Nscale agrees to buy Ray startup Anyscale for reported $1.65 billion

Nscale, led by CEO Josh Payne, has agreed to acquire Anyscale for about $1.65 billion, according to Bloomberg. The deal gives the AI infrastructure provider control of Ray, the distributed computing framework used to manage Python and AI workloads across clusters, completing Nscale's vertical integration strategy of owning data centers, GPU clusters, and software.

Nscale https://www.nscale.com/about?ref=runtimewire , founded and led by CEO Josh Payne, has agreed to acquire Anyscale https://www.anyscale.com/about?ref=runtimewire , in a deal that would give the fast-growing AI infrastructure provider control of software used to distribute demanding Python and AI workloads across clusters. Bloomberg reported https://www.bloomberg.com/news/articles/2026-07-30/nscale-to-buy-ai-software-startup-anyscale-for-1-65-billion?ref=runtimewire the transaction values Anyscale at about $1.65 billion, citing a person familiar with the private agreement. The purchase follows the architecture Payne laid out before Nscale had raised billions for AI infrastructure. In a December 2024 account of Nscale's strategy https://www.nscale.com/blog/nscale-155m-series-a-words-from-the-ceo?ref=runtimewire , Payne argued that a durable AI cloud needed to control the data center, the GPU cluster and the software. Anyscale supplies that third layer. Nscale has spent the past two years assembling power, data-center sites, networking and GPU capacity. Anyscale gives Payne a developer platform and an open-source distribution channel into the teams that decide where AI workloads run. The transaction is an agreement to acquire Anyscale, rather than a completed acquisition. Nscale targets software to improve utilization Nscale's pitch has centered on vertical integration. Payne, whose background spans capital markets, energy, data centers and supercomputing infrastructure, started with the physical constraints: finding power, constructing high-density facilities and securing accelerators. Nscale launched from stealth in May 2024 and announced a $155 million Series A https://www.nscale.com/blog/nscale-155m-series-a-words-from-the-ceo?ref=runtimewire that December. The capital buildout accelerated quickly. Nscale raised a $1.1 billion Series B https://www.nscale.com/press-releases/nscale-series-b?ref=runtimewire led by Aker ASA in September 2025, with participation from Sandton Capital, Blue Owl, Dell, Fidelity, G Squared, Nokia, Nvidia, Point72 and T.Capital. In March 2026, Aker and 8090 Industries led a $2 billion Series C https://www.nscale.com/press-releases/nscale-series-c?ref=runtimewire that Nscale said valued it at $14.6 billion. Citadel, Dell, Jane Street, Lenovo, Nokia, Nvidia and Point72 also participated. Those three rounds put $3.255 billion of announced equity behind Payne's plan. Nscale has also used acquisitions to extend control over physical infrastructure. It acquired American Intelligence & Power Corporation and its Monarch Compute Campus in West Virginia https://www.nscale.com/press-releases/nscale-acquires-american-intelligence-power-corporation?ref=runtimewire . Anyscale extends the same strategy into software. GPU ownership alone does not determine cloud economics. Scheduling, autoscaling, workload placement and developer experience affect how much expensive capacity stays productive. Idle accelerators are stranded capital, and forcing customers to rebuild applications for a particular cloud makes that capacity harder to sell. Anyscale brings Berkeley's Ray lineage Anyscale grew out of Ray, the distributed computing framework developed at UC Berkeley's RISELab in 2016 and 2017. Robert Nishihara, Philipp Moritz and Ion Stoica founded Anyscale in 2019 https://www.anyscale.com/press/founders-of-open-source-project-ray-launch-anyscale-with-usd-20-6m-in-funding-to-democratize-distributed-programmingfounders-of-open-source-project-ray-launch-anyscale-with-usd-20-6m-in-funding-to-democratize-distributed-programming?ref=runtimewire with Berkeley professor Michael I. Jordan in the founding and adviser group. The technical lineage matters. Nishihara and Moritz completed computer science doctorates at Berkeley while working on Ray. Stoica, a Berkeley computer science professor https://www2.eecs.berkeley.edu/Faculty/Homepages/stoica.html?ref=runtimewire , also co-founded Databricks and Conviva. Ray was built to let developers take Python and machine-learning applications from a laptop to a cluster without assembling a separate distributed systems stack for each workload. Anyscale commercialized that work as a managed platform spanning data processing, training, model tuning, inference and deployment. Developers write against familiar abstractions while Ray handles work across many machines. Keerti Melkote, the founder of Aruba Networks, became Anyscale's CEO https://www.anyscale.com/press/anyscale-names-industry-veteran-keerti-melkote-chief-executive-officer?ref=runtimewire in July 2024. Melkote had taken Aruba public in 2007 and led it through Hewlett Packard Enterprise's 2015 acquisition. His appointment put an experienced enterprise operator over Anyscale while Nishihara shifted toward customer and product strategy. At the time, Anyscale said Ray was used by over 50,000 developers and that adoption had grown sixfold year over year. Anyscale also said revenue had quadrupled over the preceding year. Those were company-reported figures from 2024, rather than current measures of the business Nscale has agreed to buy. Anyscale was still widening its cloud distribution shortly before the sale. In June 2026, Anyscale announced an Azure-native public preview https://www.anyscale.com/press/anyscale-on-azure-cost-efficient-and-sovereign-ai-microsoft-build?ref=runtimewire that lets customers run Ray workloads inside their own Microsoft Azure environments. That product reflects the strategic tension Nscale inherits: Anyscale's value rests partly on giving developers a portable compute layer, while Nscale benefits when that software directs more workloads toward Nscale-owned capacity. Preserving that portability will be central to the acquisition. Ray's reach comes from its usefulness across clouds, data centers and hardware configurations. Turning Anyscale into a narrow interface for Nscale infrastructure would weaken the open-source adoption funnel Payne is paying to acquire. A $1.65 billion outcome for Anyscale's backers Anyscale raised $259.6 million across four announced financings. Andreessen Horowitz led its $20.6 million Series A https://www.anyscale.com/press/founders-of-open-source-project-ray-launch-anyscale-with-usd-20-6m-in-funding-to-democratize-distributed-programmingfounders-of-open-source-project-ray-launch-anyscale-with-usd-20-6m-in-funding-to-democratize-distributed-programming?ref=runtimewire in 2019. NEA led a $40 million Series B https://www.anyscale.com/press/anyscale-announces-usd40m-in-series-b-funding-led-by-nea?ref=runtimewire in 2020. Andreessen Horowitz and Addition co-led a $100 million Series C at a $1 billion valuation https://www.anyscale.com/press/anyscale-secures-100m-series-c-at-1b-valuation-to-radically-simplify-scaling-and-productionizing-ai-applications?ref=runtimewire in 2021, followed by a $99 million extension https://www.globenewswire.com/news-release/2022/08/23/2502861/0/en/Anyscale-Unveils-Ray-2-0-and-Anyscale-Innovations-at-Ray-Summit-2022-Adds-an-Additional-99M-Funding-from-Existing-Investors-Addition-Intel-Capital-and-Foundation-Capital.html?ref=runtimewire backed by Addition, Intel Capital and Foundation Capital in 2022. The reported purchase price sits 65% above Anyscale's last publicly stated valuation. That comparison does not establish investor returns: the form of payment, liquidation preferences and retention packages will determine how the consideration is divided. For Payne, Anyscale offers a faster route to software depth than building another orchestration layer inside Nscale. For Anyscale's founders, the sale would place Ray's commercial operation inside a buyer with the capital and infrastructure to run the workloads their research anticipated. The execution test is whether Nscale can join those assets without compromising the infrastructure independence that helped Ray spread.