SemiAnalysis founder: AI inference market size may surpass oil, becoming one of the world's largest markets According to BlockBeats, on July 1st, Dylan Patel, founder of SemiAnalysis, recently stated in an interview with Sequoia Capital's podcast "Training Data" that AI inference will become one of the largest markets in the world, possibly larger than oil and accounting for several percentage points of global GDP. He believes that after each iteration and upgrade of the model, the number and value of tasks that can be completed continue to expand faster than the growth of computing power, so the shortage of computing power may exist for a long time. Patel predicts that by 2030, the combined computing power demand of OpenAI and Anthropic alone will exceed 100 gigawatts; In the next 3 to 5 years, the impact of space data centers can still be ignored, but by 2040, more than half of the new computing power may enter space. He stated that the core constraints lie in the cost of ground energy and the ability to build electricity. Once the economic viability of space deployment exceeds that of the ground, the migration of computing power to space will become inevitable. In terms of hardware and software collaborative design, Patel stated that the efficiency improvement of AI in the past three years has not mainly come from hardware, but from model layer and cross layer collaborative optimization. He cited DeepSeek as an example, whose expert model shape is specifically optimized for the NVIDIA Hopper architecture, resulting in excellent performance on Hopper but poor performance on TPU; The Anthropic model is more suitable for TPU, while the OpenAI model leans more towards the GPU route. He believes that the so-called CUDA moat is essentially not just CUDA itself, but the open-source model ecosystem generally revolves around GPU collaborative optimization. Patel also stated that Nvidia CEO Huang Renxun's strong support for emerging cloud computing providers is aimed at avoiding the monopoly of computing power by super large scale cloud vendors and promoting the formation of a multipolar market. In addition, the real-time inference benchmark system InferrenceX built by the SemiAnalysis team shows that under equivalent quality, inference costs decrease by about 60 times per year, and intelligence improves by about 40 times per watt.