SpaceX's Debut Quarter Puts a Number on Nvidia's Next Leg — and on the Capex Bill
A 92% revenue surge and an exclusive Vera Rubin commitment sent the read-through to Nvidia soaring, but $18.4 billion of quarterly capex and a policy framework that defines almost nothing are the other half of the story.
Strong top line, punished stock: capex is the swing factor
SpaceX's first public quarterly report beat expectations by nearly $1 billion, with revenue up 92% to $7.8 billion, AI-specific revenue of $2.6 billion up 213% sequentially, and $14.1 billion in contracted cloud services agreements including deals with Anthropic and Google. The shares still fell back, because companywide capex hit $18.4 billion — up from $10 billion in Q1, with AI infrastructure alone near $16 billion — and management said spending stays near those levels for two more quarters. Add the overhang of insider and early-investor share sales as post-IPO lockups expire, and the market's reaction reads as a straightforward verdict: revenue growth of this magnitude is being discounted against cash burn and supply of stock. The stock was down 22.64% over the past month before bouncing 9.43% on the report.
The Nvidia math is enormous — and that is precisely the risk
Musk committed to building SpaceX's AI platform exclusively on Nvidia's Vera Rubin architecture, said the company expects 'a very significant percentage' of Nvidia's GPUs next year, and floated 20 gigawatts of power and cooling capacity by end-2027 against 1.4 GW of nameplate compute at the end of Q2 and over 2 GW expected by year-end 2026 — conceding that with delays the number lands 'close to 15 gigawatts.' Melius analyst Ben Reitzes calculates that executing just 2–3 GW in CY27 would represent roughly $100 billion of incremental Nvidia revenue, about $2 of upside to his $13.36 EPS estimate, and holds a $400 Buy target; Jensen Huang separately projected $1 trillion in cumulative Vera Rubin revenue through 2027. Our read: the same exclusivity that makes the upside case so clean also concentrates a large slice of Nvidia's forward revenue in one customer's tentative, self-described capacity target.
Who is financing the buildout, and on what terms
The bear case circulating alongside these numbers is about balance sheets rather than demand. As laid out in one analyst commentary, Oracle Cloud Infrastructure revenue rose 46% to $9.9 billion in fiscal Q4 with remaining performance obligations of $638 billion — but total debt reached $156 billion, another $261 billion of data-center lease commitments sit off the balance sheet, free cash flow was negative $24.5 billion in fiscal 2026, and roughly half that backlog is attributed to OpenAI. CoreWeave is characterized as the same structure without a legacy business to fall back on. Treat this as one investor's argument, not reported fact — but it is the coherent counterweight to the SpaceX–Nvidia flywheel: contracted backlog is only as good as the counterparty's ability to fund it.
Washington's testing framework leaves the key terms undefined
The administration's framework for assessing AI cybersecurity risk, briefed to Anthropic, OpenAI and Google at the White House and stemming from a June executive order, sets a 30-day government review window for new models — but is voluntary, applies only to closed-source models with 'state-of-the-art' capabilities and 'national security risk,' defines neither term, excludes open models entirely, and explicitly cannot be used to restrict open models post-release. The details are not being published. For frontier labs actively seeking clarity on how to ship without triggering restrictions, this resolves little; the practical burden falls hardest on smaller closed-model providers trying to guess where the line sits.
AI's capital pull is reaching well past the megacaps
Two funding rounds this cycle show where applied AI is finding paying customers: WindBorne Systems raised a $37 million Series B co-led by Khosla Ventures and Galvanize at a $250 million valuation, running about 600 balloons across 20 launch sites into a forecasting model already bought by the U.S. National Weather Service, Air Force and Navy — with commercial expansion aimed at investment funds trading weather-driven commodity outcomes. Mitti Labs raised $9.5 million led by Aramco Ventures, with Lightspeed India, Godrej and Cisco participating, scaling its rice-paddy GeoAI platform from 8,000 farmers in 2024 to over 100,000 this season. The demand for exposure is global enough that Kenya's Nairobi Securities Exchange plans East Africa's first AI-focused ETF before year-end, shilling-denominated, referencing names like Microsoft, Anthropic and OpenAI — while its CEO says he will delay the launch if the global rally looks overdone.
Sources
- Analyst: If SpaceX Delivers Even a Quarter of Elon Musk’s Promised Compute, Nvidia Could Hit $1 Trillion in Revenue (yahoo_finance)
- Morning Bid: Outer SpaceX (yahoo_finance)
- Why Nvidia Stock Is the Big Winner From SpaceX Earnings (yahoo_finance)
- If a Bear Market Is Coming, 2 Stocks You Don't Want to Own -- and 1 You Do (yahoo_finance)
- Trump’s AI testing plan is limited and vague (the_verge)
- AI makes weather prediction better. Can WindBorne make it lucrative? (techcrunch)
- Saudi Aramco backs India’s Mitti Labs to make Asia’s rice farming more water-resilient (techcrunch)
- Kenya's stock exchange plans East Africa's first AI-focused ETF (yahoo_finance)
Not investment advice.