NVIDIA: Two years ahead of Google's TPU!
Research institution Bernstein stated after a meeting with NVIDIA's IR that NVIDIA believes its technology is two years ahead of Google's TPU. Although Google has made progress over the past decade, NVIDIA believes its more flexible platform has advantages in building large-scale cloud AI infrastructure due to the specificity of TPU design. NVIDIA is still waiting for the export regulations for H200 and the U.S. government's revenue-sharing guidance
Research institution Bernstein wrote after the meeting with NVIDIA IR:
NVIDIA is still waiting for specific details on the export of H200, and has not yet received formal guidance from the U.S. government regarding the proposed 25% revenue sharing linked to certain sales related to China.
While acknowledging Google's progress over the past decade, NVIDIA believes it is approximately two years ahead of Google's TPU.
Due to the rapid changes in artificial intelligence models, NVIDIA believes it is difficult for Google to get cloud service providers to adopt TPU, as TPU is designed for more specific types of models. In contrast, NVIDIA believes its more flexible and programmable platform remains the best choice for building large-scale cloud-based AI infrastructure.
