SemiAnalysis founder Dylan Patel said in a recent podcast that OpenAI and Anthropic could control 70% to 80% of the world’s newly added AI compute by 2028, with combined compute capacity potentially exceeding 100GW. Patel said the two companies already account for about 30% of annual new compute added globally, and that share is still rising quickly. He also argued that the business model for frontier AI labs is shifting as the revenue generated per unit of compute improves sharply. According to Patel, Anthropic is currently generating about $50 million in revenue per megawatt of compute, and that figure could climb to $100 million. That, he said, would allow OpenAI and Anthropic to buy or lease compute at prices of $25 million to $50 million per megawatt. Patel also projected that global AI-related capital spending will reach about $11 trillion from 2024 to 2029, with more than $5 trillion of that total likely requiring debt financing. He added that higher expected returns from AI infrastructure could lead tech giants to accept higher funding costs, putting pressure on overall credit rates, traditional asset valuations, and highly indebted economies.
SemiAnalysis founder Dylan Patel said in a recent podcast that OpenAI and Anthropic could take 70% to 80% of the world’s newly added AI compute by 2028, with their combined compute footprint potentially surpassing 100GW.
Patel said the two companies already represent about 30% of annual new compute added globally, and that share is still climbing quickly.
Rising economics of compute
Patel said the business model at frontier AI labs is changing, while the output efficiency of AI compute has improved sharply. He said Anthropic is currently generating roughly $50 million in revenue per megawatt of compute, and that figure could rise to $100 million.
On that basis, OpenAI and Anthropic would be able to purchase or lease compute at prices ranging from $25 million to $50 million per megawatt, according to Patel.
Capital spending outlook
Patel projected that global AI-related capital expenditures will total about $11 trillion between 2024 and 2029. More than $5 trillion of that amount, he said, would need to be financed through debt.
He added that the potential returns on AI infrastructure are much higher than those in traditional industries. As a result, tech giants may be willing to accept higher funding costs, a shift that could push up broader credit rates and weigh on traditional asset valuations as well as highly indebted economies.
Where new compute may go
Patel also said future additions to compute capacity may not be used mainly for external model inference services. Instead, a larger share could be directed toward internal research and model self-improvement inside AI labs.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
