SpaceX Spends the IPO

If You Only Read One Thing

SpaceX is not mainly using AI to improve rockets; it is using launch dominance and an $85.7 billion IPO to finance an AI utility. Its first earnings report shows the inversion in hard numbers. EA's $55 billion take-private is the mirror image: one company sold equity to buy optionality, while the other added debt to harvest certainty.

AI Ate SpaceX's Capex

SpaceX spent thirteen times as much on AI equipment as it did on space systems last quarter. Capital allocation, more than branding, has turned the company into an AI infrastructure bet wrapped in a launch and satellite business.

In its first quarterly report as a public company, SpaceX disclosed $7.814 billion of revenue, up 92% from a year earlier, and a $541 million net loss. Its AI unit generated $2.561 billion of revenue but consumed $15.828 billion of capital spending, versus $1.174 billion for space systems and $1.367 billion for connectivity. Starlink and related connectivity services remained the cash engine, producing $4.291 billion of revenue and $1.656 billion of operating income.

Bloomberg covered the extraordinary AI spending. What the headline number misses is the financing mechanism. During the first half, SpaceX generated $3.466 billion of operating cash, spent $34.487 billion on investments and raised $100.291 billion from financing, principally the IPO. It ended June with $100 billion in cash and securities, so an immediate liquidity crisis is the wrong risk. Public shareholders are funding the conversion of IPO proceeds into chips and data centers long before AI customers pay for them.

Why it matters: Starlink is not yet financing the AI build; the balance sheet is. That changes the bargain investors thought they were buying. SpaceX has combined a scarce launch network, a recurring connectivity business and a speculative compute utility so that mature assets lower the cost of capital for the new one. The structure can work because scale is arriving quickly: AI revenue grew 247%, nameplate compute reached 1.4 gigawatts, and contracted cloud sales stand at $14.1 billion. But those contracts remain small beside a single half-year's investment outflow, making continued market confidence part of the operating model rather than a background condition.

Room for disagreement: The AI unit already produced $1.146 billion of adjusted earnings before interest, tax, depreciation and amortization, despite a $1.257 billion operating loss, which suggests depreciation rather than weak demand explains much of the accounting red ink. If contracted sales convert rapidly, today's excess capacity will look like the cost of entering a market before rivals can assemble comparable power and chips.

What to watch: Track the ratio of quarterly AI infrastructure revenue to AI capital spending. A rising ratio would show that customers, rather than new shareholders, are beginning to finance the buildout.

EA Mortgages Its Hits

Electronic Arts has escaped quarterly earnings pressure by accepting a more literal deadline: the payment schedule on an enormous acquisition loan. Private ownership removes one scoreboard and installs another.

EA completed its $55 billion sale to a consortium led by Saudi Arabia's Public Investment Fund on Tuesday, ending its Nasdaq listing and paying shareholders $210 a share. The original financing plan paired roughly $36 billion of equity with $20 billion of committed debt, of which $18 billion was expected to be funded at closing. EA's fiscal 2026 results showed $7.531 billion of revenue, $8.026 billion of net bookings and $2.553 billion of operating cash flow.

The arithmetic matters more than the ownership headline. In a leveraged buyout, the buyer borrows much of the purchase price and the acquired company's future cash helps repay it, much as rent services the mortgage on a property. EA's expected funded debt is about seven times its latest annual operating cash flow. The PIF can be patient about when its equity pays off; creditors still expect interest on schedule.

Why it matters: Commentary about the deal has concentrated on Saudi influence over games. The nearer and more mechanical force is debt. EA's annual sports releases, Ultimate Team purchases and established franchises become more valuable because their cash arrives predictably, while an experimental game with an uncertain launch date becomes harder to defend. Going private can protect ambitious development from public-market volatility, but the acquisition financing rewards a portfolio that behaves more like an annuity. That shifts power toward franchises with repeatable monetization and away from projects whose payoff cannot be scheduled.

Room for disagreement: EA enters private ownership from strength: bookings grew 9%, operating cash flow reached a record and the debt market was willing to finance one of history's largest buyouts. A controlling shareholder with a multidecade mandate may approve investments that public investors would punish after one weak quarter. The relevant comparison is therefore not debt versus no debt, but fixed payments plus patient ownership versus the old pressure to meet guidance every ninety days.

What to watch: Watch EA's cash interest coverage when the first bondholder disclosure appears. That single ratio will show how much operating room remains after servicing the buyout debt.

The Contrarian Take

Everyone says: Saudi ownership is the main reason EA's creative direction will change.

Here's why that's wrong (or at least incomplete): Ownership can influence content, but debt changes which projects survive before anyone debates their themes. PIF benefits if a risky new franchise becomes a cultural hit; lenders receive the same interest whether the next game is daring or derivative. With roughly $18 billion expected to be funded at close, the repeated cash flow of sports franchises and live services becomes a financial control system. The most consequential editorial voice may be the payment calendar.

Under the Radar

• Coupang's privacy failure became a margin problem: The Korean commerce company reported a $556 million operating loss, including about $410 million of administrative fines tied to last year's data breach. Excluding those fines, the loss was $146 million. The quarter shows privacy regulation crossing from a legal contingency into unit economics: the penalty was nearly three times the underlying operating loss.
• SpaceX's first market test arrives tomorrow: About 911.5 million employee and insider shares become eligible for sale on August 6, according to Axios. The expiration does not mean every holder will sell, but it expands supply just as public investors absorb the scale of AI spending. Employee liquidity and shareholder tolerance are about to meet in the same price.

Quick Takes

Washington Gives Open Models an Exit

The White House's voluntary frontier-model framework applies to closed, state-of-the-art systems with national-security risks and explicitly leaves released open models unrestricted. That is a substantive boundary, not a drafting detail: labs can avoid the government's secret pre-release access process by making weights public, although doing so sacrifices control of the model. Yesterday's opaque framework now contains a competitive design choice. (Source)

The FCC Moves Inside Servers

The FCC is preparing restrictions on Chinese-made optical transceivers, the components that turn electrical signals into light inside data-center networks. China's Innolight holds an estimated 27% of the global market, while American suppliers lack comparable scale. Telecom security policy is moving from radios and routers into the AI cluster itself, where replacing a tiny component can still delay an entire build. (Source)

Cloudflare Gives Agents Allowances

Cloudflare introduced stablecoin wallets with virtual sub-wallets, spending caps and programmable controls for software agents buying APIs or content. The important product is not the coin; it is delegated authority. If agents become customers, the company controlling their identity, budget and payment policy can sit between publishers and machine demand much as app stores sit between developers and users. (Source)

The Thread

Capital providers are becoming product governors. SpaceX's public shareholders decide how long launch and connectivity assets can subsidize compute; EA's lenders influence which creative risks fit a payment schedule. Washington draws a boundary between open and closed models, while the FCC reaches into the components data centers can buy. These are different institutions using the same instrument: control the funding, permission or bottleneck that makes a product possible, and product strategy follows without requiring day-to-day management.

Predictions

New predictions:

• I predict: By December 31, the FCC will formally name at least one Chinese optical-transceiver maker or the product category itself in a Covered List action, proposed rule or order. This prediction is wrong if the reported drafting produces no public commission action by then. (Confidence: medium; Check by: 2026-12-31)

Issue date: 2026-08-05 · Generated: 2026-08-05 03:44 AM EDT