SemiAnalysis Founder Predicts 70-80% of AI Compute Power Will Be Controlled by Two Companies by 2028

Dynamic Beating AI News: In a recent podcast, Dylan Patel, founder of SemiAnalysis, predicted that by 2028, OpenAI and Anthropic could collectively account for 70% to 80% of global新增 AI compute capacity, with combined power consumption potentially exceeding 100 GW. Patel noted that the two companies currently represent approximately 30% of annual global新增 compute capacity, a share that is rapidly growing. He highlighted that the business models of leading AI labs are evolving, with significant improvements in AI compute efficiency—Anthropic currently generates about $50 million in revenue per megawatt of compute, a figure that could rise to $100 million. This enables OpenAI and Anthropic to procure or lease compute at premium prices of $25 million to $50 million per megawatt. More notably, Patel forecasts that global AI-related capital expenditures between 2024 and 2029 will reach approximately $11 trillion, with over $5 trillion requiring debt financing. Given the substantially higher potential returns of AI infrastructure compared to traditional industries, tech giants may accept higher financing costs, thereby pushing up overall credit rates and exerting pressure on traditional asset valuations and highly leveraged economies. Additionally, Patel believes that future新增 compute capacity may not primarily be used for external model inference services but rather increasingly directed toward internal R&D and self-improvement of AI models by the labs themselves.