2026-09-25 · Fuente：Tom's Hardware
Elon Musk says Colossus 2 takes 220,000 GB300 GPUs next week with two equal tranches to follow, pushing the site past one million GPUs as a 1.2-gigawatt plant is built.
Elon Musk says Colossus 2 will receive 220,000 GB300 GPUs by next week, with another two tranches of the same amount expected to arrive later this year. Taken together, the three deliveries would put roughly 660,000 new AI GPUs into the site over the course of the year, according to the schedule laid out by Musk.
Once the incoming hardware is racked and running, Colossus 2 will host more than a million AI GPUs — a milestone the billionaire set as a target two years ago and which the site is now positioned to clear. With systems already in operation factored in, the total fleet across Musk's AI compute operation is nearing 1.44 million GPUs.
Powering hardware at that density is a project in its own right. The firm is building a 1.2-gigawatt power plant to bring the systems fully online — a clear signal that electricity generation is being treated as core infrastructure alongside the GPU fleet itself. At gigawatt scale, energy supply sits on the critical path for how fast installed accelerators can actually be put to work.
For the accelerator supply chain, the disclosure offers a concrete demand data point: a single site absorbing 220,000 GPUs per tranche on a recurring schedule through the remainder of the year. Commitments of that size, made publicly and on a dated timeline, give chip, networking and power vendors an unusually clear read on where one of the industry's largest compute buildouts is heading.
Whether the remaining tranches land on schedule will determine how soon the million-GPU threshold is formally crossed. Either way, the trajectory — 1.44 million GPUs in operation, backed by a dedicated 1.2-gigawatt plant — now defines the reference scale for frontier AI sites.
Etiquetas: Colossus 2 · SpaceXAI · NVIDIA GB300 · Elon Musk · Data Center · AI GPUs
萬安算交所 AIXX · AIXX · Exchange verificado · Ir al mercado