How a Disagreement at OpenAI Built a $965 Billion Company

Month 1 covered four companies that reshaped their categories: Midjourney built a $500M empire with no investors. ElevenLabs went from struggling to raise $2M to an $11B voice AI platform. Perplexity built the most credible challenger to Google in twenty years without training a single LLM. Cursor became the fastest-growing SaaS company in history and sold to SpaceX for $60 billion.

Every one of those stories had a moment where the outcome looked impossible.

Month 2 starts with the company whose whole existence is a bet on an outcome most people still find hard to believe: that you can build the most powerful AI systems in the world and simultaneously make them safe. That those two goals are not in tension, they reinforce each other.

Anthropic is now worth $965 billion. It just overtook OpenAI to become the most valuable private AI company on earth. It filed a confidential S-1 with the SEC on June 1, 2026.

And it started with a disagreement in a conference room.

📌 No sponsor this week - Projects of AI just relaunched. If your product belongs in front of 2,400+ AI founders and builders with a 69% open rate, grab a slot here →

The disagreement that started everything

In December 2020, Dario Amodei walked out of the hottest company in tech.

He'd spent five years at OpenAI. He'd co-built GPT-2 and GPT-3, two of the most consequential AI systems ever created. He'd risen to VP of Research, the most senior technical role in the organisation. He had watched language models get better with each scaling step faster than anyone had predicted.

And he had concluded something his colleagues didn't fully agree on.

He believed two things simultaneously. The first: if you pour more compute into these models, they'll keep getting better, there's almost no end to this. The second: scaling alone doesn't give you safety. You don't tell models what their values are just by making them bigger.

The people at OpenAI weren't wrong about scaling. Amodei agreed with them entirely on that. What he disagreed about was how much weight to give the alignment problem, the question of how you ensure an increasingly capable AI system actually does what you want it to do and not something subtly, catastrophically different.

Most accounts agree: this group believed OpenAI was commercialising too fast and that its safety mechanisms couldn't keep up. The disagreement wasn't about the destination. It was about how carefully you should drive when you don't know what's around the next corner.

So he left.

Colleagues followed in the weeks after fourteen in total, eight of whom became co-founders.

They met on Zoom, it was the second wave of COVID, and everything was remote. Someone opened a spreadsheet of potential company names: Aligned AI, Generative, Sponge, Swan, Sloth, Sparrow Systems. And "Anthropic", a word that connotes being human-centered and human-oriented. It was also an available domain. "We like the name it is good," the team wrote in the spreadsheet. And thus, Anthropic was born.

The siblings at the center

The Anthropic story has many characters, eight co-founders, each with extraordinary credentials. But the story runs through two people: Dario and Daniela Amodei. Brother and sister. Two completely different skill sets. One shared conviction.

Dario Amodei — CEO. Born 1983. PhD in biophysics from Princeton. Started his career at Google Brain before joining OpenAI in 2015. His research background is unusual for a CEO, he's deeply technical, spent years doing the actual work of building and studying large language models, and is more comfortable talking about mechanistic interpretability than business strategy. His defining characteristic isn't charisma. It's intellectual honesty, a willingness to say "we don't know" about AI safety when most people in the industry were selling certainty.

Daniela Amodei — President. VP of Safety and Policy at OpenAI, her most senior title before leaving. She also held VP of People earlier in her tenure. Before OpenAI she spent five years at Stripe as an early employee. Her academic background is one of the most striking details in this whole story: Daniela has a BA in English Literature from UC Santa Cruz, Not computer science. Not engineering. English Literature. The President of the most valuable private AI company on earth studied literature. She runs business operations, governance, and partnerships at Anthropic and she is the reason Anthropic broke into regulated industries like finance, healthcare, and government, where trust is the product and one mistake ends the relationship. Both siblings were named to the TIME 100 list of most influential people.

