Anthropic went from $87 million to $47 billion in annualized revenue in 28 months. No technology company has ever grown this fast. The speed proves that AI demand is real. It also proves that nothing protects it.
Anthropic generated $87 million in annualized revenue in January 2024. By May 2026, that figure reached $47 billion. The trajectory in between: $1 billion by December 2024, $9 billion by the end of 2025, then $14 billion in February, $19 billion in March, $30 billion in April, and $47 billion in May. That works out to roughly 540 times growth in 28 months.
No technology company has ever scaled revenue this fast. Google crossed $10 billion in annual revenue eight years after its founding. Facebook took ten. Amazon Web Services took a decade. Anthropic reached $47 billion in annualized revenue five years after its founding, with most of the growth compressed into the final twelve months. On June 1, the company confidentially filed for an IPO at an implied valuation of $965 billion.
In April, Anthropic's Claude surpassed OpenAI's ChatGPT in business adoption for the first time. Ramp, the corporate finance platform that tracks spending across more than 50,000 U.S. businesses, measured Anthropic at 34.4 percent adoption versus OpenAI's 32.3 percent. Two years earlier, Anthropic's adoption rate was 0.03 percent. More than 70 percent of Fortune 100 companies now use tools built on Claude. Over 500 customers spend more than $1 million per year.
These numbers are extraordinary. They also carry information that gets overlooked in the celebration. The speed of a revenue ramp reveals something fundamental about the product it represents.
When Google took eight years to reach $10 billion, the slowness was the point. Enterprise software adoption requires integration work, staff retraining, workflow redesign, and data migration. Every year a customer stays deepens the relationship and raises the cost of leaving. The slow growth built an economic moat that still holds two decades later. Anthropic's growth tells a different story. Enterprises adopted Claude at unprecedented speed because adding an AI model to a workflow often requires changing a single API endpoint and a model name. The performance gains are immediately measurable. The decision to switch is reversible at negligible cost.
The evidence is already visible from the other direction. ChatGPT commanded 87 percent of global AI chatbot web traffic in early 2025. By March 2026, that share had fallen below 57 percent. A 30-point decline in fourteen months. OpenAI's product did not deteriorate. Its competitors became good enough, and users discovered that switching was trivial.
The pressure is intensifying from below. Chinese AI labs including DeepSeek, Kimi, and Zhipu now offer models priced up to nine times cheaper than Claude for equivalent work. In early June, reports surfaced that OpenAI is considering drastic reductions in token pricing, treating the Chinese cost advantage as an existential threat rather than a peripheral one. When the incumbents slash prices to match new entrants, the economics look less like software and more like petrochemicals.
This is the central tension in AI's business model. The demand is real. Alphabet's cloud backlog hit $460 billion. The five largest hyperscalers will spend more than $725 billion on AI infrastructure this year. Google processes 19 billion tokens per minute through its APIs, six times more than a year ago. Nobody questions whether businesses want AI. The question is whether AI models are products or commodities. Products have differentiation, switching costs, and pricing power. Commodities compete on cost, with margins converging toward the lowest-cost producer's break-even plus a thin premium.
Berkshire Hathaway's bet is instructive. Greg Abel wrote a $10 billion check into Alphabet's recent $85 billion equity raise. He did not invest in Anthropic or OpenAI. Berkshire has spent sixty years buying infrastructure: railroads, utilities, pipelines, energy. The logic is that durable economics live in the physical layer underneath the models, in the data centers and chips and power plants, not in the models themselves. If AI is a commodity, the money accrues to whoever controls the infrastructure. The same way it works in oil.
The fastest revenue ramp in technology history may ultimately demonstrate something its beneficiary would prefer not to advertise. Anthropic's $47 billion proves that AI demand is massive and accelerating. It also proves that the barriers protecting any single provider are close to zero. The same frictionless adoption that delivered $47 billion in 28 months can redirect it somewhere else just as fast. The speed is the signal.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)