Nvidia's Vera Rubin release is ostensibly about chips entering full production. The more durable detail is light. The company says its AI factory design includes Spectrum-X Ethernet Photonics and co-packaged optics intended to improve power efficiency and uptime inside vast multi-rack systems. [1] A syndicated version carries the same release language for the production claim. [2]

That extends the paper's earlier note that Nvidia photonics moved AI infrastructure toward uptime, and it narrows the claim made when Vera Rubin entered full production. Production is not the same as customer capacity. A five-rack POD, fall shipments and named adopters still need deployment receipts. [1]

The point is not that optics are glamorous. They are not. They are the part of the story that drags AI back from model mystique into the data center. If a million-GPU fabric depends on moving data without burning the power budget or collapsing uptime, the bottleneck is not only silicon.

X can celebrate another GPU generation. Nvidia's own release tells a more physical story. AI factories are becoming power plants with servers attached, and photonics is one of the receipts.

-- DAVID CHEN, Beijing