The Terrifying Truth About Who's Actually Losing Money

This week the financial media discovered SpaceX is losing money. They did not discover that OpenAI will lose $14 billion in 2026. They did not discover that Anthropic is being valued at $1 trillion on 40% margins with no second revenue engine. They discovered the one with a profitable satellite business, 82% of global commercial launches, and a compute infrastructure its competitors are paying $45 billion to access. That is the one they called terrifying.

The SpaceX S-1 dropped on May 20. Revenue grew 33% year over year to $18.7 billion. Expenses accelerated faster. The net loss came in at $4.9 billion. The pile-on from financial media was immediate.That read misses the point entirely. SpaceX is not one company. The S-1 breaks it into three segments with three completely different financial profiles.

Starlink generated $11.4 billion in revenue in 2025 with $4.4 billion in operating profit. It is the only profitable segment in the entire company and the cash engine subsidizing everything else. The launch business generated $4.1 billion in revenue in 2025 against a $657 million operating loss, reinvesting heavily into Starship development with no near-term path to profitability. The AI segment, covering xAI, Grok, and X, generated $3.2 billion in revenue in 2025 against $6.4 billion in operating losses, with capex accelerating to $7.7 billion in Q1 2026 alone. That is where the majority of losses live.

Now here is what the headlines missed entirely. In the S-1 is a single number that reshapes the entire investment case. Anthropic agreed to pay SpaceX $1.25 billion per month for dedicated access to the Colossus and Colossus II data centers, covering more than 200,000 NVIDIA GPUs and hundreds of megawatts of power. The contract runs through May 2029.

The total value of that single contract is $45 billion. That is $15 billion per year.

SpaceX built $18.7 billion in total annual revenue across three segments over twenty years. One compute contract with one customer adds $15 billion per year on top of that. The segment the media is calling a money pit just signed a deal worth nearly as much as the entire company's current revenue. And the customer signing it is the company the media is calling the responsible alternative to SpaceX's reckless spending. Anthropic has no chips of its own. SpaceX built the infrastructure. The critics missed the business model inside the loss.

The moat underneath all of this is the part the media coverage misses entirely. SpaceX captured 82% of commercial launch market share in 2025, completing 161 of 196 total commercial launches globally. Blue Origin, its closest competitor, has been grounded by the FAA since April 30. No Western alternative exists for heavy lift at SpaceX's cost, cadence, or reliability. That is not a competitive advantage. That is a structural lock-in that took twenty years and billions in private capital to build. The launch monopoly is the foundation that makes everything else possible, including Starlink's orbit economics, Starship's development runway, and the ability to absorb xAI's losses without external financing pressure.

Before we look at the companies getting favorable coverage, here is the full picture side by side:

_{^{*ARR reflects current monthly revenue multiplied by twelve. Full-year actual reflects what was collected across the entire year. For companies growing as fast as Anthropic and OpenAI, both numbers matter. The audited S-1 gives SpaceX a transparency advantage neither competitor can match until they file. Anthropic reports revenue on a gross basis. OpenAI's April 2026 internal memo alleged this overstates comparable revenue by approximately $8 billion. Anthropic maintains it is the principal in these transactions, which justifies gross recognition under standard accounting rules. Adjusted on a net basis the figure would be approximately $22 billion.}}

Now apply the same logic to the companies getting favorable coverage.

OpenAI spent approximately $1.69 for every dollar it earned in 2025. Full year revenue came in at an estimated $13.1 billion against approximately $22 billion in total spend. Projected losses hit $14 billion in 2026 alone. The company does not expect to reach positive cash flow until 2029 to 2030. The current private valuation sits at $852 billion. There is no Starlink. There is no launch business. There is ChatGPT, an API, and a compute dependency on Microsoft infrastructure that already costs more than it generates. When that bill comes due at scale, there is no profitable segment anywhere in the business to absorb it. That is the risk nobody is writing about.

Anthropic is the most credible pure-play AI bet of the three. The growth numbers are extraordinary. The company went from $2.2 billion in full year 2025 revenue to $30 billion in annualized run-rate by April 2026, driven almost entirely by enterprise adoption of Claude Code. Investor offers have pushed its implied valuation to $800 billion, with a Q4 IPO rumored at potentially $1 trillion.

The concern is not the growth. The concern is the structure. Gross margins have compressed to approximately 40%. Inference costs are running above internal projections. There is no second revenue engine. The entire valuation sits on one variable: the model stays ahead long enough for the revenue to compound. If model pricing compresses, if a competitor closes the capability gap, or if enterprise procurement slows, there is nothing else to fall back on. And the $15 billion per year Anthropic is paying SpaceX for compute is a real cost sitting inside that structure.

xAI inside SpaceX is losing money. That is true. But the entity it sits inside runs the only profitable satellite internet business at scale and is collecting $15 billion per year from Anthropic alone to provide the compute Anthropic cannot build fast enough itself. The critics are quoting an xAI segment loss. They are ignoring a company with a launch monopoly, a profitable satellite network, and a compute business its competitors are paying $45 billion to access.

This is what my April post argued when the Cursor acquisition was announced. The stack was the point. Compute at Colossus. Models at xAI. Distribution at X. Developer tools at Cursor. Starlink as the cash engine funding the entire buildout. The S-1 confirmed every element of that thesis in audited financials for the first time. What looked like an aggressive M&A strategy in April is now a filed prospectus with audited numbers behind it.

The market will price all three of these companies in 2026. One of them has a profitable infrastructure business, a launch monopoly, and a customer base that includes its own competitors. Two of them are burning pure capital and asking investors to believe the model trajectory holds. The one with the hedge is being called terrifying. The ones without it are being called the future.

The investors who will do best here are the ones reading segment financials, not headlines.

Bashar Aboudaoud
Managing Member, UpRound