Dylan Patel: The scale of the AI reasoning market will surpass oil.

CoinWorld News, Semianalysis founder Dylan Patel stated that AI inference will become one of the largest markets in the world, exceeding the size of oil and accounting for several percentage points of global GDP.
He predicts that by 2030, the combined computing power demand of OpenAI and Anthropic will exceed 100 gigawatts.
Patel believes that the core constraints lie in ground-level energy costs and power construction capabilities, and that migrating computing power to space will become inevitable.
He pointed out that AI efficiency improvements mainly come from model-level and cross-layer collaborative optimization, and that the DeepSeek expert model is specifically optimized for Nvidia's Hopper architecture.
Additionally, Patel stated that Jensen Huang's support for emerging cloud computing providers is to avoid the monopolization of computing power by hyperscale cloud vendors.
Inferences show that inference costs decrease by about 60 times per year, and intelligence per watt improves by about 40 times.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 5
  • Repost
  • Share
Comment
Add a comment
Add a comment
OutsiderOfZhiyuandao
· 5h ago
Energy costs are the bottleneck; in the end, AI comes down to who can secure cheap electricity.
View OriginalReply0
PeonyMemo
· 5h ago
"Jensen Huang's move to support small cloud vendors is quite ruthless, as antitrust actions have now reached the cloud."
View OriginalReply0
BittersweetArb
· 5h ago
DeepSeek is specifically optimizing for Hopper. Domestic models are increasingly understanding hardware.
View OriginalReply0
FoldedYield
· 5h ago
Space computing power sounds sci-fi, but the electricity cost is really unsustainable.
View OriginalReply0
  • Pinned