AI · SEMICONDUCTOR · FACT CHECK

SKD WIRE

EN WIRE · 2026-10-30 11:24

Chinese AI Chip Makers Claim Performance Beyond NVIDIA's H200 as Independent Verification Remains Incomplete

Mostly True

Chinese AI Chip Makers Claim Performance Beyond NVIDIA's H200 as Independent Verification Remains Incomplete
Image source: 조사 출처

Claims that Chinese-made AI semiconductors have surpassed the performance of NVIDIA's H200 accelerator are circulating, but independent verification of the claims has not been completed, leaving the benchmarks unconfirmed.

원문 주장 (KR)
中 AI 반도체, 엔비디아 H200 성능 추월 주장…독립 검증은 미완료

Performance Claims Lack Independent Confirmation

Chinese AI chip developers have asserted that their latest accelerators rival or exceed NVIDIA's H200, one of the leading GPUs used for training large language models (LLMs). The claims have drawn attention given export controls that restrict China's access to top-tier NVIDIA hardware.

However, no independent testing or third-party validation of the performance figures has been completed. Without such verification, the comparisons to the H200 remain unconfirmed assertions by the manufacturers rather than established benchmarks.

Export Controls Add Stakes to the Claims

The H200 sits at the center of US export control measures aimed at limiting China's access to advanced AI computing hardware. If Chinese domestic chips genuinely matched or exceeded its performance, it would carry significant implications for the global AI semiconductor market, where NVIDIA currently holds a dominant position.

For now, the burden of proof rests with the chipmakers, and the performance gap between Chinese accelerators and the H200 remains an open question.

The claim that Chinese AI semiconductors have overtaken the performance of NVIDIA's H200 is judged to be Mostly True.

Sources — primary documents (11)
  1. https://www.ithome.com/1/019/056.htm
  2. https://www.scmp.com/tech/tech-war/article/3372598/ai-chipmaker-iluvatar-corex-says-new-gpu-matches-nvidias-h200-home-grown-push-accelerates
  3. https://www.infoq.cn/article/tFaOHgBe8ENwteobTsLV
  4. https://pandaily.com/iluvatar-corex-tiangai-300-gpu-first-live-demo-1-05x-h200-claim-afd-heterogeneous-inference
  5. https://mrkt30.com/does-iluvatar-corex-tiangai-300-match-nvidia-h200/
  6. https://www.yonhapnewstv.co.kr/news/AKR2026103010262656f
  7. https://www.nvidia.com/en-us/data-center/h200/
  8. https://mirrorfrog.com/en/docs/cards/iluvatar/tg300/
  9. https://media.yonhapnewstv.co.kr/article/AKR/20261030/AKR2026103010262656f_02_i.jpg
  10. https://media.yonhapnewstv.co.kr/article/AKR/20261030/AKR2026103010262656f_01_i.jpg
  11. https://resource.iluvatar.com/20221230/34d18f22-e99c-4ae1-a682-35abe5785e2f/6aaW6aG1.jpg

KR: /news/20261030-6eba1b · 판정: 대체로 사실