Newsletter · · Ashutosh Agarwal

Anthropic Passes OpenAI as the Most Valuable Startup at Nearly a Trillion Dollars - The Raise - Week of Jul 20–Jul 26, 2026

Startups and venture newsletter for the week of July 20 to 26, 2026. A staggering run of raises led by Anthropic passing OpenAI at a near trillion dollar valuation, alongside billion-dollar rounds for Fireworks AI and Travis Kalanick's Atoms, plus founder stories from Toast and Frozen One.

The Raise

Week of Jul 20–Jul 26, 2026: Anthropic Passes OpenAI as the Most Valuable Startup at Nearly a Trillion Dollars

The week AI's mega-round machine hit overdrive, Anthropic vaulted past OpenAI to become the most valuable startup on earth at nearly $1 trillion, and it wasn't even the only ten-figure round of the week.

Some weeks the funding news trickles. This was not one of those weeks. Between the podcasts we listen to for a living, the last seven days delivered a genuinely staggering run of raises: a near-trillion-dollar valuation, three separate rounds north of a billion dollars, and a supporting cast of AI companies, a Travis Kalanick comeback, and, because balance matters, a very fancy gas station. Below, every notable raise we heard, then two founder stories worth your time: the man who rebuilt the technology behind 1 in 5 American restaurants, and two twenty-somethings who turned a $250K check from a guy at a house party into a Target rollout.

