GPUs for AI: the 2026 lineup, explained
The whole datacenter line compared: NVIDIA H100, B200, B300, GB300 and AMD's MI355X. Memory, generation, best fit, and how to choose the right one.
Learn
Plain-English guides to the hardware, pricing, and market behind European AI compute.
NVIDIA B300: the Blackwell Ultra GPU, explained
288 GB of HBM3e, 15 PFLOPS of FP4, and the accelerator behind today's frontier clusters. Specs, B300 vs B200, what runs on one node, and how to get it in Europe.
NVIDIA B200: the Blackwell workhorse, explained
192 GB of HBM3e and native FP4, the volume Blackwell part. B200 vs H100, B200 vs B300, what fits on a node, and how to reserve it in Europe.
NVIDIA H100: the Hopper workhorse, explained
80 GB of HBM3 and the market's reference GPU. Specs, H100 vs A100, H100 vs H200, what fits on a node, and how to reserve it in Europe.
AMD Instinct MI355X: the CDNA 4 accelerator, explained
288 GB of HBM3e and an open ROCm stack, the credible non-NVIDIA option. MI355X vs B200 and B300, ROCm for inference, and how to reserve it in Europe.
No guides in this topic yet.