Anthropic is heading into what could be the largest IPO in history with a problem it probably did not want analysts to notice: businesses are not buying its most expensive model. According to spending data from 70,000 companies collected by payments firm Ramp, and reported by the Financial Times, Fable 5, Anthropic’s flagship and priciest model, accounts for only around 11 percent of total enterprise spending on Anthropic tools. More than two months after launch, that number has stopped growing.
That is a significant break from how enterprise AI adoption has worked. Until recently, corporate buyers defaulted to the most capable model available, assuming that more power meant better results. That assumption has now cracked. Analysts and investors close to Anthropic say the shift is straightforward to explain: Fable 5 is expensive, and older models handle most real business tasks well enough. So companies are staying put on cheaper tiers.
This matters well beyond Anthropic’s balance sheet. The entire frontier lab model, where labs justify spending billions on training runs by selling access to the most advanced outputs, depends on customers actually wanting the frontier. If enterprise buyers are satisfied with last year’s model at half the price, the financial logic of racing to build the next one gets harder to defend. Miles Clements, a partner at Accel, which has invested close to $1 billion in Anthropic, said the era where customers reliably chose only the most powerful option “was not a durable era.” That is a striking thing for a major investor to say out loud.
Fable 5 also had a troubled start. The Trump administration forced Anthropic to pull the model shortly after its early June launch, citing national security concerns. It was cleared for relaunch on July 1, and political risk has since faded as a primary concern for buyers. But the damage to momentum was real. Slower initial uptake, combined with the data retention rules imposed by the administration that further restricted Fable’s rollout, left the model playing catch-up in a market that had already moved on.
The competitive picture has also shifted. OpenAI, which had a sluggish first half of 2026, recovered sharply after releasing GPT 5.6 in July. That model is priced significantly below Fable 5, and OpenAI’s annualised revenue has since jumped 35 percent, now exceeding $40 billion. Meanwhile, cheap open-weight models from Chinese labs have given enterprise buyers even more alternatives. The result is a pricing war that Anthropic’s cost structure may not be built to win.
There are signs Anthropic is adapting. Opus 5, a smaller and cheaper model released in late July, has already surpassed Fable 5 in enterprise spending share, per Ramp’s data. That is partly good news, it shows customers are staying in the Anthropic ecosystem. But it also means the company is generating less revenue per customer than the Fable 5 adoption curve would have implied.
The headline numbers still look strong. Anthropic’s annualised revenue hit $65 billion in July, up from $47 billion in May, and nearly sevenfold higher than the start of the year. The company recorded its first adjusted operating profit in Q2 and has signaled it expects to be profitable again in Q3. It also has 6,000 customers spending $100,000 or more annually. But July revenue came in below the most bullish investor projections, which had penciled in $80 billion annualised.
With a potential $2 trillion valuation on the table and an IPO that could arrive as soon as next month, every data point now carries extra weight. Ramp’s lead economist Ara Kharazian put it bluntly: predicting Anthropic’s trajectory even a few months out is nearly impossible. The company is growing fast, but the model that was supposed to anchor that growth is not performing the way the bull case required.
