- Essays··6 min read
The Interconnect Nobody Can Buy
UALink 2.0 was ratified in April 2026; no production-ready silicon exists. For procurement decisions targeting Q1 2027 delivery, that leaves NVIDIA NVLink over InfiniBand or RoCE on Ethernet — each with known performance trade-offs, HBM supply constraints, and operational costs that never appear on the CapEx comparison slide. The interconnect you want and the silicon you can order are not the same thing.
ai-infrastructureinterconnectnvidiaualinkRead - Essays··12 min read
Sunday Essay · The Substrate Bill Comes Due
For a decade the AI conversation was about the model; the substrate was a logistics problem handed to the cloud team. The last six weeks changed that. A rack that draws a quarter-megawatt, memory sold out before the GPU ships, grid connections you cannot procure inside an investment horizon, and an interconnect that quietly became a vendor product again. The bill has come due.
ai-infrastructuredata-centersnvidiapower-and-coolingRead - Essays··7 min read
field notes: the data centre learns to flinch
The demos are real and the peer review is real. What's unproven is whether a utility planner will write a 500 MW flexible load into a resource-adequacy filing and sleep at night. The tariffs will decide this before the silicon does.
energy-aidata-centresgrid-flexibilitydemand-responseRead