The founding team also included some of the most credentialed AI researchers alive:

• Tom Brown — first author on the original GPT-3 paper

• Jared Kaplan — led scaling laws research at OpenAI, establishing mathematically how models improve with size

• Chris Olah — widely regarded as the world's leading researcher on neural network interpretability

• Jack Clark, Sam McCandlish, Ben Mann — all senior OpenAI researchers

Eight people. Every single one of them had helped build the technology they were now leaving to make safer.

The thesis that most investors found contradictory

They registered Anthropic in California in January 2021, with $124 million in their first outside funding round, a Series A.

The seed investors - Jaan Tallinn (co-founder of Skype), Dustin Moskovitz (Facebook co-founder), and others were betting on a thesis most venture capitalists found difficult to parse: build the most powerful AI systems in the world and simultaneously figure out how to make them safe.

The contradiction, as most investors saw it: if you really believed AI was dangerous, wouldn't you slow down? And if you were going to slow down, how would you compete with OpenAI, Google, and everyone else who wasn't?

Dario's answer and it's an answer worth sitting with, was that the only way to make AI safe is to be at the frontier. You can't develop safety techniques for systems that don't exist yet. You need to build the powerful systems, study them, understand what's inside them, and develop the tools to steer them before someone else builds them without those tools.

Safety-focused labs staying behind the frontier don't make AI safer. They just cede the frontier to labs that don't care about safety.

This was a genuinely controversial position in 2021. It still is. But it's the intellectual spine of everything Anthropic has built.

The SBF chapter - the detail nobody dwells on

In 2022, a significant financial push came from Sam Bankman-Fried. Through Alameda Research, he led a $580 million round at the time, one of the largest single investments in AI history.

FTX collapsed in November 2022. Bankman-Fried was arrested, convicted, and sentenced to 25 years in prison.

Anthropic returned the money or more precisely, the bankruptcy proceedings unwound the investment. The company survived the reputational hit, partly because they'd already established their research credibility and partly because the broader AI moment of 2023 overwhelmed everything else.

It's worth mentioning not because it defines Anthropic, but because it's a real chapter in the company's history that gets quietly skipped in most accounts. The fastest-growing AI company in the world had SBF as an early major backer. That happened.

Claude named after a mathematician who founded information theory

Anthropic's flagship model is named Claude, in honor of Claude Shannon, the mathematician who founded information theory.

The choice of name is intentional in a way that's easy to miss. Shannon's work on how information is encoded, transmitted, and understood is foundational to the mathematics that underlies modern AI. Naming their model after him is Anthropic signaling something about how they think about the work: as part of a longer scientific tradition, not just a product release.

Claude launched in March 2023, three months after ChatGPT. The positioning was direct: a safer, more honest, more steerable AI assistant. Not "we won't help with bad things" as a constraint but "we've designed this from the ground up to be aligned with what you actually want, including when what you actually want is the truth."

The Constitutional AI approach where Claude is trained to evaluate its own outputs against a set of principles was Anthropic's technical answer to the alignment problem they'd left OpenAI to solve. Whether it works well enough remains an active research question. It works better than most alternatives that existed at the time.

The growth trajectory that rewrote every forecast

What happened to Anthropic's revenue over seventeen months is something that financial analysts are still struggling to frame properly.

In December 2024, Anthropic had $1 billion in annualised revenue. Seventeen months later, May 2026 that number reached $47 billion. A 47x increase in 17 months.

Let me put that in context. Salesforce took 24 years to reach $36 billion in annual revenue. Anthropic went from $1B to $47B in seventeen months.

In April 2026, Anthropic overtook OpenAI in annualized revenue. The company whose stated mission is safety not market dominance, not consumer adoption, not winning the AI race, became the highest-revenue AI company on earth.

The engine behind most of this growth: Claude Code. Launched as a standalone product in February 2025, Claude Code became generally available in May 2025, hit $1B in annualized revenue by November 2025, and reached $8B in annualised revenue by May 2026. Claude Code now accounts for approximately 54% of the AI coding market, a category that barely existed 18 months ago.

Total users: 245 million monthly active users as of July 2026.

