"SpaceX Q2 Earnings Beat the Narrative: $100B Cash and $25B Compute Contracts Rewrite the AI Story"

To investors,

SpaceX beat on every number in its first quarterly earnings report as a public company. Revenue hit $7.8 billion versus $6.8 billion expected. Adjusted EBITDA - earnings before interest, taxes, depreciation, and amortization, a rough proxy for cash generation - came in at $3.5 billion versus $2.0 billion expected. Capital expenditures came in slightly under estimates at $18.4 billion. The AI segment lost $1.26 billion versus $2.4 billion expected.

And the stock fell anyway.

The narrative is simple: SpaceX is spending too much on AI, the returns are years away, and a company that lost $5 billion last year shouldn't be building data centers. The lockup expires in two days. The float is thin. The bears have a script.

The script is wrong. It reads the numbers but misses the deal flow underneath them.

SpaceX has already signed compute lease agreements with Anthropic and Google that could generate more than $25 billion annually once fully ramped. That's more than SpaceX's entire $18.7 billion in revenue for all of 2025.

The company sitting on over $100 billion in cash isn't burning through money. It's building the most scarce asset in the global economy - coherent, high-density compute capacity - and selling it at a premium before the rest of the market can catch up.

The compute contracts change everything.

In May, SpaceX agreed to give Anthropic access to roughly 325,000 Nvidia GPUs across its Colossus data centers for $1.25 billion per month. Weeks later, it struck a similar deal with Google - about 110,000 GPUs for $920 million per month.

Combined, that's $2.17 billion per month. Annualized, it's $26 billion per year.

Neither contract is yet reflected in Q2 revenue. Both are still ramping. But they represent the most important data point in the earnings report - more important than the revenue beat, more important than the AI loss, more important than the stock price.

They prove SpaceX is no longer just an AI model developer. It is a hyperscale compute landlord. And it's one that doesn't need to ask customers to wait in line - customers are paying billions to get in now.

Demand for AI compute continues to outpace supply. Building a data center at this scale takes years, permits, power access, and massive capital. SpaceX built its first Colossus cluster in 122 days by converting an existing factory. The Memphis-area complex now runs roughly two million square feet and one gigawatt of compute power, with plans to reach one million GPUs.

Other companies are trying to follow. Meta is reportedly in talks to lease computing power to Anthropic in a deal that could be worth up to $10 billion over two years. But Meta's data centers weren't purpose-built for this. SpaceX's were.

Starlink is the engine.

Here's the financial architecture that the bears are missing.

The connectivity business generated adjusted EBITDA of $2.6 billion for the quarter, beating the $2.4 billion estimate. Starlink revenue in 2025 was $11.4 billion.

Starlink is the cash engine. SpaceX uses its profits to fund AI infrastructure and Starship development. It's not a subsidy. It's a reinvestment loop.

$18.4 billion in capex this quarter sounds enormous. But $15.8 billion of that - 86% - went to AI infrastructure. The rest covered Starlink satellites and launch operations. If you annualize that AI spending, it's roughly $63 billion per year. That's the build cost of one of the largest compute infrastructures on Earth.

Once the Anthropic and Google contracts are fully live and the Pentagon deal - which SpaceX is reportedly negotiating - comes through, the revenue math flips. $25 billion plus in annual compute revenue against a company that already does $18.7 billion in total revenue means the AI business alone could represent a 100%+ expansion of the current top line.

That's not an expense problem. That's an infrastructure arbitrage.

The counterargument is understandable but misfires.

Critics argue that $63 billion in annualized AI capex is unsustainable. Will Rhind, founder and CEO of GraniteShares, put it bluntly: "Starlink is executing beautifully, but it cannot single-handedly fund a $30 billion annualized AI capex program."

The concern is valid on its face. Most companies can't fund $30 billion in annual capex from one cash-generating division without borrowing.

But SpaceX is not most companies. And it has three sources of funding that critics don't account for.

First, the cash position. SpaceX reported more than $100 billion in cash following its notes offering in late June. That's not a runway - it's a war chest. Even if AI capex runs at $63 billion annualized for the next two years, the company has the liquidity to fund the buildout without forcing equity dilution or taking on unsustainable debt.

Second, the compute contracts themselves are becoming self-funding. Every month Anthropic and Google pay SpaceX, that cash flows back into building more capacity. The AI segment lost $6.4 billion on an operating basis in 2025, but it also generated $3.2 billion in revenue. That loss was a transition cost - building the infrastructure before the tenants move in.

Third, the IPO raised more than $10 billion in primary proceeds. Combined with the notes offering, SpaceX just pulled tens of billions from public markets to de-risk the capex timeline.

Bill Birmingham, managing director at REX Financial, said the SpaceX AI business is "not going to be a free cash flow story now, soon, or maybe even ever."

That may be true for the AI segment in isolation. But a company that generates $25 billion+ in new annual revenue from compute leases, sits on $100 billion in cash, and owns the launch infrastructure to deploy orbital data centers is playing a different game than a pure software company burning cash on GPUs.

What to watch next.

The lockup expires August 6. Roughly 20% of shares - triple the current tradable float - will be eligible to sell. The stock has fallen sharply from its post-IPO high, and selling pressure is real.

Short-term supply dynamics are separate from the long-term thesis. The lockup is a mechanical event. It doesn't change the compute contracts, the Starlink cash flow, or the $100 billion cash position.

What matters after the dust settles is three things:

1. Whether the Anthropic and Google contracts ramp as expected and when they hit full utilization.
2. Whether the Pentagon deal for AI data center capacity closes and at what scale.
3. Whether Starlink can continue growing subscribers while expanding into enterprise and government markets that carry higher margins.

If those three questions trend positive, the AI spending stops looking like a liability and starts looking like the most important infrastructure bet in technology.

The bears are looking at the capex line and calling it a burn rate. The data says it's a buildout.

Pick your poison.