taught the team something honest about themselves:

"We are not very good at finding product market fit, or maybe luck plays a big role. Probably both things are true. But we have become pretty good at engineering, design, monetization, marketing … And so we should be able to buy product market fit from people. And they get a good price. We get a good asset. We can make more valuable."

Their first purchase cost about $10,000. It was an iPhone app for customizing your keyboard, built by one developer, with almost no revenue. What they were really buying was users and a good spot in the App Store. They rewrote it from scratch, made it better, and did the same thing again, then again, at larger and larger sizes.

How it's paid for. This is where the story gets unusual by Silicon Valley standards. Bending Spoons has put "pretty much 100%" of its free cash flow (the cash left over after running the business) into buying more companies, from the start. It began borrowing in 2017 or 2018 with basic bank loans and later moved to larger institutional loans. It raised very little from outside investors. When it went public, Ferrari said, it had raised only about half a billion dollars in new equity at a valuation of roughly $20 billion, and nearly all of that came in the six months before the IPO. Today he puts the market value "roughly" around $40 billion ("I haven't checked the ticker since we IPO'd"), and says that with Miro included, the company runs at close to $4 billion in annual revenue. The debt numbers are tidy: a blended interest rate of about 9%, fully hedged so rising rates won't raise it, maturing in 2031, with debt at about 2.5 times a year's operating profit. He says past returns on its deals have been "consistently above 25%" before counting the borrowed money.

The part that makes people uncomfortable: very small teams. Chamath asked the question everyone asks about Bending Spoons: how do you cut most of an acquired company's staff and still have the product work? Ferrari said they found out by accident. Early on, the people who sold them apps usually kept their teams, so Bending Spoons built small internal teams to run the products. Later, when they bought companies that came with large existing teams, they "couldn't explain why you necessarily needed more people." His rule now:

"We find that generally you're more likely to get that level of performance if you have very, very small teams, super high bar for talent and sense of ownership."

The central team is about 800 people, and about three-quarters of them are engineers, researchers or product people. Every business they buy gets moved onto a shared "operating system" of more than 50 in-house tools covering AI model management, recruiting, A/B testing and more. Staff move between businesses: a year rebuilding AOL's email systems, then seven months on Vimeo's subscriptions. That variety is also how they hire. Last year they got 800,000 job applications and hired fewer than 300 people.

Why he thinks private equity can't copy it. Private equity firms buy companies, run them separately and then sell them. Ferrari's point is that this makes shared technology and shared teams impossible for them: "What do you do when you sell it to your private equity competitor? Do you license it to them?" And the thing that took Bending Spoons a decade to build can't be rushed: "You could probably do it in five years … but not in two months."

Chamath's verdict was that Bending Spoons is the first large-scale success in tech of the "buy and improve" playbook made famous by Danaher, Roper, Amphenol and Berkshire Hathaway. The lesson for founders is simpler: a failed company, a forgiving investor and $40,000 can be enough, if you're honest about what you're good at.

Also Heard

A two-year-old company files to go public at $35 billion. Nscale, a UK "neocloud" (a company that rents out Nvidia chips by the hour for AI work, like CoreWeave), filed to list on the New York Stock Exchange under the ticker NSCL. The Financial Times reports it is aiming for a valuation of about $35 billion, up from roughly $14.6 billion in its private round in March. Run the Numbers went through the filing line by line:
- Headline number: $103.4 billion of "take-or-pay" contracts (customers commit to pay whether or not they use the capacity). Microsoft committed up to $43.8B through 2033 and Anthropic up to $44.6B. Together that's about 85% of the total, and only $2.6B (about 2%) is live today.
- The Anthropic deal was signed August 25 and lets Anthropic walk away from parts of it if capacity is delivered late. From the filing: "we have not obtained binding commitments for any of the financings required to fund performance under the Anthropic Services Agreement."
- The finances: first-half 2026 revenue of $141M (up 1,252%). Gross margin was about minus 35% before depreciation and minus 159% after it, meaning it currently costs more to deliver the service than customers pay. The company lost just over $1 billion in the half.
- Red flags: "going concern" language (auditors' wording for real doubt about whether a company can keep operating without new money), weak accounting controls, and a largest customer that made up 52% of first-half revenue.
- The people: CEO Josh Payne, 32, got an IPO stock award worth about $350M, and almost half of it depends on how many megawatts of capacity he actually builds. The board includes Sheryl Sandberg and Nick Clegg. (Run the Numbers, "The NScale IPO | A Neocloud S1 Breakdown" (Sept 24).)
How to run a raise, from a founder who has done it well. Taktile's CEO (whose wife used to be a venture capitalist) described fundraising as mostly a matter of managing investors' fear of missing out: "There's one kid in the sandbox and then once there's one toy and everyone wants to have the toy." In his view only about 50 investors in the world really matter for a round like his, "they all talk and they all know each other," and for junior VCs "the main currency … is information." His approach is to run a tight process with a couple of "stalking horses" (early bidders who set a floor price): "Let someone run fast and then squeeze it." He also has a warning about later rounds: "now VCs become more bankers than believers." He was open about the company's hardest moment, too. Taktile's first product, a tool for managing machine-learning models, wasn't working, and he had to persuade the team and investors to scrap it and rebuild when "the market has just turned from 2020 to 2023." Fintech Leaders (Sept 22).
Why Fleet.co waited seven years to sell a share. Seyvan Marjan and his co-founder met at the startup studio Rocket Internet, where they helped launch Jumia (sometimes called "the Amazon of Africa," now listed in New York). The model that let Fleet.co skip venture capital is clever. It signs 24- or 36-month rental contracts, then sells each contract to a bank, which pays the full value up front minus about 10%. The bank carries the risk of a customer not paying, and suppliers ship straight to customers, so Fleet.co holds almost no inventory. The company is profitable and brings in cash from day one, so growth pays for itself: "When you don't need working capital to grow, then you can grow without external funding." It has done this across roughly 50,000 devices, grew 60% in 2024 and 90% in 2025, and runs at 25–30% EBITDA margins. Top Founders (Sept 24).
$400 in the bank to $50M in recurring revenue. Jacqueline Samira, founder of Howdy, which builds software teams for U.S. companies across 11 offices in Latin America (more than 600 people), gave a clear breakdown of her fundraising. Before Y Combinator, "no one wanted to talk to me." After YC (7% of the company for about $125–150K), she had 93 investor meetings coming out of Demo Day, raised $2.9M, and a year later raised $18M, for $21M in total. She still owns about 35% and her co-founder about 23% (they started 60/40). Howdy is cash-flow positive again with $9M in the bank, and it keeps a flat 15% margin on every placement. Her advice, looking back: "If you have the ability to not fundraise, it's always better because you just make better business decisions." And on choosing investors: "Who you have on your cap table makes all the difference." Young and Profiting, "How She Scaled Howdy From $400 Left to $50M in Recurring Revenue" (Sept 23).
Bootstrapped to a $47.5M exit, after being turned down by VCs. Ashley Tyrner-Dolce started FarboxRx to deliver healthy food to low-income families. Every VC she pitched wanted her to become a meal-kit company, which would have meant building for customers who could afford $10 a meal. One partner at a well-known San Francisco firm asked her, "Do poor people want to eat healthy food?" After striking out, she had what she calls "the burrito moment" at San Francisco airport. The company survived by cutting costs, with her and her COO doing "75 jobs" and her giving up her own salary "more times than I ever want to remember" to make payroll. The cash that came in during COVID paid for a full pivot, from selling to consumers to working with Medicare and Medicaid health plans, which pay for the food. She walked away from several term sheets and buyout offers that would have gutted the mission, and later sold for $47.5 million. Her main tip for fundraising: after every "no," ask the VC which other investors might be a fit. Build Mode, "How a first-time founder bootstrapped her way to a $47.5 Million Exit with Ashley Tyrner-Dolce, FarboxRx" (Sept 24).
Grüns sold to Unilever about three years after launch. The gummy-nutrition brand, which already had around 130 employees, was bought by Unilever. The Startup CPG host said it "holds the record for fastest exit or at least largest, fastest exit in CPG history" (CPG means consumer packaged goods, like food and toiletries). No price was given on the podcast. Co-founder Juliet Choi credits an internal rule the team calls "taste to retain": samples go out weekly and get a verdict within about 24 hours, because customers only come back if the product tastes good. The Startup CPG Podcast, "How 'Taste to Retain' Built a Billion-Dollar Supplement Brand" (Sept 26).
What OpenAI's "free money" for YC startups actually costs. On This Week in Startups, Bryant Chou, a repeat unicorn founder (Calacanis mentioned his time at Webflow) who went back through Y Combinator with a new AI website-builder called Ploy, explained a deal Sam Altman offered every YC company this spring: about $1.5–2 million in OpenAI credits in exchange for an uncapped note with no MFN. In plain English: OpenAI gets a future stake in the company with no ceiling on the price at which that stake converts into shares, and without the right to match better terms other investors might get. Calacanis worked out that at a $300–400M valuation this is about 50 basis points (0.5%) of the company. Chou said "a good portion of my batch" took the deal and he wasn't worried about OpenAI seeing his data. Calacanis told founders to be careful anyway. This Week in Startups, "A rogue OpenAI agent hacked Australia's government. Does this matter? | E2342" (Sept 25).
A new mega-fund. Talia Goldberg said on TBPN that her firm, Bessemer Venture Partners, just raised $5.75 billion: $1.75B for early-stage companies, "often way before there's even revenue or sometimes even a company name," and $4B for growth rounds, where the firm plans to lead with large checks rather than "peanut buttering" small ones across many companies. Her reasoning: "Returns are concentrating … in fewer, larger winners." TBPN, "Jensen vs NYT, New Model Reactions …" (Sept 23).
Anthropic's IPO keeps slipping. (An ongoing story; here's what's new this week.) The listing, once aimed at October at a $2 trillion valuation, is now reportedly November or later, according to the All-In hosts citing the Wall Street Journal. They pointed to leadership's public comments about extinction risk as material problems for the IPO filing. They also noted that Anthropic's founders reportedly own only about 2% each, unusually low for founders at IPO. On his podcast, Patrick Boyle argued that even with extremely generous assumptions, about $65B in revenue at 31 times sales can't be justified by future cash flows. On the other side, family-office investor Vishal Verma told The Neon Show he bought into Anthropic at an $18 billion valuation and is sitting on a 55x paper return in 18 months. "Everything's on paper," he added. All-In, "Anthropic IPO at Risk, Meta's Muse Pop …" (Sept 26); Patrick Boyle On Finance, "Is Anthropic Worth Two Trillion Dollars?" (Sept 27); The Neon Show, "Anthropic, Sequoia Investor: 55x Returns in 18 Months …" (Sept 22).
Oura sets its IPO terms. (We covered the filing on Sept 14.) The smart-ring maker's shareholders are offering 50 million shares at $40–44 each, and most of the money goes to existing shareholders rather than the company. The Equity hosts called it "a big payday" for early backers. One of the Pivot hosts admitted he's "doing everything I can to try and find shares in this IPO." Equity (Sept 25); Pivot, "Trump TV, Meta's Muse Charm, and Xi in D.C." (Sept 25).