The funding rounds that broke every record

The capital Anthropic has raised is difficult to comprehend even when you see the numbers:

• Series A (May 2021): $124M. Jaan Tallinn (Skype co-founder), Dustin Moskovitz (Facebook co-founder), Eric Schmidt.

• Series B (Apr 2022): $580M. Led by SBF/Alameda Research. (Later unwound.)

• Series C (May 2023): $450M. Google led. Spark Capital.

• Series D (Dec 2023): $750M. Google ($300M), Spark Capital.

• Amazon investment (2023–2024): $8B total across three tranches.

• Series E (Mar 2024): $2.75B. Amazon led.

• Series F (Sep 2025): $13B. ICONIQ led, Fidelity, Lightspeed. Valuation: $183B.

• Series G (Feb 2026): $30B. GIC and Coatue led. Valuation: $380B.

• Series H (May 2026): $65B. Altimeter, Dragoneer, Greenoaks, Sequoia. Valuation: $965B.

Total raised: approximately $125 billion, the most ever raised by a private technology company.

The Series H alone $65 billion is the largest single venture capital round in history.

The $965 billion valuation makes Anthropic the most valuable private AI company on earth ahead of OpenAI's last reported figure of approximately $852 billion.

The IPO that could make history

On June 1, 2026, Anthropic filed a confidential S-1 with the SEC at a valuation of $965 billion.

A Nasdaq listing could come as early as October 2026.

Anthropic is reportedly on track for its first ever profitable quarter in Q2 2026, a different narrative from OpenAI's high-revenue, high-loss profile heading into public markets.

If the IPO prices at or above the Series H valuation, Anthropic becomes the first AI company to list at a valuation above $1 trillion joining Apple, Microsoft, NVIDIA, Amazon, and Google in a club that has never had an AI-native company.

The company that started with a disagreement about safety in a Zoom call during COVID. The company that couldn't get most investors to understand its thesis in 2021. The company that had SBF as an early backer and survived it.

Nearly a trillion dollars.

What I keep coming back to

There's a version of the Anthropic story where the safety thesis is just positioning, a way to differentiate from OpenAI, to win the trust of regulated industries, to look responsible while racing to the frontier like everyone else.

I've read everything I can find on Anthropic. I don't think that's what happened.

The people who founded this company, the researchers, the safety scientists, the interpretability team genuinely believe that AI systems are getting more capable faster than our ability to understand and control them. They genuinely believe that if nobody solves alignment, the consequences could be catastrophic. They left a company that was, by any measure, more successful and more resourced to work on this problem specifically.

That doesn't mean they're right. The alignment problem may be more tractable than they fear, or less tractable than they hope. The Constitutional AI approach may work or it may not.

But the conviction is real. And in Silicon Valley where the stated mission is almost always a wrapper around a business objective, a founding team that walked away from OpenAI over a genuine intellectual disagreement about existential risk is something different.

The founding story doesn't fit the usual startup template. Most breakout companies start with a market insight or a technical breakthrough. Anthropic started with a disagreement about risk.

Nearly a trillion dollars later that disagreement turned out to matter.

That's the Anthropic origin story. On Friday I'm going inside the product - Constitutional AI, how Claude Code captured 54% of the AI coding market, the Amazon and Google partnerships, and what every builder can steal from how Anthropic won enterprise trust without winning the consumer brand war. See you then.

Hit reply — did you know the SBF chapter of this story? Or the Claude Shannon name origin? Tell me what surprised you most. I read every reply.

- Projects of AI

P.S. - The revenue trajectory I keep staring at: $1 billion ARR in December 2024. $47 billion ARR in May 2026. That's 47x in 17 months. Salesforce took 24 years to reach $36 billion. I've covered a lot of fast-growing companies in this newsletter. Nothing comes close to this trajectory. Nothing.

Enjoying this? Forward it to one person building in AI. That's the only growth strategy I have.

Want to reach 2400+ AI founders and builders? Sponsor Projects of AI