2026-06-08 — views · Alphabet Inc. · AI cloud compute / GPU capacity · MC ~$4.5T

Alphabet is renting 110,000 GPUs from SpaceX for $920M/month — the $30B compute deal that re-rated GOOGL and exposed the supply crunch

An SEC filing for SpaceX's IPO revealed Google will pay $920M/month (>$30B) for ~110,000 Nvidia GPUs through 2029. Piper Sandler and UBS lifted GOOGL targets — but trimmed EPS. The bottleneck is supply, not demand.

What shipped

On June 5, 2026, buried in the free-writing prospectus (FWP) for what is set to be the largest IPO in history, SpaceX disclosed a compute-supply contract that tells you more about the AI equity trade than any earnings call this quarter. The counterparty is Alphabet. Google has committed to pay SpaceX $920 million per month from October 2026 through June 2029 — roughly 33 months — for access to approximately 110,000 Nvidia GPUs plus associated CPUs, memory, and infrastructure housed in SpaceX’s Memphis data-center footprint. At the full run-rate that contract is worth more than $30 billion.

The reason this is a stocks story and not a space story: a hyperscaler with one of the deepest balance sheets on earth is renting GPUs from a rocket company to bridge a capacity gap. The filing frames the deal as covering Gemini Enterprise demand that had “exceeded near-term supply.” Read that twice. Alphabet guided 2026 capex to $180–190 billion, and it still cannot self-provision fast enough.

The terms that matter

The structure is unusually lender-friendly to Google, which is the tell.

Term	Detail
Monthly fee	$920M (full run-rate from Oct 2026)
Term	Oct 2026 → Jun 2029 (~33 months)
Total value	>$30B
Hardware	~110,000 Nvidia GPUs + CPUs/memory
Location	SpaceX Memphis data center
Ramp	Reduced fees through September during buildout
Delivery gate	If full GPU capacity slips past Sept 30, Google gets a one-month grace period, then may terminate outright or accept delivered capacity at a pro-rata fee cut

That delivery gate is the part the market under-weights. Google did not sign a take-or-pay; it signed an option on capacity with a hard performance milestone. The economic risk of GPUs arriving late sits with SpaceX, not the customer. When demand is this tight, the customer still gets to dictate downside protection — which tells you the bottleneck is supply, not willingness to pay.

Why analysts moved on Alphabet, not SpaceX

The equity that re-rated on this news was GOOGL, trading near $376 when the deal surfaced. Two desks moved in the same week, both citing the cloud/compute crunch rather than the headline contract:

Piper Sandler (Thomas Champion) raised its target from $425 to $445 on June 1, lifting 2027 search revenue estimates 5% on AI Overview citation-share work.
UBS (Stephen Ju) rebuilt its Google Cloud model on June 2, raising 2026 cloud revenue +24% and 2027 +34% — putting his 2027 number 16% above Street consensus — while trimming EPS ~1% (2026) and ~7% (2027) to reflect the heavier rent-and-build mix.

That EPS trim is the honest part. Renting another company’s GPUs at $920M/month is margin-dilutive versus owning silicon outright; UBS is paying for revenue visibility (Cloud backlog cleared $460 billion in Q1, with cloud revenue up 63% YoY at a 33% operating margin) with near-term earnings. The consensus target range now spans roughly $413–$493.

The SpaceX side, and the comparable

SpaceX prices its IPO at $135/share, ~555.6 million shares, an ~$75 billion raise at a $1.77 trillion valuation, listing on Nasdaq as SPCX on June 12, 2026. The Google contract is one of two anchor compute deals underwriting that number. The other: Anthropic at $1.25 billion per month through May 2029 for the xAI Colossus 1 facility in Memphis (>220,000 Nvidia GPUs, ~300 MW), worth ~$45 billion. Combined contracted compute revenue is ~$75 billion through 2029 — a recurring, infrastructure-style cash flow stapled onto a launch business.

Practitioner note

I would not chase GOOGL on the contract headline; the re-rating already happened and the cleaner signal is the EPS cut sitting underneath the revenue upgrade. What I would actually do is treat this filing as a dated, public data point in the GPU-supply thesis and act on the second-order names: the supply gap is bullish for Nvidia unit demand and for anyone selling power, cooling, and interconnect into Memphis-scale buildouts. On SPCX itself I would not buy the IPO print — a $1.77T valuation where two compute contracts are doing this much narrative work needs at least one post-IPO quarter to see whether that $75B of contracted revenue converts at the implied margin, and the Sept 30 delivery gate is a live headline risk between now and then.

Under-considered angle

Everyone is reading this as “AI demand is insatiable.” The sharper read is counterparty inversion: a $2-trillion-plus hyperscaler is now a tenant of a private rocket company for frontier compute, and it negotiated termination rights rather than ownership. If the most capital-rich buyers in the world are choosing to rent GPUs with an escape hatch instead of building, that is a quiet vote that they expect either supply to loosen or the technology to depreciate fast — neither of which is priced into the “buy every GPU-adjacent name” reflex. The contract that looks like a demand flex is also, structurally, Alphabet hedging its own conviction.