[KG] H100 — risk

{
  "entity_slug": "h100",
  "entity_name": "H100",
  "entity_type": "product",
  "title": "H100 faces Blackwell heat, Cerebras speed claims",
  "narrative": "Nvidia’s H100 remains the most deployed AI accelerator, used by Super Micro, PayPal, DARPA, and Google Cloud, and even powering Shanghai’s underwater data center. But the graph reveals mounting pressure: Blackwell is marked as a direct competitor, and Cerebras WSE-3 claims 10x training speed over H100 on GPT-scale models. Meanwhile, Nvidia’s own NVFP4 on Blackwell cuts JAX training by 1.8x, signaling internal cannibalization. H100’s reliance on EUV lithography ties it to ASML’s capacity, a supply-chain bottleneck. The US cleared H200 sales to 10 Chinese firms, but H100 remains restricted—limiting India’s $8B AI market access. TensorWave’s $350M raise for AMD clusters shows buyers hedging. H100’s volume is unmatched, but its moat erodes as competitors ship faster, cheaper alternatives.",
  "key_points": [
    "H100 used by Super Micro, PayPal, DARPA, Google Cloud, and Shanghai underwater data center",
    "Direct competitor Blackwell and Cerebras WSE-3 claim significant speed advantages",
    "Nvidia’s own Blackwell NVFP4 improves JAX training 1.8x, threatening H100 upgrade cycle",
    "EUV lithography dependency creates supply risk; US export policy limits China sales to H200 only",
    "TensorWave’s $350M AMD cluster raise signals customer diversification away from Nvidia"
  ],
  "angle": "risk",
  "neighborhood_size": 8,
  "generated_at": "2026-06-14T15:41:39.945271+00:00"
}