OpenAI Fell Short of Its Own Targets as Compute Costs Piled Up: Report

OpenAI is facing a reckoning over the gap between its ambitions and its finances, experts told Decrypt, after the Wall Street Journal reported Monday that the company missed key internal targets for ChatGPT users and revenue while CFO Sarah Friar privately warned that ballooning compute costs could outpace the money coming in.

Friar raised the alarm after the company fell short of its goal of reaching one billion weekly active users for ChatGPT by the end of last year, a milestone it never hit and never announced, unsettling some investors, the WSJ reported.

“When the dust settles, I think companies will find out something they already knew—a lot of the work still depends on human judgment, collaboration, and contextual understanding that AI can’t yet replicate,” Alice Li, Investment Partner at Foresight Ventures, told Decrypt.

Li sees the current pressure as an internal rebalancing within the tech sector, not a leading indicator of a broader macro downturn.

OpenAI’s spending

OpenAI has locked in roughly $600 billion in future data-center spending, accumulated through years of aggressive dealmaking under Altman’s thesis that compute scarcity was the true constraint on AI growth.

Friar told other company leaders she is worried that revenue may not grow fast enough to cover those contracts, the report said.

Board directors have reportedly grown more probing about the data-center deals and have questioned why Altman continues to pursue even more computing capacity despite the slowdown.

Anthropic has quietly overtaken OpenAI on share trading platform Forge Global, where it now trades at roughly $1 trillion against OpenAI’s approximately $880 billion, according to Forge CEO Kelly Rodriques, the first time its rival has commanded a higher implied valuation.

Markus Levin, co-founder of DePIN network XYO, told Decrypt that reading a market crash into these numbers misreads the underlying data.

He noted that by the end of 2025, roughly 84% of the world’s working-age population had still not used generative AI tools, and only around 44.8 million people held paying AI subscriptions globally.

“Conflating a slow, uneven adoption curve with an imminent market reckoning reflects a tunnel vision the data is pushing back against,” Levin said.

The disruption, he pointed out, is real but narrowly concentrated, driven more by tech-sector over-hiring cycles and cost corrections than by automation sweeping through the broader economy.

“A rational repricing phase is almost inevitable—market sentiment tends to move ahead of fundamentals, and expectations need to be recalibrated,” Li said.

She frames current valuations as priced ahead of time rather than fundamentally broken, with fundamentals likely to catch up if capability development stays on track.

Decrypt has reached out to OpenAI for comment.

Pavel Bezhin, CFO at AI development company Napoleon IT, told Decrypt the pattern is familiar from prior technology cycles, and the outcome is not predetermined.

“In human history, such breakthroughs have indeed often preceded crises and recessions, but they have never been their direct cause,” he said.

Bezhin pointed to the dot-com crash as the relevant lesson: economic systems built on outdated models fail to adapt, and it is that failure, not the technology itself, that triggers collapse.

“If global financial institutions have learned the right lessons from the dot-com crash, discussions about recession and systemic collapse will remain nothing more than cautionary tales,” he added.

OpenAI’s IPO ambitions are caught in the middle of all this, with Altman pushing for a public listing by year-end, while Friar has privately cautioned that the company’s internal controls are not yet built for the reporting standards public markets demand. On prediction market Myriad, owned by Decrypt‘s parent company Dastan, users place a 64% chance on Anthropic carrying out its IPO before OpenAI.

Away from the boardroom, Altman spent last week apologizing to the community of Tumbler Ridge, British Columbia, after OpenAI acknowledged it had banned a ChatGPT account tied to the suspect in a February mass shooting that killed eight people without ever notifying law enforcement.