This Week's Rounds

  • Anthropic, $65 billion at nearly a $1 trillion valuation (Series H). The AI lab behind Claude officially passed OpenAI to become the most valuable startup in the world. What's striking is how the money came together: this was less a cash raise than a land-grab for computing power. Roughly $15 billion came from cloud and chip partners, including $5 billion from Amazon, plus guaranteed capacity commitments, next-generation AI chips from Google and Broadcom, memory from Micron, Samsung and SK Hynix, and access to GPUs inside SpaceX's "Colossus" data centers. In plain terms, Anthropic is trading equity for a guaranteed supply of the silicon and power it needs just to keep operating. The company's revenue "run rate" (its most recent month's sales, annualized) has reportedly crossed $47 billion, versus OpenAI's last-known $852 billion valuation on about $24 billion. One healthy caveat the hosts stressed: take that $47 billion "with a grain of salt," because a big chunk of it is billed through cloud providers who keep a cut before the cash reaches Anthropic, so the headline number flatters the real, net revenue. Both companies, notably, are still losing enormous amounts of money. (Elon Musk Podcast, "Anthropic overtakes OpenAI as most valuable startup" (2026-07-23).)
  • Fireworks AI, $1.5 billion at a $17.5 billion post-money valuation. Led by Index Ventures, Gavin Baker, Lightspeed, and 20VC. Fireworks runs "inference" (the plumbing that actually serves AI models to customers once they're built) and the growth numbers are the story: over $1 billion in revenue run-rate, reached in about three and a half years, now processing roughly 40 trillion tokens a day, up from 15 trillion, with founder Lin Qiao expecting to hit $2 billion by year-end. 20VC's Harry Stebbings recounted writing Fireworks a $10 million check after a 15-minute meeting, a bet on the team and the sheer size of the inference market. (The Twenty Minute VC (20VC), "OpenAI and Anthropic Threatened by Kimi? ... Stripe Buying Paypal: What You Need to Know" (2026-07-23); backstory in "Are OpenAI and Anthropic Overvalued? ... with Lin Qiao, Founder and CEO @ Fireworks" (2026-07-20).)
  • Atoms, $1.7 billion, with investors including Andreessen Horowitz and Uber. This is Travis Kalanick's new venture, an "industrial AI and physical automation" company that recently bought Anthony Levandowski's autonomous-trucking startup Pronto (aimed at self-driving mining equipment). The TechCrunch hosts were candid about the mix of substance and hype: "part of me was somewhat surprised that they were able to raise $1.7 billion, including from Uber" (Kalanick's old company, from which he departed acrimoniously a decade ago) while wondering aloud "how much of Andreessen Horowitz's money did they pay just to be able to go on the posting spree they've gone on the last 24 hours of, like, 'Travis is back, baby.'" Pronto, they conceded, does have real customers among big mining companies. (Equity, "'AI communism', rogue models, and the why Kimi K3 spooked Wall Street" (2026-07-24).)
  • CuspAI, $450 million. An AI-for-materials-discovery company (think: using AI to invent new materials, grounded in real lab experiments). Founder Chad Edwards, who previously built a quantum-computing company, is doing something unusual for a software business: baking royalties into customer contracts, biotech-style, so CuspAI shares in the upside when a material it discovers goes into production. "When we started the company and we pitched this business model to investors, they said, 'Chad, forget about it. This is never going to work. You're never going to get big corporates paying you royalties on materials,'" he recalled, then spent two years proving them wrong. He described his own job as "a little bit analogous to a VC," running a portfolio of near-term and frontier materials bets. The round follows a $100 million Series A last year and comes alongside a large NVIDIA partnership. (TBPN, "China's AI Push Sounds Alarms, Airplane Seat Debate, Chris Best Joins ..." (2026-07-21).)
  • Augustus, $180 million Series B at a $1 billion valuation. Founder Ferdinand Dabitz is building what he calls an "AI bank," a modern version of the unglamorous but essential business of "correspondent banking" and dollar clearing (moving dollars around the world for other banks and fintechs). The pitch is automating away the "tens of thousands of analysts in the wire room" that a Citibank runs. Augustus has been conditionally approved as a U.S. national bank, "the eighth company since 2010" to clear that bar, per Dabitz, and already clears billions of euros for large fintechs. (TBPN, "China's AI Push Sounds Alarms, Airplane Seat Debate, Chris Best Joins ..." (2026-07-21).)
  • Ent.ai, $100 million, coming out of stealth. An AI-native endpoint-security company (endpoints = the laptops, servers and devices attackers try to break into). Founder Brandon and co-founder Lou left Microsoft to build what they call a "programmable endpoint" that augments the existing security tools rather than replacing them. He was frank about why a young company needs nine figures: cracking the Fortune 500 "takes a lot of effort... it's just expensive to build a product," and expensive again to land big enterprise accounts. He says he's already hired 75+ people without paying a recruiting fee. (This Week in Startups, "An AI that watches your every click may be the future of work | E2314" (2026-07-20).)
  • Socket, Series C at a $1 billion valuation. Founder Feross Aboukhadijeh is tackling software supply-chain attacks, the growing problem of hackers poisoning the free, open-source code that nearly every app is built on top of. As he put it, today "you can't even get 'Hello World' to show up on the screen without installing a thousand-plus dependencies," each a potential way in. Socket says it detects over a thousand attacks a week and protects more than 20,000 organizations, including OpenAI, Anthropic and Vercel. (Village Global Podcast, "How One Hacked Library Can Take Down Thousands of Companies | Feross Aboukhadijeh (Socket)" (2026-07-23).)
  • Sila, roughly $300 million. A battery-materials company (silicon anodes that make batteries hold more charge), now diversifying its customer base well beyond electric vehicles to names like Panasonic, Mercedes, Whoop, and drone and satellite makers, a survival move the hosts flagged as common as EV demand softens. (Equity, "'AI communism', rogue models, and the why Kimi K3 spooked Wall Street" (2026-07-24).)
  • Poolhouse, $55 million at a $100 million valuation. "Topgolf, but for billiards." The tables are rigged with tracking cameras, machine-learning projectors and AI-enabled cue sticks that beam guide-lines onto the felt to help beginners. The real business plan is to license that smart-table hardware and software to cruise ships, casinos and pool halls at a target of $1,000 per table. Worth noting for investors: the founders are twin brothers Steve and Dave Jolliff, who previously built Topgolf (sold to Callaway for $2 billion) and then Puttshack, the definition of "rinse-and-repeat" founders. (The Best One Yet, "'Big Dumb Line', Our retail line obsession. Trump's $100k pay-to-tweet. TopGolf for Pool. +TechNeck" (2026-07-20).)
  • Maggie's Refuel, $2 million. For sheer contrast with the billions above: a venture-backed "nice gas station" in Los Angeles serving lattes and local pastries, with EV charging and an order-ahead app. Its early investors are Starbucks and 7-Eleven executives. Named, naturally, after the founder's French bulldog. (The Best One Yet, "'Glammed Gas', Costco's 1st gas station. World Cup's $15B triumph. Chanel's oldest shirt. +The 5 Coffee Rule" (2026-07-21).)

