SpaceX IPO Pre-Announcement Large Orders: Google Pays $920 Million Monthly to Rent 110k NVIDIA GPUs

SpaceX Announces a Second Major Computing Power Deal One Week Before Its IPO: Google Will Pay $920 Million Monthly Starting October 2026, Renting About 110,000 NVIDIA GPUs, with the Contract Running Until June 2029 for a Total of Approximately $30 Billion.
(Background: Anthropic Will Pay SpaceX $1.25 Billion Monthly in Exchange for xAI Excess Computing Power)
(Additional Context: Musk Announced That xAI Is No Longer an Independent Company; Renamed “SpaceXAI,” It Is Positioned as an AI Product Under SpaceX)

One Week Before SpaceX’s IPO, in the form of a regulatory filing submitted to the SEC, SpaceX disclosed the details of a transaction: Google will pay $920 million per month starting October 2026 to obtain the right to use approximately 110,000 NVIDIA GPUs, CPUs, memory, and related hardware, with the contract term running until June 2029. Based on this, the total contract value is approximately $30 billion.

Several key terms in the contract are worth noting. Google’s access to computing capacity will be “gradually activated at preferential rates before September.” If SpaceX fails to deliver the committed number of GPUs by September 30, 2026, after a one-month grace period Google may immediately terminate the contract, or it may choose to accept the actual delivered amount and adjust the monthly fee proportionally. Both parties retain a flexible exit clause: after December 31, 2026, either party may terminate the contract with 90 days’ advance notice.

The filing does not specify which data center Google is using. Musk previously said Colossus 2 would be reserved for xAI’s own use, but the content of Google’s contract differs from that, and the details remain unclear.

SpaceX’s Colossus: From xAI’s Internal Computing Power to a Global GPU Rental Platform

Colossus 1 is located in the outskirts of Memphis, Tennessee. It was originally a supercomputing cluster built by xAI for its own AI research. After xAI was incorporated into SpaceX earlier this year, Musk ended up with a batch of idle computing power—just as demand in the AI industry happened to fill the gap.

About a month ago, Anthropic was the first to sign a computing power leasing agreement with SpaceX: $1.25 billion per month in exchange for all available computing capacity of Colossus 1, estimated at about 220,000 GPUs, with the contract term also extending to 2029. Google’s deal is about half the size of Anthropic’s—110,000 cards—but the unit cost and terms are similar to those of Anthropic.

Taken together: Anthropic’s $1.25 billion + Google’s $920 million = $2.17 billion per month, which works out to approximately $26 billion annually (about $260 per billion). SpaceX’s Colossus data center is evolving from the infrastructure of an AI company into a new cloud computing provider centered on renting GPU clusters.

Google Has Computing Power—So Why Rent It?

Google’s official explanation is “temporary bridging.” In a statement, Google Cloud said this is a “short-term, immediate agreement,” aimed at “ensuring we have bridging compute capacity (temporary transitional compute: renting external resources to get through demand peaks, then exiting once our own data centers are built) to address an unexpected surge in customer demand for Gemini Enterprise.”

A data center typically takes 18 to 36 months to move from groundbreaking to production. During this window, market demand won’t wait—especially even if Google is among the world’s top builders of data centers.

The relationship between Google and SpaceX also goes far beyond this computing power contract. Google is a long-term investor in SpaceX, and after the IPO the market value of its equity stake is expected to exceed $100 billion. The two sides are also reportedly exploring the possibility of jointly building orbital data centers—one of SpaceX’s important long-term plans after the IPO.