Anthropic Was Founded by People Who Left OpenAI Over a Safety Disagreement. Now It Is Filing for a $1 Trillion IPO and Beating OpenAI at Its Own Game.
By Creatives Takeover · June 16, 2026
The rebels built the better company. Here is how.
The Meeting That Started Everything
In December 2020, a group of researchers sat together on Zoom and decided to walk away from the most influential AI lab in the world.
Dario Amodei was the VP of Research at OpenAI. His sister Daniela was the VP of Safety and Policy. They had helped build GPT-2 and GPT-3, the models that proved large language models could do things nobody expected. They understood better than almost anyone what was coming next. And that understanding was exactly the problem.
The disagreement was not about capability. It was about priority. Dario and his colleagues believed that as AI systems became more powerful, safety research needed to scale alongside them. Not as a footnote. Not as a PR function. As a core architectural commitment baked into how models were trained and deployed. OpenAI, in their view, was not treating it that way.
So they left. Dario, Daniela, and seven other OpenAI researchers resigned and started building from scratch inside a Zoom call during the second wave of a global pandemic. They picked a name from a spreadsheet that included options like Aligned AI, Sponge, Swan, and Sparrow Systems. They chose Anthropic, a word that connotes being human-centered, and which happened to have an available domain name in early 2021.
Five years later that decision is worth nearly a trillion dollars.
The Company Nobody Took Seriously at First
The early criticism of Anthropic was sharp and consistent. It was a company founded by idealists who had left the most commercially aggressive AI lab in the world to build something more responsible. The implied reading was that responsibility and competitiveness were in tension. That prioritizing safety meant accepting a slower, smaller, less commercially viable outcome.
The first version of Claude was finished in the summer of 2022. Anthropic did not release it immediately, citing the need for further internal safety testing and a deliberate desire to avoid triggering a race to release increasingly powerful AI systems before anyone was ready. In an industry where speed was everything, Anthropic chose to wait.
That decision looked cautious at the time. In retrospect it was the foundation of the trust that became their primary commercial advantage.
Claude launched publicly in March 2023. By August 2025, Anthropic's annualized revenue had crossed $5 billion. By December 2025 it was $9 billion. The numbers after that are where the story becomes genuinely unusual.
The Revenue Curve That Has No Precedent
Between January 2025 and May 2026, Anthropic's annualized revenue grew from $1 billion to $47 billion. That is a 47-fold increase in 16 months.
Venture capitalists who reviewed the company's financial data before its Series H stated that such growth is unprecedented. The firm had studied more than 200 software company IPOs and found no comparable trajectory.
The primary engine of that growth was not consumer adoption. It was enterprise. Over 1,000 enterprise customers now spend more than $1 million per year on Claude, up from roughly a dozen two years ago. That number doubled in under two months between February and April 2026 alone. Eight of the Fortune 10 are Claude customers. Approximately 80 percent of Anthropic's revenue comes from businesses rather than individual consumers, compared to roughly 40 percent for OpenAI.
Enterprise revenue is structurally different from consumer revenue. It is stickier. It renews on annual contracts. It expands as organizations integrate Claude more deeply into their workflows. And it does not disappear overnight when a competitor launches a better consumer product. The 80-20 split in Anthropic's favour was not an accident. It was the result of building a product that enterprise buyers trusted for the exact reason Anthropic was founded: a genuine commitment to safety, interpretability, and reliability.
The product that accelerated everything was Claude Code. Launched publicly in mid-2025, it became the fastest-growing product in Anthropic's history, hitting $1 billion in annualized revenue within six months of launch and reaching $2.5 billion in run-rate revenue by February 2026. It is an AI coding agent embedded directly into developer workflows, replacing hours of manual work with minutes of AI-assisted output. Weekly active users doubled between January and February 2026 alone.
The Moment Anthropic Passed OpenAI
In April 2026, Anthropic's enterprise AI market share reached 34.4 percent, surpassing OpenAI's 32.3 percent for the first time. US corporate spending data tracked by Ramp Economics Lab shows that Anthropic's share of combined enterprise spending on OpenAI and Claude had grown from 10 percent in early 2025 to over 65 percent by February 2026.
