Arizona: $265 billion and 12 facilities
Let me write down the number CFO Wendell Huang gave CNBC: another $100 billion, bringing the total commitment to $265 billion. The first fab is operating and yielding 'as good as' the flagship in Taiwan; the second is ready to move in equipment, a third is under construction, and a fourth plus the site's first advanced-packaging facility are in prep.[1]
These aren't bubble numbers; it's poured concrete and installed equipment. But Huang himself names the constraint: a shortage of construction workers and infrastructure limits. The bill has to be read in labor-hours too, not just dollars.[1]
Helios: the first real rack against Nvidia's 95% share
AMD's Helios packs MI455X GPUs, EPYC Venice CPUs, Pensando networking and ROCm software into one rack; Microsoft, Meta, OpenAI, Oracle and TCS are on the customer list. Meta's commitment is concrete: up to 6 gigawatts of AMD GPUs over time, with 1 gigawatt on Helios racks.[2]
Nvidia still holds more than 95% of the data-center GPU market; Futurum Group's 20-25% share estimate may be optimistic, but shipping genuinely starts in the second half of 2026. We'll keep reading the bill in megawatts and ship dates.[2]
Frozen v2: 6-10x per watt, but still on paper
Google's chip, codenamed 'Frozen v2,' rests on The Information's anonymous sources; Google neither confirmed nor denied it. There's a 2028 target and a claimed 6-10x efficiency gain per watt — which is a reminder, once again, that there is no cloud, only machines wired to somebody's transformer.[3]
TSMC gave poured concrete, AMD gave a ship date; Google, for now, has given only a leak and a 3% stock pop. I don't want to put all three in the same sentence — one is a construction site, one is a rack, one is still a slide. Next quarter will show which of them turns into actual shipments.[1], [2], [3]