Recursive Superintelligence signs $410M AWS deal
Recursive will spend most of its $650 million round on AWS compute for automated AI research, with no AWS equity stake disclosed.
By Ingrid Halvorsen · Venture Capital Reporter
· 4 min read
The Recursive Superintelligence AWS deal commits $410 million over multiple years to Amazon Web Services, giving the AI lab cloud capacity for its automated research platform. The size of the commitment matters because it absorbs most of the $650 million Recursive raised in May 2026 at a $4.65 billion valuation, underscoring how quickly frontier AI budgets are shifting from headcount to compute.
Recursive said AWS will provide infrastructure for running research loops that can propose experiments, execute them, evaluate results and decide the next step. The companies also plan to work together on infrastructure for large-scale automated AI research. AWS is not taking an equity stake in the company, according to the announcement.
Richard Socher, Recursive’s chief executive and co-founder, said the partnership gives the company enough infrastructure to run self-improving AI research at a scale available to few teams. He claimed Recursive has already shown that its automated AI research system can beat years of human optimization, and said the benefit should grow as research loops run on their own. The company did not disclose unit economics for the AWS commitment or the exact term of the contract beyond saying it is multi-year.
Why did Recursive Superintelligence sign an AWS deal?
Recursive is trying to make AI systems that improve through repeated automated experimentation, a model that can consume large amounts of compute if each loop generates more tests. Socher told TechCrunch the AWS agreement may be one of the smaller compute contracts Recursive signs in the next several years, and said the company is spending more on compute than on hiring.
The May 2026 financing was led by GV and Greycroft, with AMD Ventures and Nvidia also participating. That round put Recursive among the better-funded AI labs focused on self-improving systems, though the company has not yet released a generally usable product. Socher expects the first usable products to arrive by October 2026.
Jason Bennett, AWS vice president and global head of startups and venture capital, said self-improving AI creates rising compute demand because each research loop leads to another experiment. AWS said its infrastructure will let Recursive run those loops in parallel, using its cloud reliability, security and purpose-built compute. The statement did not specify how much of the work will use Nvidia GPU clusters versus AWS Trainium chips.
What is Recursive Superintelligence building?
Recursive describes its system as an automated version of the scientific method: it selects experiments, runs them, checks outputs and chooses what to test next. In its first public results, the company said the system surpassed a two-year human leaderboard record and outperformed a live human-AI collaboration, posting top results on NanoChat, NanoGPT Speedrun and SOL-ExecBench benchmarks.
The company is based in San Francisco and London and has more than 25 researchers and engineers. Socher, who also leads You.com and founded AIX Ventures, is one of seven co-founders. The group includes Yuandong Tian, formerly a research scientist director at Meta’s Fundamental AI Research lab; Tim Rocktäschel, a University College London AI professor and former Google DeepMind principal scientist; Alexey Dosovitskiy, a co-author of the Vision Transformer paper; and Jeff Clune, creator of the Darwin Gödel Machine at Sakana AI.
Recursive is one of several well-funded groups testing different routes toward more capable AI systems. Ilya Sutskever’s Safe Superintelligence is centered on safety, Yann LeCun’s AMI Labs is working on world models, and David Silver’s Ineffable Intelligence is prioritizing reinforcement learning. Recursive’s current signal to the market is that it is treating compute access, rather than a larger staff, as the main constraint.
For AWS, the contract gives it another AI lab customer as it competes with Microsoft Azure and Google Cloud for frontier workloads. Evolvance Market Research valued the global AI infrastructure market at $142.8 billion in 2026 and projected it could reach about $947.46 billion by 2035, a 23.4% annual growth rate. Recursive’s next test is whether that spending produces compounding research gains, rather than faster cloud burn.
This story draws on original reporting from TechFundingNews.
