Anthropic reportedly signs a $45 billion, Nvidia-powered lease after backing AMD

Anthropic, the artificial intelligence lab backed by Google and others, has reportedly signed a far-reaching infrastructure lease tied to Nvidia-powered computing, a decision that comes only about a month after the company was described as placing a major bet on AMD chips. The development, first reported by Reuters and republished by Yahoo Finance, indicates that Anthropic’s procurement strategy has not followed a straight line, at least in the eyes of market watchers tracking the chip war.

According to the report, Anthropic’s new commitment is for about $45 billion and is described as Nvidia-powered. The lease is said to have been signed one month after an earlier, much larger wager involving AMD, which framed Anthropic as challenging Nvidia in a market where the dominant suppliers are fiercely competing for AI datacenter designs and long-term capacity.

The Reuters report, as cited by Yahoo Finance, said a person familiar with the matter indicated the lease size and the direction of the Nvidia-linked deal. The figure is significant not just because of its headline amount, but because leases of this magnitude are typically used to secure a pipeline of compute systems and chips for training and running frontier models, rather than acting as a short-term spot purchase.

What the announcement does not fully spell out in the coverage is as important as what it does. The report characterizes the deal as a lease and describes it as Nvidia-powered, but it does not, in the text available here, provide additional contract specifics such as the exact mix of systems, the start and end dates of delivery, or whether the lease includes related networking and storage components that datacenter operators often bundle into AI infrastructure programs.

For Nvidia, the reported lease would represent another major announcement that the company’s AI datacenter platform remains deeply embedded with the largest model builders. For AMD, the sequence would suggest the company is still in the fight but faces hurdles in converting perceived chip bets into durable, large-scale compute commitments at the very top end of demand.

The broader sector context is that competition among AI chip suppliers increasingly plays out less as a one-time procurement and more as a multi-quarter contest for access to scarce production capacity, reference designs, and the full stack required for AI workloads. Even when a buyer tests alternative hardware, the operational reality of scaling training runs and inference deployments can push teams toward the platforms they believe offer the best combination of performance, ecosystem support, and supply reliability.

It is also notable that Anthropic’s reported shift would not necessarily imply a public repudiation of prior decisions. Companies often experiment with different hardware generations or allocate workloads across platforms, depending on performance targets, software compatibility, and how quickly systems can be delivered. But without access to the underlying contract terms, it is not possible to determine whether the lease is replacing AMD deployments, supplementing them, or serving a distinct portion of Anthropic’s computing needs.

Going forward, investors and industry observers will likely watch for any confirmation or additional details from Anthropic, Nvidia, or the lessors involved, including delivery timelines, the specific Nvidia product families tied to the lease, and how much of Anthropic’s overall compute roadmap it covers. The next datapoint may also come from datacenter supply announcements and customer infrastructure updates that can clarify whether the reported $45 billion figure reflects a single implementation phase or a multi-year purchasing arc.