ARTIFICIAL INTELLIGENCE

INTELLIGENT TECHNOLOGY

The company, which is developing recursive self-improving AI capable of automating scientific discovery and knowledge generation, will use AWS as its primary cloud provider to power increasingly complex research workloads. The collaboration aims to provide the computing scale required for AI systems that continuously refine their own capabilities through autonomous experimentation.

Richard Socher, CEO and co-founder of Recursive, said the partnership would enable the company to conduct AI research at a scale achieved by only a handful of organisations worldwide.

“ We’ ve already shown our automated AI research system can outperform years of human optimization. As our research loops run autonomously, that advantage will compound. We’ re excited to partner with AWS and make our system and its infrastructure the fastest at invention generation,” he said.

Recursive emerged from stealth in May with a valuation of US $ 4.65 billion, backed by GV, Greycroft, AMD Ventures and NVIDIA. The company was founded by former AI researchers from OpenAI, Google DeepMind, Meta AI, Salesforce AI and Uber AI, bringing expertise across natural language processing, robotics, computer vision, reinforcement learning and automated AI systems.

According to AWS, self-improving AI significantly increases computing requirements because every completed research cycle generates the next experiment. The collaboration is intended to provide the elasticity, security and specialised computing infrastructure needed to execute thousands of research loops in parallel.

Jason Bennett, Vice President and Global Head of Start-ups and Venture Capital, AWS, said Recursive had already demonstrated research progress that exceeded years of community-driven optimisation.

“ AWS gives them the elasticity to run those loops in parallel at massive scale, and the reliability, security and purpose-built compute to turn months of sequential research into days of parallel discovery,” he said.

Recursive recently reported that its automated AI research platform surpassed a two-year human leaderboard record while outperforming a collaborative human-AI approach across three benchmark tests: NanoChat, NanoGPT Speedrun and SOL-ExecBench.

Under the agreement, Recursive and AWS will also co-develop infrastructure specifically designed for large-scale automated AI research, supporting future generations of selfimproving systems.

Socher said recursive self-improving AI represented a fundamental shift in computing, arguing that AI capable of improving its own code could ultimately automate the scientific method itself. He added that the partnership with AWS provides the infrastructure needed to pursue that long-term vision while scaling AI research beyond traditional human-led development. • www. intelligentcio. com

INTELLIGENT CIO NORTH AMERICA

29