A 30-Person Startup Just Committed $410 Million to Amazon Before Shipping Anything
Recursive Superintelligence, an AI startup that emerged from stealth in May 2026 with $650 million in funding, has signed a $410 million multi-year compute deal with Amazon Web Services. The company has fewer than 30 employees. It has no released product. And it just spent the majority of its fundraising on cloud bills before shipping a single thing.
That is either a bold bet on the future of AI research or the most expensive leap of faith in recent startup history. Possibly both.
What Is Recursive Superintelligence?
The company was founded by Richard Socher, former chief scientist at Salesforce and founder of AI search engine You.com. His co-founders read like a who’s who of serious AI research.
Yuandong Tian led reinforcement learning and LLM reasoning work at Meta’s Fundamental AI Research lab. Tim Rocktaschel is a professor of AI at University College London and a former principal scientist at Google DeepMind. Alexey Dosovitskiy co-authored the Vision Transformer paper that reshaped computer vision research in 2020. Josh Tobin came from OpenAI. Peter Norvig, co-author of the standard university AI textbook, serves as an adviser.
The team has fewer than 30 people total, with offices in London and San Francisco.
The goal is recursive self-improvement. Build an AI system that can analyse its own performance, identify weaknesses, design fixes, and implement them without a human engineer involved. Then let it run. Then let it get better at getting better.
“Our unique approach is to use open-endedness to get to recursive self-improvement, which no one has yet achieved,” Socher said after the May launch.
The $410 Million AWS Deal
On Tuesday July 28, 2026, Recursive Superintelligence announced the $410 million compute deal with Amazon Web Services. The multi-year agreement is meant to provide flexibility as the company scales up its self-improving systems.
“Partnering with AWS gives us the infrastructure to run self-improving AI research at a scale very few teams in the world are capable of. We’ve already shown our automated AI research system can outperform years of human optimisation. As our research loops run autonomously, that advantage will compound,” Socher said.
The math here is striking. Recursive raised $650 million at a $4.65 billion valuation. It is now committing $410 million of that to a single cloud provider. That leaves roughly $240 million for everything else, including salaries, offices, and research tools.
Socher told TechCrunch he expects this to be the smallest compute deal the company signs in coming years. “This is likely going to be one of the smallest compute deals we’re going to sign in the next few years,” he said.
Why Spend on Compute Instead of People?
Most AI startups spend their early funding on hiring. Researchers, engineers, product managers, operations.
Recursive is doing the opposite. The company’s compute-intensive methodology means that instead of spending heavily on headcount and operations, the bulk of its budget goes directly into raw computing power. The logic is that if you are building AI systems that improve themselves autonomously, the limiting factor is not the number of researchers you have. It is the amount of compute those systems can access.
Run the research loops at scale. Let the AI iterate faster than any human team could. Ship products from the output.
It is a clean theory. Whether it works in practice is what the next few months will show.
Products by October
Socher has put a public stake in the ground. “We are excited to build really amazing products that people can use, and you will see those within a few months, not within a few quarters or years. In October or so, you’ll see some actually tangible, useful things that you’ll be able to play around with,” he told TechCrunch.
That is a specific timeline from a company that has released nothing yet. October 2026 is roughly three months away. The pressure is real.
What AWS Gets from This
The deal is not just good for Recursive. For AWS, bringing Recursive on board helps attract top AI labs seeking options beyond Microsoft Azure and Google Cloud. AWS uses custom Trainium chips and Nvidia GPU clusters to compete in this space.
Amazon has been trying to position AWS as a serious alternative for frontier AI labs. Most of the biggest names in AI have existing relationships with Microsoft or Google. Recursive gives AWS a marquee research lab to point to, one focused on one of the most discussed research directions in the field.
The Bigger Picture
Recursive is not alone in this space. Ilya Sutskever’s Safe Superintelligence focuses on safety as the path to advanced AI. Yann LeCun’s AMI Labs works on world models instead of scaling language models. David Silver’s Ineffable Intelligence prioritises reinforcement learning.
What separates Recursive is its approach and its resource allocation. Others are hiring large teams and pursuing diverse research directions. Recursive wants to automate the process of building AI itself, and it is betting most of its capital that compute is the variable that matters most.
The global AI infrastructure market was valued at $142.8 billion in 2026. It is projected to reach $947 billion by 2035. Every major AI lab is making large compute commitments right now. What makes Recursive unusual is doing it at this scale, this early, with this few people, before a single product ships.
If the self-improving loops work the way Socher believes they will, October’s product release will be the first proof. The AI coding tools already helping developers today give some sense of where autonomous AI development could eventually lead.
