Anthropic, Nscale Eye Second $45B Compute Deal
Anthropic and data center startup Nscale are discussing a second deal worth $45 billion for computing power, reports say. This comes just months after they announced their first deal of the same size in August. If they agree, their total commitments would reach $90 billion. This would be one of the largest investments in AI infrastructure ever and shows how much computing power companies need as they race to build more advanced AI models.
Anthropic and Nscale are reportedly in advanced talks about another $45 billion computing deal. This could double the size of their partnership just two months after they first announced it. TechCrunch first reported the talks. If the deal goes ahead, the total value of the computing infrastructure commitments between the company behind Claude and the data center startup would reach $90 billion.
The possible deal shows how costly it is to develop the most advanced AI systems. Training and running large language models takes huge amounts of computing power. Companies like Anthropic are racing to secure enough capacity years ahead of time. Anthropic's recent decision to give 5% equity to a public benefit fund has led to speculation that the company may eventually go public through an initial public offering (IPO). Securing computing infrastructure would be an important step toward a possible stock market listing.
Nscale became an unexpected contender in the AI infrastructure market when it announced its first $45 billion deal with Anthropic in August. The company is building large data centers designed specifically for AI work. These centers have power and cooling systems built to support tightly packed groups of graphics processing units (GPUs), which are needed to train modern AI models. Traditional cloud providers such as Amazon Web Services, Microsoft Azure, and Google Cloud lead the market. But specialized infrastructure companies are finding opportunities by building facilities designed for particular needs.
The timing of these talks stands out because of recent changes in the AI market. Replit quietly ended its partnership with Claude months ago, citing high costs. This has raised questions about Anthropic's pricing. But Anthropic's willingness to commit tens of billions of dollars to infrastructure suggests it expects demand for its models to remain strong, even as it competes with OpenAI, Google, and other companies.
Anthropic's need for computing power also reflects its technical goals. The company has presented itself as a leader in AI safety. Recently, it allowed independent auditors to examine its work, giving them unusually broad access, including access at the employee level. Building AI models that are both safer and more capable requires extensive testing and attempts to find weaknesses. These tasks need significant computing power beyond what is required to train the models.
The market for AI computing power is growing very quickly. NVIDIA recently pledged $250 million for AI safety infrastructure in the UK, while Google supported AI research in Europe with a $3 million computing deal. These investments are much smaller than Anthropic's possible $90 billion commitment. Still, they show that securing computing power has become a major priority across the AI industry.
Nscale's rapid growth also shows how data center development is changing. Building traditional infrastructure can take years, but AI companies need computing power sooner. Startups like Nscale promise to build facilities faster and tailor them to specific needs. However, questions remain about whether they can deliver on commitments this large. Nscale will need to secure power agreements, allocations of GPUs from NVIDIA, and enough construction capacity to complete a project on a scale that few companies have attempted.
For Anthropic, securing computing power through 2030 and beyond would give it more time to compete with rivals that have greater financial resources. OpenAI's vice president of infrastructure recently left the company, which could give Anthropic a chance to hire experienced staff or gain a strategic advantage. Anthropic is also facing regulatory challenges, including a legal appeal in federal court against a Pentagon blacklist related to AI.
The details and schedule of the deal are still unclear. The August announcement gave few details about how payments would work, when the computing capacity would become available, or whether the commitment meant reserving capacity or buying actual infrastructure. A second $45 billion deal would likely have similar terms, but the speed of the talks suggests that both sides feel pressure to act. Anthropic needs computing power to stay competitive, while Nscale needs major customers to support its ambitious plans to expand.
People watching the industry will be looking to see whether other AI companies make similarly large deals. OpenAI, Google DeepMind, and Meta all need huge amounts of computing power. Securing long-term access to it could become essential to staying competitive. The key question is whether enough infrastructure can be built to meet demand, or whether a shortage of computing power could slow AI development across the industry.
The possible second $45 billion deal between Anthropic and Nscale shows that infrastructure has become one of the biggest limits on AI development. If their combined commitments reach $90 billion, the partnership could change the competitive landscape and push rival companies to secure their own long-term computing capacity. It remains to be seen whether Nscale can deliver on plans of this size. But Anthropic's willingness to make such a large commitment suggests it sees access to computing power as essential to its future. As AI models become larger and more complex, deals to build and secure the infrastructure behind them may become just as important as the algorithms themselves.