AI

SpaceX Lands $26B in Annual AI Compute Deals with Anthropic, Google

The rocket company is monetizing its Colossus data center infrastructure while its own AI model lags behind competitors.

Omega Editorial· July 19, 2026· 3 min read

SpaceX has secured two major agreements to rent computing capacity from its Colossus data center complex, contracts that together could generate approximately $26 billion in annual revenue—exceeding the company's total 2025 revenue.

In May, the rocket manufacturer agreed to provide Anthropic with access to roughly 325,000 Nvidia GPUs for $1.25 billion per month. Shortly after, SpaceX struck a similar deal with Google for about 110,000 GPUs at $920 million monthly, according to details first reported by Fortune.

The agreements offer immediate returns on infrastructure SpaceX has already built, helping justify the company's $1.8 trillion valuation following its record-breaking IPO last month. For a company facing pressure to demonstrate how it will achieve ambitious long-term goals—including orbital data centers and Mars colonization—the compute rental business provides a concrete near-term revenue stream.

Why it matters

SpaceX's pivot to selling AI infrastructure reveals how companies with existing data center assets can capitalize on the severe shortage of computing power for training large language models. The deals demonstrate that SpaceX can generate substantial income from resources it already owns while pursuing longer-term ventures, addressing investor concerns about when moonshot projects will deliver returns. However, 90-day cancellation clauses in both contracts mean this revenue could prove temporary if cheaper alternatives emerge or if SpaceX needs the capacity for its own AI development.

The AI segment's mixed performance

Despite generating $18.7 billion in revenue last year—with $11.4 billion from Starlink connectivity and $4.1 billion from launches—SpaceX's AI division tells a more complicated story. The segment contributed $3.2 billion in revenue but recorded a $6.4 billion operating loss.

Part of the challenge stems from Grok, SpaceX's large language model acquired through the February purchase of xAI. Training and operating Grok requires massive investments in GPUs, electricity, and infrastructure, yet the model has lagged behind competitors from Anthropic and OpenAI in both capabilities and revenue generation.

Built-in flexibility and risk

The compute rental model gives SpaceX options. The company can lease capacity to external customers, redirect it to train newer Grok versions, or allocate it to Starlink's internal operations. But this flexibility cuts both ways, according to Sridhar Tayur, a professor of operations management at Carnegie Mellon University.

Both contracts include 90-day cancellation provisions. While this allows SpaceX to reclaim computing power if Grok's needs increase, it also permits customers to exit if cheaper capacity becomes available elsewhere.

Growing market interest

Other tech companies are exploring similar strategies. Meta is reportedly negotiating with Anthropic on a deal potentially worth up to $10 billion over two years to lease its own computing infrastructure.

SpaceX is also in discussions to provide the Pentagon with data center capacity worth billions for running AI models, The Wall Street Journal reported. If finalized, the Pentagon deal would strengthen the case that compute rental represents a sustainable business line rather than a temporary opportunity.

The foundation for this business traces back to 2024, when xAI brought its first Colossus cluster online in 122 days by converting an existing factory. The company has since expanded its Memphis-area complex to roughly 2 million square feet across Colossus and Colossus II, providing about 1 gigawatt of compute power with plans to eventually house 1 million GPUs.

Sean Cray, a senior analyst covering telecom, media and technology at Moody's, noted that demand for AI compute continues to outpace supply as companies build increasingly powerful models. Building large data centers typically requires years of development and substantial investment, making SpaceX's existing capacity attractive to customers needing immediate access—even at premium prices.

These details were first reported by Fortune.

#spacex#ai infrastructure#gpu rental#data centers#anthropic#enterprise ai

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in AI

AI· 3 min read

TSMC commits $100B more to Arizona amid AI chip demand surge

The chipmaker will expand its U.S. footprint to 12 fabs despite construction workforce shortages, CFO says.

Via AI Watch · Jul 20, 2026
AI· 3 min read

TSMC commits $100B more to Arizona, raising total to $265B

The world's largest contract chipmaker is accelerating U.S. capacity amid what CFO Wendell Huang calls a multi-year AI demand megatrend.

Via AI Watch · Jul 19, 2026
AI· 3 min read

Five AI Writing Detectors Tested: Most Work, Two Fail Badly

A hands-on evaluation reveals which tools reliably distinguish human from machine-generated text—and which ones can't tell the difference.

Via AI Watch · Jul 19, 2026