Test report DSG-4623 · Rev F · tested October 10, 2026
AI Datacenter InfrastructureDevice under test
SpaceXAI to add 660,000 GPUs, nearing 1.44 million total
SpaceXAI will add 660,000 AI GPUs this year, nearing 1.44 million units in operation, and is building a 1.2-GW power plant to bring the fleet online.
- Read
- 2 min
- Words
- 332
- Node
- 20nm
- Operator
- Priya Raman
Spec summary
- SpaceXAI plans to add 660,000 AI GPUs this year.
- Total fleet will approach 1.44 million GPUs in operation.
- The company is building a 1.2-gigawatt power plant to bring the systems fully online.

Elon Musk's SpaceXAI plans to add another 660,000 AI GPUs this year, bringing its installed base close to 1.44 million units in operation, Tom's Hardware reports.
The deployment schedule depends on power. The company is constructing a 1.2-gigawatt power plant to bring the systems fully online.
What does the GPU expansion mean in practice?
A fleet approaching 1.44 million GPUs places SpaceXAI among the largest single-operator AI compute installations in the industry. The additional 660,000 units scheduled for this year represent an increase of roughly 84% over the 780,000 GPUs implied as the current baseline.
At that scale, several constraints dominate delivery:
- Power capacity. Feeding more than 1.4 million GPUs requires dedicated generation on the order of gigawatts, which explains the 1.2-GW plant now under construction.
- Deployment sequencing. Systems can be physically installed before the power infrastructure reaches full output, so the "in operation" count will lag the shipped count.
- Supply. An order of 660,000 units in a single year competes directly with allocations from other hyperscalers.
Why build a dedicated 1.2-gigawatt plant?
Grid interconnection queues and utility contracts are now a limiting factor for AI data center buildouts. By building its own generation capacity, SpaceXAI decouples GPU installation from external power availability.
The 1.2-GW figure sets the scale: a plant of that size can power a GPU estate well beyond one million units, accounting for cooling, networking and overhead in addition to accelerator draw.
What happens next?
The firm expects to bring the full complement of nearly 1.44 million GPUs into operation once the power plant comes online. The timeline for the plant's completion will determine when the 660,000 new units transition from installed to fully operational status.
The expansion marks one of the largest announced accelerator deployments of the year and signals that Musk's AI venture is scaling compute at a pace comparable to the largest cloud providers.
via Google News: GPU datacenter (Source)
More from Priya Raman
Same lot · LOT-C1C6
- DSG-95397nmNVIDIA and SB Energy Plan 10-GW AI Data Center Complex in Ohio
- DSG-214128nmCerebras to supply 100 MW of AI chips to cloud startup Gimlet Labs
- DSG-503714nmEmerald AI Draws Google, Nvidia, Anthropic for Grid-Flexibility Work
- DSG-944345nmVolantis Raises $88 Million for Laser-Linked AI Memory Chips
- DSG-39717nmNvidia Confirms 4.25GW OpenAI Commitment at Ports-Pike