Nscale acquires Anyscale for approximately $1.65 billion to advance in the AI sector.

      Nscale, the London-based AI-cloud firm that manages its own power and data centers, is advancing its capabilities through acquisitions. It has reached an agreement to purchase Anyscale, the startup responsible for the widely utilized Ray software, with Bloomberg estimating the deal at approximately $1.65 billion.

      Nscale announced the acquisition on Thursday but did not disclose the financial details. Bloomberg provided the figure, citing an insider familiar with the transaction. Anyscale’s team of around 200 employees, located across the US, Europe, and India, will become part of Nscale. The acquisition is expected to finalize in the latter half of 2026.

      The rationale behind a power company acquiring software lies in Nscale’s ongoing strategy of owning the underlying layers beneath the hardware: the energy supply, the facilities, and the equipment. By acquiring Anyscale, Nscale is positioning itself within the software that optimizes the workload of costly GPUs.

      This strategic move is driven by profit margins. Renting out raw GPUs on an hourly basis is a competitive commodity market, with rivals vying for the same Nvidia hardware. The real financial benefits exist at the layer that manages job distribution across thousands of chips simultaneously.

      Anyscale's platform, based on the open-source Ray framework, effectively handles this distribution of data preparation, training, fine-tuning, and inference across multiple systems. The company asserts that its platform can reduce the total cost of ownership by up to 90% compared to a collection of disparate tools.

      “Most infrastructure providers simply purchase GPUs and lease them,” remarked Nscale’s CEO and founder, Josh Payne. He emphasized that Nscale takes a different approach by building and owning every layer: from energy and data centers to computational resources and software.

      Dan Bathurst, Nscale's product chief, clearly articulated the offering, stating that engineers can now come to Nscale as “a one-stop shop for all their training, fine-tuning, and inference services.”

      Nscale is part of the so-called “neocloud” movement, a new wave of AI-centric data center startups chasing the growing market. Originating from a cryptocurrency-mining business in early 2024, it has rapidly expanded, surpassing 1GW of capacity and securing significant computing contracts.

      The company has also secured funding to facilitate further growth. In July, it finalized a $900 million revolving credit facility, in addition to a $2.5 billion investment for UK data centers. There are plans for an initial public offering, possibly later this year, and its board includes former Meta executives Sheryl Sandberg and Nick Clegg.

      Competitors are making similar moves. In May, Nebius invested $643 million to acquire Eigen AI, a company focused on reducing AI operational costs. Recently, chipmaker Qualcomm bought the compiler startup Modular. “Everyone is looking to move up the stack,” Bathurst told Bloomberg, indicating the drive to become the comprehensive provider as funding pours into AI infrastructure.

      What Nscale is purchasing specifically includes the commercial platform, the engineering talent, and the clientele, which features companies like Coinbase, Runway, and Bedrock Robotics. Anyscale will also become part of the foundation.

      Anyscale comes with considerable momentum, reporting a 70% growth in revenue during its most recent quarter. CEO Keerti Melkote described the merged entity as “the first full-stack AI hyperscaler.” The underlying strategy is to control every aspect from energy to the trained model, which aligns with the larger trend of building national AI factories.

      Whether this strategy will be successful largely depends on timing. Anyscale operates with current capacity. In contrast, Nscale’s largest deployment, a 1.35GW facility for Microsoft in West Virginia, is not set to come online until 2027.