OpenAI has committed more than $1.4 trillion to building out its data center infrastructure in the next 8 years. It’s projected to lose $74 billion in 2028 alone, according to a new report from the Wall Street Journal citing internal documents from the company. On the opposite end of the spectrum, AI startup Anthropic is chugging along with a fraction of the hype that Sam Altman’s company enjoys, and it will reportedly manage to break even that same year.
The documents seem to suggest very different paths to profitability for the companies. Anthropic, which most recently raised money from investors at a nearly $200 billion valuation, has quietly been building up a subscriber base of corporate customers. According to a report from WSJ earlier this month, about 80% of Anthropic’s revenue comes from business clients, of which it has over 300,000 customers. OpenAI claims to have about one million enterprise subscribers and more than seven million total ChatGPT for Work “seats” (a figure it happened to share just days after Anthropic’s numbers were made public).
But, according to the Journal’s reporting, OpenAI’s margins on those subscriptions are much smaller than those of Anthropic. One would think, given that, OpenAI might be spending its funding frugally until it figures out a clear path to profitability. Instead, the company is going all out on building out data centers and scooping up chips to train and power its models.
It’s also straight-up burning cash with the strategy of just getting people hooked on its ecosystem in the hopes that they can eventually be monetized. Earlier this week, Forbes reported that Sora 2, the company’s video generation model that blew up with millions of people creating 10-second clips loaded with copyright infringement cases in waiting, costs the company about $15 million per day, which works out to an annualized rate of $5 billion.
It seems like OpenAI’s approach is something closer to the ZIRP-era of Big Tech, where firms were happy to run deep into the red because they were getting money for nothing with zero interest. That led to the Uber-like approach of driving prices down to unrealistically low levels to capture as much of an audience as possible before eventually flipping the switch once you’re the only game in town.
It’s not clear that the path is really there for OpenAI, but it sure seems to think it’s viable. Per WSJ, the company’s own internal documents show a rapidly growing revenue base that will lead to the company turning a profit by 2030, which would still be more than halfway into its eight-year commitment to build out expensive data center operations. On its way to finally eking into the black on its bottom line, it’ll reportedly burn about 14 times as much money as Anthropic.
The problem for Anthropic and any other company operating in the AI space is that OpenAI just might take everyone down with it. The company’s financial commitments tie it tightly to lots of players that are key to the AI sector, and its failure would result in a pretty major domino effect. It’s no wonder OpenAI is out lobbying for government guarantees on its spending.
