The Day the AI Money Printer Bifurcated
SpaceX filed its S-1 last week with a simple message for investors: buy into the AI age. The filing — reported by Reuters — showed a company with $4.69 billion in quarterly revenue and an ambition to become the most valuable public company in American history, with a $1.75 trillion IPO price tag, per Fortune. What the filing showed next to that ambition was less flattering.
SpaceX lost $1.94 billion in operating income in the first quarter. Its satellite internet business, Starlink, generated $1.19 billion in operating profit. That profit did not just vanish into rocket development. According to the filing, xAI, the artificial intelligence company Musk acquired in February, accounted for $2.47 billion in losses on $818 million in revenue. Three dollars lost for every dollar taken in. The AI rocket company is also, by the numbers, an AI money pit.
The burn is not unusual for a young AI company. What makes it notable is the timing. SpaceX is asking public market investors to buy an AI story while the filing shows xAI is the load-bearing loss center inside that story. The $1.19 billion Starlink generated in operating profit was not enough to cover SpaceX's total operating loss, meaning the profitable division was itself consumed by the rest of the company.
The contrast with the rest of the industry is sharp. Anthropic confirmed this week it is on track to generate $10.9 billion in revenue in the second quarter, more than double its $4.8 billion in the first, and is expected to post its first operating profit of approximately $559 million in the same period. Nvidia posted $81.62 billion in revenue for the quarter, up 85 percent year over year, and raised its guidance for the next quarter to $91 billion. OpenAI won a jury verdict clearing its path to an IPO on May 18th. And OpenAI's model solved a central problem in discrete geometry — first posed by Paul Erdős in 1946 — marking the first time a major open problem in mathematics was solved by AI.
All four events happened within 48 hours. What they collectively exposed is that the AI industry's financial promises are no longer theoretical, and they are not uniformly distributed. The money printer is not running for everyone. Some companies are printing. Some are burning through what the printer generated.
xAI is burning. At a 3-to-1 loss ratio, it is absorbing Starlink's profit and then some. SpaceX is bringing that burn rate to public markets at a moment when the investment case for AI infrastructure has never been more accepted and the questions about which companies actually generate returns on that infrastructure have never been more urgent.
The catch is in the fine print of Anthropic's arrangement with SpaceX. Anthropic has agreed to pay SpaceX $125 million per month for compute access, according to Reuters, a figure that represents the largest single infrastructure cost for a frontier AI lab we have seen disclosed publicly. The agreement includes a 90-day termination clause for either party. That means Anthropic's first profitable quarter rests on a compute supplier it can walk away from in three months — and a supplier whose owner, Elon Musk, is simultaneously xAI's owner and OpenAI's active legal antagonist in a case that ended with a jury verdict on May 18th. The 90-day window is not a threat. It is an open dependency that the profit figure does not account for.
The structural tension does not end there. SpaceX's own filing — in the SEC S-1 — makes the case for a $26.5 trillion AI total addressable market, the bulk of its valuation thesis. That TAM claim is partly grounded in the premise that AI infrastructure demand is effectively unlimited. But xAI's $2.47 billion quarterly loss on $818 million in revenue is the clearest public evidence of what that demand costs to serve at current scale. For infrastructure investors and VCs trying to model unit economics across the industry, the Anthropic profit and the xAI burn are now a single data point: the industry's first confirmed profit depends on infrastructure owned by the industry's highest-profile loss-maker, and that infrastructure relationship can be dissolved on ninety days' notice.
What the 48 hours between May 18th and 20th actually showed was not a bifurcated AI industry in the abstract. It showed that the financial architecture of AI's leading companies is built on relationships that are also legal conflicts, and that the profit and loss figures carry dependencies the filings do not fully disclose.