Recursive Superintelligence signs $410 million computing deal with Amazon

On Tuesday, AI company Recursive Superintelligence announced a $410 million computing deal with Amazon Web Services. The company, which came out of stealth with $650 million in funding in May, is focused on building open-ended, self-improving systems, a potentially compute-intensive approach to AI research. This multi-year agreement is intended to provide flexibility as the company seeks to scale up these systems.

Recursive’s $410 million outlay represents the bulk of the company’s funding to date, but on a call with TechCrunch, founder and CEO Richard Socher emphasized that he expects this to be the first of many such deals. Today’s announcement is “probably one of the smallest computing contracts we’ll do in the next few years,” Socher said.

Recursive’s focus on self-improving AI systems means that more of the budget traditionally devoted to headcount and operations can be redirected to computing as the company looks to automate its own product development processes.

“For us, it’s not about the number of employees, it’s about the number of agents,” Socher said.

In contrast to the hybrid investment arrangements of major research institutes, Amazon’s involvement has no investment component. But the scale of the effort is so large that AWS can devote significant resources to supporting Recursive’s unique needs, which could help bring in other foundation-level AI companies in the future.

“Part of the agreement is that we will jointly develop infrastructure specifically for these types of companies,” said Jason Bennett, vice president of startups and venture capital at AWS.

Recursive self-improvement (RSI) has long been seen as a tipping point for AI, with some expecting an explosion of progress once AI can improve without human involvement. But as more labs and companies pursue the idea, the specific requirements have become hazy, with some predicting an imminent breakthrough and others characterizing self-improvement as continuous.

But for Recursive, the goal is to harness the power of RSI to develop a real product, and Socher plans to release the earliest samples by the end of the year.

“We’re excited to be able to develop really great products that people can use that they’re going to see within months, not quarters or years,” Socher says. “In October or so, there will be some tangible, useful things that you can actually use and play with.”