Anthropic’s IPO is a test of whether AI can pay for itself
When Anthropic lists on Nasdaq later this month, investors will be buying more than shares in the company behind Claude. They will be placing a bet on one of the biggest unanswered questions in technology: can frontier AI make money at the scale it costs to build?
On the surface, Anthropic’s amended IPO filing answers with a resounding yes. Revenue reached US 17 billion in the third quarter alone, more than 12 times what it earned a year earlier. More than 11,900 customers now pay it US$100,000 a year, up from about 950. Few companies in history have grown this fast at this size.
Look a little closer, though, and the picture gets more complicated. Growth slowed from about 144% in the second quarter to about 47% in the third. Price cuts and discounts took about US$600 million of revenue. And after a brief spell of operating profit, Anthropic swung back to a US$567 million operating loss, largely because it paid US$5 billion for short-term computing power that costs “significantly more” than capacity bought under long-term contracts.
That last detail is the crux of the AI economy. Demand is not the problem. Anthropic says plainly that customer demand exceeds the compute it has available. The problem is that serving that demand requires enormous, upfront and largely irreversible investment. The company has committed about US$895 billion to compute and infrastructure, roughly half of it due before the end of 2030.
That figure deserves to sink in. It is larger than the GDP of most countries, committed by a company that was founded only five years ago. It reflects a belief, shared across the industry, that whoever controls the most computing power will control the future of AI.
Optimists have a strong case. Anthropic’s adjusted operating income still rose in the third quarter, to US$2.5 billion. Its losses are driven by a temporary shortage of cheap compute, not by a lack of customers, and its Anthropic Direct infrastructure programme is designed to replace expensive short-term capacity as new data centres come online. If that works, margins could widen sharply.
Sceptics have an equally strong case. Prices are falling as competition intensifies, and about a fifth of revenue comes from just two customers. The risk factors in the filing are unlike anything investors have seen in a prospectus, from the possibility that a catastrophic AI incident could exceed what insurers can cover, to the US export-control order that forced the suspension of two of its most advanced models earlier this year. A company whose products can be switched off by a government directive carries a kind of regulatory risk that few other firms face.
That is why this listing matters well beyond Wall Street. It will set a price on how much the market believes AI’s revenue can outrun its costs. A strong debut will make it easier for other AI companies, including OpenAI, which is expected to list next year, to raise the vast sums they need. A weak one could force the whole industry to slow down and prove its economics before spending more.
“While revenue might be soaring, computing costs are also rising”, Wael Makarem, financial markets strategist lead at Exness, explains. “Customer concentration is an important matter. According to Anthropic’s IPO filing, roughly 25% of Claude’s maker’s revenue comes from top five customers. This is counterbalanced by the company’s compute capacity, which is pivotal to its revenue growth and tech development in the long term. While Amazon, Google and Microsoft have given it a capacity boost, it’s not enough to sustain or provide the compute capacity Anthropic needs to serve a larger customer base. Also, most of its data centers are based in the US, where there’s a massive backlash against AI, not to mention the security issues that Anthropic’s AI agents have raised, from hacking into computer systems to blackmailing users. It would be interesting to see what the future holds for Anhropic and AI tech at large.”
For the rest of the world, Africa included, the lesson is sobering. The global AI race is being financed in hundreds of billions of dollars, while much of the continent is still trying to add data centre capacity measured in tens of megawatts. The companies building frontier models are not just selling software; they are building the infrastructure on which the next economy will run. Countries that do not own any part of that infrastructure will be customers, not participants.
Anthropic’s IPO will not settle the debate over AI’s business model. But it will be the clearest test yet. The numbers show a company with extraordinary demand and extraordinary costs, racing to make the first outgrow the second. Investors are about to decide how confident they are that it will win that race.
Go to TECHTRENDSKE.co.ke for more tech and business news from the African continent and across the world.
Follow us on WhatsApp, Telegram, Twitter, and Facebook, or subscribe to our weekly newsletter to ensure you don’t miss out on any future updates. Send tips to info@techtrendsmedia.co.ke