Founder Story of the Week: How Toast Rebuilt the Restaurant

If you've paid a restaurant check on a handheld tablet lately, you've touched Toast, a company that now powers nearly 1 in 5 restaurants in America and generates more than $2 billion in annual revenue. On How I Built This, co-founder and CEO Aman Narang walked through an origin story that is really a lesson in surviving the "no."

It started as a failure. In 2012, Narang and his co-founders Steve Fredette and John Grimm, all veterans of a Boston e-commerce company, Endeca, that Oracle bought for about $1 billion, built a simple app to let diners pay the check from their phones. They tested it at a bar downstairs from their old office, got about 50 people to pay at the launch party, and then... nothing. "It was actually quite buggy," Narang admitted, and "we were never able to figure out how to get users to pay with any sort of velocity after that first day."

The pivot came from listening. As they hung around Boston restaurants trying to save the app, they noticed something: "Nobody really liked their point-of-sale system." The existing technology was a tangle of disconnected software: one restaurateur, Chris Kane of Finale Desserts, showed Narang a business where the point-of-sale, payments, accounting, inventory, payroll, scheduling and online ordering were all separate systems that didn't talk to each other. The detail that stuck: in 2013, online orders were arriving "into a fax machine." So they made the bold call to stop building one feature and instead replace the entire operating system of a restaurant, cloud-based, running on their own Android hardware.

Then came the fundraising wall. Their old boss, Endeca founder Steve Papa, put in the first money, "I think it was $500K to get started. He believed in this idea." Narang's own read on it: "I think he believed in us more than the idea." Institutional VCs were another matter. The feedback was brutal and repetitive: restaurants "are a tough business... they don't buy online," margins "are low," they're "not tech-forward," and "you're trying to build a platform that's going to take you years to build," all while taking on entrenched giants like Micros and NCR. Nobody in the ecosystem wanted to integrate with them either: "Nobody wanted to integrate with us because we were nobody... So we just said, to hell with it, let's just try to build as much as we can on our own."

What actually unlocked scale wasn't a marquee VC, it was a food distributor. Papa connected them to Gordon Food Service (GFS) out of Michigan, an old Endeca customer, and that partnership got Toast out of Boston and into Miami and Chicago.

The most Toast anecdote of all: an early nightclub owner asked what happens if the system goes down at 1 a.m. Narang promised "great 24/7 support." The owner immediately called the support number on the website, a Google Voice line that rang all six or seven employees' phones at once. "We would wait for 2 or 3 rings to see if someone else would pick it up because no one really loved picking up the support calls," Narang laughed. On the third ring he answered: "I told you I've got really good 24/7 support." That customer is still with Toast. His takeaway for founders, the thing that nearly sank them in the early days wasn't competition, it was their own ability to execute and scale: "It was clear that our execution was getting in the way... of us actually scaling this company." (How I Built This with Guy Raz, "Toast: Aman Narang. How a Long Wait for the Dinner Check Launched a $2 Billion Business." (2026-07-20).)

A Second Founder Story, Because It's a Masterclass in the Raise: Frozen One

If Toast is the marquee name, Frozen One is the one every early-stage founder should study for the mechanics of raising with almost nothing to show. On the Startup CPG podcast, co-founders Alan Chen and Conner Mennig narrated a raise that went from a house party to a Target rollout in about a year. Their product: a high-protein ice cream, 40 grams of protein, under 400 calories, that started with a Ninja Creami machine in Conner's apartment.

The first check set everything else in motion. "There's the first round... which was we had nothing but, like, a story and a product. We had absolutely no financials. We had absolutely no retail commits." The believer was "a local investor in Austin that Conner actually met at, like, a house party... a prominent trader in Austin. And we ended up raising our first $250K from him." They raised it on a SAFE (a common early-stage instrument that converts to equity later), and that first check "set the terms for everybody else."

