Anthropic customer demand is now the central question for investors, according to a new analysis. As Anthropic prepares to go public, the company has become the fastest in history to reach $50 billion in revenue, but the simple models used to value AI growth no longer work.
How Claude Code changed the story
Funda AI looks back a year. At that time, rental and contract prices for H100 and A100 GPUs were sliding every month, and stories about Microsoft trimming orders and canceling data centers were everywhere. OpenAI still drew most of the attention.
The newsletter gives Claude Code the credit for the recovery and points to three changes. Teams now expect AI to write all of their code, not half. Engineers supervise dozens of agents instead of one. And each agent can work for hours, sometimes rebuilding software it found online.
Funda AI says the gains stacked on each other and lifted Anthropic’s monthly growth to 50%.

New math for AI revenue
Funda AI says this changed two assumptions. Monthly spending per person no longer had to stop at the price of a ChatGPT subscription. And revenue per gigawatt of data center capacity could climb from $10 billion to $20 billion, or even $50 billion.
So the market began estimating annual recurring revenue (ARR) from future compute capacity instead of extending a straight line. If revenue per gigawatt holds steady, the split between training and inference capacity is enough to derive an ARR path.
Why that model is breaking
The authors say the picture has grown more complicated for two reasons. Competition between OpenAI and Anthropic is intensifying, and their combined ARR growth is slowing.
Anthropic’s S-1 also shows inference’s share of total costs falling sharply from the second quarter of 2026, with training reaching 75% of costs that quarter. That makes it harder to infer ARR from data center capacity or to judge the return on AI spending as a whole. Funda AI compares the closer approach to AGI to the final sprint of a 100-meter race, where training takes a growing share of the effort.
The newsletter praises Anthropic, calling it one of the great companies of this century and saying it takes AI safety seriously even when that could hurt revenue.
Questions the research set out to answer
Funda AI says it did field research before the IPO, and admits it still lacks complete answers. Its list of open questions:
- Whether startups, tech firms and traditional enterprises are buying at different rates, and where each group heads next.
- The effect of KV caching and model routing on ARR.
- Whether today’s budgeting for tokens resembles the software cost-cutting of 2022-2023.
- What rivalry from OpenAI and open-weight models does to ARR.
- Which applications lock customers in.
- Which uses beyond today’s could add revenue in 2027.
- Whether recent NVIDIA news changes the picture for Anthropic.
Several of these will need more company disclosures, the authors write.
The anniversary framing
The newsletter opens by noting that four years have passed since ChatGPT launched and set the AI narrative in motion. It says the story is approaching its biggest moment now that Anthropic is preparing to go public, a week before that anniversary.
Funda AI’s own track record is part of the pitch. It reminds readers of a series of reports it published during the months when GPU prices were falling and data center cancellations were in the news.
It adds that it expects further company disclosures to be needed before several of its questions can be settled.
What this means for Anthropic IPO watchers
This is analysis from an outside newsletter, not company guidance, and its figures such as the $50 billion revenue and 50% monthly growth are the authors’ claims. Still, the core point is concrete. If training eats a larger share of costs each quarter, revenue forecasts built on data center capacity will miss, and the debate about Anthropic customer demand will shift to which customers stay.
The next evidence will come from the company’s first public earnings reports, where investors can test these questions against reported numbers.
Source: fundaai.substack.com