Elon Musk says xAI's models will be as advanced as those of OpenAI and Anthropic by Christmas, and he is staking that claim on a datacentre buildout of 1.44 million GPUs across Memphis by year-end.

"We will keep accelerating," said Musk. "Our AI efforts are only 3 years old, vs 6 and 10 years old for Anthropic and OpenAI. If our second derivative remains strong, we will reach pole position in about 6 months."

The "second derivative" in Musk's phrasing means the rate of acceleration — the argument being that a younger company improving faster will overtake slower-moving incumbents.

The centerpiece is the Colossus 2 datacentre in Memphis. It currently runs on 110,000 NVIDIA GB200 GPUs and 440,000 GB300 GPUs. Musk's roadmap calls for adding 660,000 more GB300s before the end of the year, delivered in three batches of 220,000.

The first 220,000 GB300s should come online by the end of this week. A second batch of the same size is due to be operational by the end of November, with the final 220,000 arriving before year-end. That would give Colossus 2 a total of 1.2 million Blackwell-generation GPUs.

Colossus 2 sits alongside the original Colossus facility, which currently holds 230,000 GPUs: 150,000 H100s, 50,000 H200s and 30,000 GB200s. Combined, the two sites would reach roughly 1.44 million GPUs at year-end.

That fleet spans three NVIDIA generations — Hopper's H100 and H200 alongside Blackwell's GB200 and GB300 — and it is the Blackwell installation that carries the near-term scaling plan. All of the 660,000 planned additions through December are GB300s.

Musk's timeline puts model parity at the moment that final batch lands. Whether the 1.44 million-GPU installed base actually translates into frontier-level model performance by Christmas is the open question — but if the three GB300 tranches arrive on schedule, xAI will operate one of the largest single-company compute footprints in the industry, and the competitive gap it needs to close will be measured against rivals racing to scale their own clusters.