Then Target changed the math. In mid-January, Target told them: there's a shelf reset in three months, we want you in 1,500 doors, and we'll take a Target-exclusive flavor. "Of course we can. Let's go do it tomorrow," was the answer, and they immediately went into fundraising mode, because a rollout like that costs money. The problem: they'd ended the prior year with just $29,000 in revenue. "We probably had 15 VC calls" in two weeks, Chen said, and got "so many people just saying... 'it's too early.'"

The yes came from angels who move fast. An angel group (principals Chris Robb, Adam Spriggs and Jared Cohen) took the first call and emailed back two hours later asking to talk again the next day. That became an oversubscribed $2 million round that let them actually deliver on Target. Along the way they leaned on a velocity story from Central Market, a Texas grocer whose buyer Chen had cold-called. His blunt tally on the grind: "We got 100 nos before we got yes, but every single one of those nos taught us something." And the throughline: "Every single thing that's happened to us has been because of relationships."

Where they are now: closing a $5.5 million round with Brand Foundry, an introduction that came, of course, through that same angel group. This round is about building the team: they've already hired a COO who was previously VP of Supply Chain at Van Leeuwen Ice Cream (where he commercialized 90+ flavors) and are hiring a VP of Sales and a head of marketing. "We're smart enough to know what we don't know," Chen said. (The Startup CPG Podcast, "Founder Fundraising Journey: Frozen One, Alan Chen & Conner Mennig" (2026-07-25).)

Also Heard

  • China's open-source labs rattled everyone. The week's other big money story was Beijing-based Moonshot AI, whose Kimi K3 model "moved markets on Friday." Moonshot is reportedly working to close a funding round at roughly a $30 billion valuation and eyeing a Hong Kong IPO within six months, with China International Capital Corp and Goldman Sachs involved. (A separate podcast pegged the valuation closer to $20 billion, the number is still in motion.) The rise of cheap, capable Chinese open-weight models is exactly what has OpenAI and Anthropic executives lobbying Washington. (Bloomberg Intelligence, "Moonshot Plans IPO in Six Months After China AI Breakthrough" (2026-07-20); Moonshots with Peter Diamandis, "The Hugging Face Breach, Moonshot AI Valued at $20B, and Living to 1,759 Years Old | EP #273" (2026-07-24).)
  • How Anthropic got here, the $4 billion bet. For context on the trillion-dollar headline: Menlo Ventures' Matt Murphy recalled leading Anthropic's Series D with over $500 million back in 2024, at a $4 billion valuation, pre-revenue and pre-launch, largely on the strength of founder Dario Amodei's pedigree and an early lock-up of Google and Amazon as investors and distributors. From $4 billion to nearly $1 trillion in about two years is the single cleanest illustration of how fast this cycle is moving. His lesson for founders today: "a great model isn't enough." (Equity, "Menlo Ventures' Matt Murphy says the lesson for founders now is that a great model isn't enough" (2026-07-22).)
  • A private-equity exit finally happens. Away from AI, Blackstone is taking sandwich chain Jersey Mike's public, looking to sell up to $1.1 billion in an IPO that would value it around $8 billion, a fast turn given Blackstone only closed its buyout in early last year, and a rare sign of life in a sluggish market for private-equity exits. (Bloomberg Intelligence, "Moonshot Plans IPO in Six Months After China AI Breakthrough" (2026-07-20).)
  • The quiet trend: founders structuring rounds for control. On a niche distribution podcast, Conor Leen of Lantern (an AI demand-forecasting startup he and two Stanford classmates built after doing 200 interviews with distributors and recruiting a CTO who ran Walmart's applied-AI forecasting lab) described structuring his recent venture round specifically to preserve full founder control and board seats while still paying top-of-market. A small data point, but a recurring theme in a market where founders with real traction are pushing back on terms. (Distribution Talk, "What If Machine Learning Could Prevent Your Next Inventory Crisis? With Conor Leen, Lantern" (2026-07-22).)