AI giants are learning a hard lesson about pricing power

Anthropic, until Friday’s White House move, looked like one of the more rationally valued companies in its peer group

Based on its annual revenue run-rate of nearly $50bn, Anthropic’s proposed $1tn valuation, assuming little or no debt, would represent a multiple of 20 © Samuel Boivin/NurPhoto/Getty Images

Published10 hours ago

First comes innovation, then comes negotiation. Anthropic’s unfortunate skirmish with the US government, which slapped an export ban on its most advanced AI models on Friday, shows that for any industry, creating impressive products is of limited use if a company can’t sell them profitably. That’s a fact that investors chasing thirteen-digit AI valuations might previously have missed.

Anthropic’s case is, to be fair, an odd one. The White House’s curb on the company’s newest models comes in the context of an already-fractious relationship between the two sides. But it raises a wider question that could weigh on Anthropic and its chief rival OpenAI as they head towards initial public offerings. What if being good at AI doesn’t equate to delivering handsome returns to shareholders?

The stakes couldn’t be higher. Elon Musk’s SpaceX, which also has designs on creating leading-edge AI, went public on Friday valued at more than 60 times its expected $30bn revenue for this year. Across the market, credulity is at historic levels. Palantir, an exemplar of market ebullience, crested last year at 96 times sales. By contrast, older tech companies such as Facebook owner Meta Platforms, Google parent Alphabet and Musk’s Tesla have never exceeded 25 times, according to LSEG data.

Anthropic, until Friday’s intervention, looked like one of the more rationally valued AI companies. Co-founder Dario Amodei’s drive to sign up corporate users for its efficiency-enhancing tools has paid off. Based on its annual revenue run-rate of nearly $50bn, Anthropic’s proposed $1tn valuation, assuming little or no debt, would represent a multiple of 20. That’s roughly where Google started its public life in 2004, and Tesla in 2010.

Even when a company’s products are not taken off the market by political edict, profit depends on being able to extract decent prices from consumers. In the AI world, chipmaker Nvidia is one that manages this with ease, as a 75 per cent gross margin shows. AI model makers, though, may have less ability to dictate terms. OpenAI is considering drastically cutting what it charges users, The Wall Street Journal has reported, suggesting a price war could be in the offing.

Falling prices would be particularly concerning given AI giants — even the likes of Meta and Alphabet — can’t easily compress what they pay for access to chips, servers and human talent. OpenAI is in a slightly better place than Anthropic on that front, since it has spent the past couple of years furiously signing agreements to rent server space.

SpaceX — for all of its quirks — has a core business that does actually have pricing power. Its space launch and satellite communications are practically without peer; the latter makes up all of the group’s profit. The catch is that its jumbo $2.3tn valuation is largely a bet on selling yet-to-be-invented AI applications, over which it has much less price-setting clout.

Granted, Anthropic has bigger fish to fry. For a company that has been de facto blocked from selling its most powerful AI models, questions about the profitability of future revenue might seem academic. But even if Amodei manages to secure a speedy truce with the White House, Friday’s intervention is a reminder that it takes more than dazzling technology to justify a sky-high valuation.