Google-SpaceX $30B GPU lease deal

Two deals, one address. SpaceX now collects $1.25 billion a month from Anthropic and another $920 million a month from Google for compute running inside the same xAI-built Colossus campuses it acquired in an all-stock merger ^[3]. Together those leases throw off roughly $2.17 billion a month, about $26 billion a year, against SpaceX's ~$18 billion in total annual revenue ^[4]-- a side business that already dwarfs the rockets. The striking part is who the tenants are: both Google and Anthropic build models that compete with xAI's Grok, yet they are now paying rent to xAI's owner for the hardware to do it.

Across X, the dominant read crystallized into a single image -- SpaceX became the AI war's landlord, collecting rent from both armies -- with a recurring 'Elon called it' note of admiration for a rocket firm that pivoted into selling shovels mid-gold-rush. Google itself was careful to call this 'bridge capacity' for Gemini Enterprise demand 'even higher than we expected' ^[1], language that frames a $30 billion commitment as a stopgap until its own buildout catches up.

Google has backed SpaceX since 2015 and holds a reported ~6% stake that could be worth more than $100 billion after the IPO, with a Google executive sitting on SpaceX's board ^[4][5]. That ownership turns a straightforward lease into something messier: a chunk of every $920 million payment flows back toward a company Google partly owns, and the disclosure landed just days before SpaceX's June 12 IPO roadshow ^[2].

The skeptical community reaction zeroed in on exactly this. Across Reddit's investing and technology forums the prevailing take was that the timing looks engineered -- a marquee, contracted-revenue headline to dress up SpaceX's balance sheet before listing, with Google positioned to benefit from a richer valuation on its own stake. The contract's own escape hatch fed that suspicion: both sides can walk with 90 days' notice after December 31, 2026 ^[6], which readers framed as a 'tell' that this is less an ironclad multi-year bet than an option dressed up as a commitment.

The cleanest contrarian read flips the scarcity narrative on its head. SpaceX is leasing ~110,000 GPUs to Google and 220,000+ to Anthropic ^[1][7]-- hundreds of thousands of chips it apparently does not need to run Grok itself. If xAI were genuinely compute-constrained, the argument goes, it would be hoarding those clusters, not renting them out by the gigawatt. That observation surfaced repeatedly in the social conversation, with a YouTube and Reddit through-line questioning whether the industry's 'we can't get enough GPUs' framing survives contact with a vendor this eager to sublet.

The sober counterpoint, also well represented, is that the scarcity is real and these are rational moves within it: a rented GPU is a 'double win' because you get the capacity and your competitor doesn't, and short-term leasing lets an owner monetize clusters faster than waiting on its own model roadmap. Either way, the deal reframes industrial-scale GPU capacity as a tradeable asset class rather than a strategic moat.

Google appears to be paying a premium. It is committing $920 million a month for ~110,000 GPUs, while Anthropic pays $1.25 billion a month for 220,000+ ^[1][7]-- roughly 74% of Anthropic's monthly bill for about half the GPUs, implying a materially higher per-GPU rate. (Widely circulated specifics like an exact dollars-per-GPU-hour figure or a precise multiple were debated in the social conversation but not confirmed in the underlying reporting, so treat them as estimates rather than facts.)

Whatever the exact ratio, the premium plausibly reflects urgency -- bridge capacity bought against demand that outran Google's own buildout -- and a market where reporting says the cost of building AI infrastructure jumped more than 50% in two weeks, from roughly $50 billion to $75 billion per gigawatt ^[8]. For SpaceX the recurring revenue is real, but analysts caution the landlord business carries an operator's curse: heavy ongoing capex, GPU lifespans that force constant reinvestment, and commoditization that erodes pricing power ^[9]. Morningstar went further, calling the company overvalued and suggesting investors could buy cheaper after the IPO ^[4].