The valuation followed the revenue. Anthropic closed a $65 billion Series H in late May 2026, pushing its post-money valuation to $965 billion, edging past OpenAI on secondary markets for the first time. On June 1st, just four days after closing that round, Anthropic filed a confidential S-1 with the SEC, becoming the first major AI lab to formally begin the public market process. A listing on the Nasdaq or NYSE is expected as early as October 2026, potentially making it the first AI company to debut at a one-trillion dollar valuation.
For context: in January 2025, Anthropic was valued at approximately $60 billion. Eighteen months later it is approaching a trillion. The multiple expansion is unlike anything the venture market has seen in a generation.
The Structural Advantage Nobody Predicted
The reason Anthropic is beating OpenAI in the enterprise market is the same reason critics thought it would lose.
Enterprise buyers do not choose AI infrastructure the way consumers choose an app. They run procurement processes, involve legal and compliance teams, negotiate data handling agreements, and ultimately choose the vendor they trust with their most sensitive workflows. That trust is not built on benchmark scores. It is built on track record, transparency, and a demonstrated commitment to the things that keep a general counsel comfortable.
Anthropic's entire founding thesis was that safety was not a constraint on capability but a precondition for trustworthy deployment. That argument felt philosophical in 2021. By 2026, with AI systems embedded in legal review, financial analysis, healthcare documentation, and customer service for Fortune 10 companies, it became the most commercially relevant argument in the industry.
The gross margins tell the same story. Anthropic's inference infrastructure gross margin jumped from 38 percent a year ago to over 70 percent today. OpenAI is projected to spend $125 billion per year on model training by 2030. Anthropic's projection for the same period is approximately $30 billion. Same race. Four times the cost difference. And Anthropic projects positive free cash flow by 2027, three years before OpenAI expects to reach profitability.
The Risks That Remain Real
None of this means the story is without tension.
Anthropic reports revenue from its cloud partners, AWS, Google Cloud, and Microsoft Azure, on a gross basis, counting the full amount billed through those channels as revenue and booking partner payouts as expenses. That accounting practice inflates headline figures relative to net-reporting peers, and the full S-1 will need to resolve how investors should interpret those numbers compared to the software companies they are used to valuing.
The company is also deeply dependent on the cloud infrastructure of its largest investors. Amazon has committed over $8 billion to Anthropic. Google has committed billions more. Both are also competitors in the AI model market. The dependency and the rivalry exist simultaneously, and public market investors will price that risk into any valuation.
And the competitive pressure is real. OpenAI is not standing still. xAI's Grok is embedded across the X platform with distribution that Anthropic cannot replicate. Google DeepMind's Gemini runs natively on the world's largest consumer hardware ecosystem. The enterprise lead Anthropic holds today was earned. It will need to be defended every quarter.
Five Things This Story Teaches Founders
The contrarian position is only valuable if you hold it long enough to be proven right. Anthropic's safety-first positioning looked like a liability in 2021 and became the source of its commercial dominance by 2026. The founders who quit when the market did not immediately validate their thesis would have missed the entire payoff.
Slower launches can build faster companies. Anthropic finished Claude in 2022 and waited until 2023 to release it. That decision cost them early press cycles and first-mover advantage in the consumer market. It bought them the trust that became 80 percent enterprise revenue.
Enterprise stickiness compounds in ways consumer growth does not. One thousand customers each spending over a million dollars a year is a more durable business than ten million consumers each paying ten dollars a month. The revenue looks smaller until the renewals start coming in.
The product that breaks out is rarely the one you planned. Claude Code was not Anthropic's founding vision. It was a developer tool that found product-market fit faster than anything else the company had ever built and became the engine of its most significant growth period. Build the infrastructure, then watch where usage concentrates.
A clear founding principle is not just a values statement. It is a go-to-market strategy. Dario Amodei did not leave OpenAI to build a better chatbot. He left to prove that safety and capability were not in conflict. That proof, delivered consistently over five years, became the reason enterprise buyers chose Claude over every alternative. The mission was the moat.
The people who left OpenAI over a disagreement about how AI should be built are now filing to go public at a valuation that has never existed before in the history of AI companies. The argument they were making in December 2020 on a Zoom call is the argument the market has decided to believe.