Compute Atlas

Large-scale compute, told as a supply chain.

Rankings tell you how fast a machine is. They do not tell you who built it, whose fabric it runs on, which of those vendors still exists under its own name, or which listed company carries the exposure. Compute Atlas is the layer underneath: a parts graph with systems, sites and companies as first-class nodes, every fact carrying its own source.

What this answers that a ranking cannot

A ranked list is system-centric. You can read one machine's spec string, but you cannot ask it a question about companies, and the spec string is where the interesting structure is buried. Four examples, all answerable here and none answerable there:

Which vendors appear in the largest systems, weighted properly. Machine-count share is the number everyone quotes and it systematically understates whoever supplies the biggest machines. Weighting by installed FLOPS changes the ordering. See it →

Who a 2012 supplier is today. Cray, SGI, Bull, Mellanox and Thinking Machines all appear in this dataset. Four of them fold into companies you can buy today; one does not fold anywhere, because it went bankrupt in 1994. Nobody has published that normalization, and every long-run chart that ignores it is wrong. See HPE's inherited history →

Where the domestic-substitution line actually moved. Tianhe-1A was American compute on a Chinese fabric in 2010. Sunway TaihuLight was fully domestic silicon in 2016. Tianhe-2 was rebuilt with domestic accelerators after an export control blocked the upgrade it had planned. That is a supply-chain story that the spec strings contain and never state. See it →

What is missing entirely. The largest AI clusters built in the last three years mostly never submitted HPL results, so they are absent from every public ranking. We carry ranked and unranked systems in one taxonomy with explicit confidence tiers, rather than pretending the ranked world is the whole world. See the one calibration point we have →

Interconnect share by installed FLOPS

Rolled up to today's corporate parent, so Mellanox counts toward NVIDIA and Cray's fabrics count toward HPE.