Footnote MI350-81 dates the benchmark to 7 July 2026 and names the comparison: ROCm 7.0 against a preview build of ROCm.ai, averaged over three models on the same MI355X box. Nvidia is not in the comparison at all.
2d ago
315 PFLOPS, 128 GB of SRAM, 40 PB/s. Every one of those numbers is a rack of 256 chips, and the chip is somebody else's.
4d ago
A company that has raised over a billion dollars does not have a Series A. The label, the halved valuation and the identity of the lead investor all point at the same thing.
Aug 18, 2026
Taalas hard-wires one model per chip and changes two mask layers to swap it. AMD disclosed no price, and the deal is still waiting on regulators.
Aug 9, 2026
Korea's memory giants are commercializing Compute Express Link modules — a layer between DRAM and storage — that they say can hold huge AI models close to GPUs at near-DRAM speed and lift inference throughput.
Jul 20, 2026
The Paris startup, backed by Yann LeCun, wants to break Nvidia's software lock-in with an inference server that runs open-weight models across five chip families from one stack — free, though its alpha is narrower than the pitch.
Jul 9, 2026
