Meta and Anthropic discuss tens of billions of dollars in computing power cooperation to create a new model for monetizing AI infrastructure
Meta is in in-depth negotiations with artificial intelligence (AI) start-up Anthropic, and the two parties plan to reach a computing power leasing cooperation agreement worth up to US$10 billion in the next two years. This move shows that Meta is accelerating its entry into the cloud computing and computing power rental market, and that the core of competition in the international AI field is shifting from single model research and development to infrastructure monetization. According to people familiar with the matter, this cooperation was first proposed by Anthropic in June this year and is currently being evaluated by Meta. According to preliminary plans, Anthropic will pay monthly lease payments to Meta in installments over two years. Although both companies declined to comment officially and talks are still in the early stages, multiple media outlets including Fox Business independently confirmed the contact. Affected by this news, Meta's stock price fell by 6% during the trading day, then recovered part of the loss, and finally closed down by about 2%. Senior executives of cloud services (AWS) further highlight their determination to transform. For Anthropic, which is in a period of rapid expansion and plans to conduct an initial public offering (IPO) in October this year, ensuring a stable supply of computing power has become the key to its commercialization. In May this year, Anthropic signed a three-year, US$4.5 billion computing power contract with the US Space Exploration Technology Company (SpaceX) to obtain the computing power of its large data center in Tennessee. Industry experts believe that Anthropic is trying to reduce its dependence on a single partner by establishing a distributed infrastructure network with multiple hardware suppliers, thereby demonstrating a more flexible supply chain structure to the market before the IPO road show. Market analysis reiterated that this potential transaction highlights the structural changes in the current global AI industry chain. Meta's Llama large model is in direct competition with Anthropic's Claude large model, but the extreme scarcity of computing power has blurred the boundaries of competition between technology giants at the infrastructure level. This business model of "building your own base and renting excess computing power to competitors" is similar to the path that Amazon achieved through AWS in its early days. As the computing power layer gradually develops into an independent and profitable industry, the factors that determine the return on AI investment in the future will depend even more on the monetization efficiency of the infrastructure.