Elon Musk said on September 25, 2026 that Colossus 2, xAI's data center near Memphis, could more than double its current fleet of Nvidia chips before the year is out — the most detailed announcement yet about how fast that facility is scaling. Outlets including Yahoo Finance, Invezz, and BigGo Finance confirmed Musk's message the same day, at a moment when xAI is pushing hard to close the compute gap with OpenAI, Google, and Anthropic.
Colossus 2 currently runs on a mix of Nvidia GB200 and GB300 chips — around 550,000 units in total, according to industry reports — already making it one of the largest AI compute sites in the world. Musk indicated that another batch of 220,000 GB300 chips would go live in the days following the announcement, with a second batch of similar size planned for November.
According to Musk, if the planned batches land as expected — including a third batch toward December that he described as "if we're lucky" — Colossus 2 could end 2026 with more than double the Nvidia chips it has today, approaching or exceeding a million GPUs at a single site. xAI had previously set a goal of equipping its Memphis campus with roughly a million GPUs during 2026, so Musk's announcement is essentially a reaffirmation of that target with a more concrete timeline.
The race to stockpile cutting-edge Nvidia chips isn't just a corporate vanity metric: every batch of hundreds of thousands of GPUs that one lab locks up for itself is capacity that isn't available to the rest of the AI cloud market. This announcement lands amid tension already familiar to readers of this blog — the cloud GPU shortage we've been documenting in parallel, with providers like RunCloud warning of longer wait times to access high-end AI compute.
For a company in Guatemala or Latin America renting AI compute through a cloud provider, the consolidation of capacity among a handful of deep-pocketed labs (xAI, OpenAI, Google, Anthropic) has an indirect but real effect: less leading-edge GPU supply available for mid-tier providers, which pushes prices and provisioning times upward. The practical recommendation remains the same as for the GPU shortage in general: diversify compute providers, negotiate capacity commitments ahead of demand spikes rather than waiting for them, and evaluate whether your workload really needs the newest model or chip or whether a more accessible compute tier is enough.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel