Woofun AI reports that SemiAnalysis founder Dylan Patel stated on the Sequoia Capital podcast 'Training Data' that AI inference may become one of the world's largest markets, potentially surpassing oil and contributing several percentage points to global GDP. Patel argues that task completion speed and value expansion outpace computing power growth, suggesting a long-term supply shortage.

Patel projects that OpenAI and Anthropic's combined power needs will exceed 100 gigawatts by 2030, with over half of new compute potentially moving to space by 2040 due to ground energy constraints. He noted that recent efficiency gains stem from model-level optimization rather than hardware, citing DeepSeek's NVIDIA Hopper-specific design.

Additionally, SemiAnalysis' InferenceX benchmark indicates inference costs drop roughly 60 times annually, while intelligence per watt improves by approximately 40 times.