AI: Trying to Make the Numbers Add Up
SpaceX valuations and OpenAI's leaked earnings reports makes for hard sums
My wife is a maths teacher and she tells me that some equations have no solution.
Looking at the numbers behind SpaceX and OpenAI, I am starting to think investors may have found a few of their own. Ahead of SpaceX’s IPO last week, I struggled to make the valuation stack up. Then OpenAI’s 2025 financial accounts were leaked - and somehow the numbers look even harder to explain.
But sometimes you’ve got to sit down with a calculator and give it a go anyway, so lets try and make the numbers make sense.
SpaceX: The Moonshot Premium
Last week saw the SpaceX IPO blast off in financial markets. Shares initially soared from an IPO price of $135 to almost $220 before gravity began to take hold, with reports emerging that the stock had already attracted significant short-selling interest.
The challenge for investors is that SpaceX’s valuation appears to depend heavily on the success of several ambitious “moonshot” projects. These range from expanding Starlink’s global communications network to more speculative ventures such as space-based data centres and, ultimately, reusable missions to Mars. While these opportunities may be enormous, they remain inherently uncertain.
This creates a valuation problem. Morningstar estimated prior to the IPO that more than $1 trillion of implied company value could disappear if some of the firm’s most ambitious projects fail to materialise. In other words, a substantial proportion of today’s valuation reflects profits that investors expect SpaceX to generate many years into the future rather than cash flows being earned today.
The chart below illustrates the challenge. Even under aggressive assumptions of 50-100% annual earnings growth, SpaceX continues to trade at valuation multiples well above those of the broader technology sector. Investors are therefore not simply betting on strong growth. They are betting that SpaceX successfully delivers on some of the most ambitious commercial projects ever attempted - and, as a reminder, the Starship rocket launches (one of the most important divisions in SpaceX) currently have a 50% success rate on their launches.
OpenAI: Where Did the Losses Go?
To make matters worse, OpenAI’s 2025 financial statements were leaked this week. Technology commentator Ed Zitron highlighted that the company spent approximately $34 billion on research, development and sales initiatives (see the breakdown below) during the year while generating just $13 billion in revenue. The result was an operating loss of more than $21 billion in a single year - but there are more losses.
Zitron highlights that when accounting for interest expense and the costs associated with changing the corporate structure, OpenAI reported a consolidated net loss of more than $60 billion for 2025. That figure is so large that it exceeds the annual GDP of many sovereign nations like Paraguay or Tunisia!
Yet the final loss attributable to OpenAI was reported at a much lower $38.5 billion. More than $21 billion of losses were allocated to noncontrolling interests within the firm’s complex ownership structure.
Put differently, that means that means that OpenAI is effectively transferring a large portion of these losses to a separate controlling entity - but who that economic owner is and how they might bare those losses remains a concerning mystery.
The Pathway to Profitability
Tech firms often make losses. Airbnb lost around $5.6 billion between 2018 and 2021 before becoming consistently profitable. Uber accumulated roughly $32 billion of losses between 2014 and 2022 before eventually reaching profitability.
The issue is not that technology firms lose money. It is the scale of the losses now being tolerated by investors. OpenAI lost more money in a single year than Uber lost in eight years. Similarly, SpaceX’s high risk ventures into space and its desire to play catch-up in AI could also see the firm hemorrhage cash for a long period of time.
Yet despite these mounting losses, valuations continue to climb. Investors are being asked to assign ever-higher values to businesses generating ever-larger deficits in the hope that future profits will eventually justify today’s spending. That may prove correct. But the gap between present-day economics and future expectations is becoming increasingly difficult to ignore.
Disclaimer: The information provided in this blog post is for general informational purposes only and should not be construed as financial, investment, or professional advice. The content is not intended as a solicitation, recommendation, endorsement, or offer to buy or sell any securities or financial instruments. Any reliance you place on such information is strictly at your own risk. Always consult with a qualified financial advisor or professional before making any investment decisions. The author and the website assume no responsibility for any losses or damages that may result from the use of or reliance upon the information provided in this blog post.
Your wife's observation is doing more work than the rest of the piece combined. Some equations have no solution. OpenAI's revenue was $13 billion. Its costs were $34 billion. The equation doesn't need the numbers to "add up" in some future quarter. It needs revenue to nearly triple before the losses even stop growing, and that's before anyone accounts for the $60 billion hole already dug. Uber took eight years to lose $32 billion and eventually found profitability by cutting costs and raising prices. OpenAI lost nearly double that in twelve months and its costs are accelerating because the compute gets more expensive with every model generation, not less.
The $21 billion in losses allocated to noncontrolling interests is the detail nobody is pressing on hard enough. That's the GDP of a sovereign nation quietly absorbed by an entity the public can't identify. Who is holding those losses? What are the terms? What happens to the parent company if that entity decides it's had enough? That's not a footnote. That's the load-bearing question in the entire capital structure, and it's sitting in a line item most readers will scroll past.
Great stuff as always!