xAI's November GPU Target Doubles to 420,000 -- Colossus Fleet Eyes 1.44 Million by Year End
xAI is planning to activate 420,000 Nvidia GPUs in November 2026, nearly double the 220,000-unit estimate that had previously circulated for the same month. The figure, reported by The Information, reflects an accelerating buildout across xAI’s Colossus 1 and Colossus 2 data center clusters.
Cluster Composition
Elon Musk laid out the hardware breakdown himself:
Colossus 1
- 150,000 H100
- 50,000 H200
- 30,000 GB200
Colossus 2
- 110,000 GB200
- 440,000 GB300
A separate batch of 220,000 GB300 chips was expected to come online this week, with another tranche following in December. If the schedule holds, the combined fleet reaches 1.44 million GPUs by year end — up from the 1.1 million figure reported in September when the Colossus 2 GB300 expansion was first announced.
The Capital Commitment
Evercore ISI projects roughly $53 billion in total AI capital expenditure for xAI across the second half of 2026, based on a $40-per-watt build-cost model. That figure covers the full second half, not a single quarter, and reflects the cost of building out the compute capacity being activated, not just GPU line items. Even spread across six months, it underlines the financial scale of the buildout at a time when GB300 allocation is constrained across the industry.
The Execution Gap
The headline numbers carry a significant caveat. GPU activation announcements in this sector routinely slip. SpaceX’s own latest 10-Q filing claimed an additional 0.4 GW of capacity deployed — capacity that observers noted was not actually operational at the time of filing. The distinction matters because it creates a wide spread in analyst estimates: some project xAI at 2.7 GW server power by year end; critics put current operational capacity at roughly 1 GW.
Until xAI reports 420,000 GPUs as running — not installed, not planned — the November target is best read as a schedule under active execution, not a completed fact.
Context
xAI’s compute buildout is the infrastructure foundation for Grok and for compute-rental contracts the company has signed with external customers. The pace of activation directly constrains what can be trained and when. At 1.44 million GPUs, xAI would control one of the largest concentrated compute fleets on the planet, with a hardware mix spanning three Nvidia generations — H100, H200, and GB200/GB300 — that creates meaningful operational complexity in scheduling and utilization.