NVIDIA Now Sells the Number That Decides If Your AI Factory Pays
AI factory yield is the number that decides whether a GPU cluster earns. NVIDIA Minerva is now the instrument that produces it.
NVIDIA Is Prepaying for Capacity That Ships in 2028
NVIDIA paid $1.1bn up front and a multibillion advance to expand NVIDIA supply chain capacity. Substrate, test and optics output lands in 2027 and 2028.
Broadcom Signed a Decade of VMware. The Silicon Under It Does Not Wait.
Broadcom signed a 10 year VMware Cloud Foundation agreement with NTT DOCOMO, its first of that length in Japan. No contract value was disclosed.
NVIDIA Cut AI Factory Build Time to 14 Days. The Power Queue Did Not Move.
NVIDIA DSX POD ships a validated Vera Rubin AI factory in containers and claims 14 days to deploy. The grid still sets the real date.
One GPU in Every Nine Sits Spare in Meta’s New Rack
The Meta Prometheus supercluster runs 64 of every 72 GPUs and parks 8 as spares. That rack geometry adds 13 percent to the capacity you buy.
Nscale Booked $103 Billion of Compute and Runs 25,000 GPUs
The Nscale IPO wants $30bn on $103bn of contracts. At the last filing, 5.4 percent of its contracted GPUs were running.
Amazon Signed 960 MW of Reactors Where the Grid Was Already Built
Twelve small modular reactor units on a retired coal site say more about grid queues than about reactors. Amazon funds it and would buy the power.
Azure VMware Solution’s License-Included Path Ends in 12 Days
Azure VMware Solution licensing shifts on 31 October. PayGo SKUs with VCF included retire, and six months of migration work will not fit in 12 days.
Microsoft Walked From 23,000 GPUs Because the Grid Says 2030s
Nscale's Essex site lost its anchor tenant over AI data center power that arrives in the 2030s. Contracted capacity is not billable capacity.
A 501B Open Model Fits on Four GPUs. The Context Window Still Sets Your Bill
Reflection's Beam shows the shrink is easy and the KV cache memory is the hard part. About 19 long contexts per node, not two hundred.