Google Secures 110,000 Nvidia GPUs from SpaceX in $920M/Month AI Compute Deal
The relentless demand for AI compute capacity is driving hyperscalers to secure resources through innovative, large-scale agreements, signaling continued infrastructure expansion.
The story
The scramble for AI compute capacity reached new heights this week as Google inked a deal to rent a massive cluster of Nvidia GPUs from SpaceX. Starting October 2026 and running through June 2029, Google will pay SpaceX $920 million per month for access to approximately 110,000 Nvidia GPUs, alongside CPUs and memory, housed in SpaceX's data centers. This agreement highlights how even major cloud providers, traditionally relying on their own extensive infrastructure, are now turning to external partners to meet the surging demand for AI processing power, particularly for platforms like Google's Gemini Enterprise. The deal, valued at roughly $30 billion in revenue for SpaceX, underscores a significant shift in AI infrastructure procurement, resembling strategic supply agreements more than typical cloud contracts. It reflects the current reality where the bottleneck isn't just the chips themselves, but the complete, power-intensive sites required to run them, pushing companies to acquire capacity wherever it can be found.
Silicon
RTX Spark
Maker: Nvidia
What: Superchip combining a 20-core Grace CPU and a Blackwell-architecture RTX GPU, delivering 1 petaflop of AI compute and up to 128GB unified memory.
For Whom: Windows AI PCs from OEMs like ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI for local AI agents and creative workloads.
Xeon 6+
Maker: Intel
What: Data center CPU built on Intel 18A process, offering performance density and power efficiency for cloud-native and agentic AI workloads.
For Whom: Data centers, with new rackscale AI infrastructure developed with SambaNova and Foxconn for high-performance AI inference.
Crescent Island
Maker: Intel
What: New AI inference chip designed to use cheaper LPDDR5 memory and air cooling.
For Whom: AI data centers, with sampling expected in the second half of 2026.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| AI data center campus | SoftBank | 3.1 GW by 2031 (part of a broader 5 GW target) | Hauts-de-France region, including Dunkirk, Bosquel, and Bouchain, France |
| Multi-site data center development | Edged US | $2 billion in financing, including a $1.3 billion multi-site bond, and a 200 MW campus | Atlanta, Chicago, and Council Bluffs, Iowa, U.S. |
| AI data center campus | Digi Power X and Cerebras Systems | 40 megawatts, with Phase 1 (15 MW) by mid-December 2026 | Columbiana, Alabama, U.S. |
Supply & policy signals
TSMC CEO C.C. Wei warned that AI chip supply shortages will persist for 'several years' due to explosive demand, impacting logic chips, memory, packaging, and cooling systems.
Implication: The entire AI supply chain is unprepared for current demand, suggesting prolonged constraints and potential bottlenecks across various components, not just advanced processors.
Widening shortages in DRAM and High-Bandwidth Memory (HBM) are driving up prices for consumer electronics, with Samsung reporting supply 'falls far short of customer demand.'
Implication: The reallocation of memory production towards high-throughput AI components is creating supply pressure on consumer devices, potentially leading to higher costs and longer procurement timelines across industries.
An industry coalition urged the U.S. government to address the AI-driven memory chip shortage, warning of price increases and disruptions across automotive, medical, and telecommunications sectors.
Implication: The unprecedented memory consumption by AI data centers is creating a structural shift in the memory market, affecting industries far beyond AI and potentially requiring federal intervention to ensure adequate supplies for non-AI sectors.
What we'll be watching
- Intel's next earnings update on July 23, 2026, for insights into their AI chip and data center market performance.
- The launch of new thin and light laptops and compact desktops equipped with Nvidia's RTX Spark superchip this Fall.
- Intel's Crescent Island AI inference chip beginning to sample in the second half of 2026.
- Nvidia's Vera Rubin next-generation GPU system, expected to launch in the second half of 2026.
- Continued deployment of Microsoft's Maia 200 AI accelerator in its data centers beyond Iowa, including Phoenix, Arizona.
Reporting + analyst voices: grounded via Google Search at publish time.