Rohit Prasad, the executive who has led Amazon’s artificial intelligence initiatives and overseen the creation of its homegrown Nova AI models, is leaving the company at the end of the year.

In a memo Wednesday morning, Amazon CEO Andy Jassy named Peter DeSantis, a 27-year company veteran and top cloud infrastructure executive, to lead a new organization that combines its Nova and model research teams with custom silicon and quantum computing.

AI researcher Pieter Abbeel, who joined the company last year when Amazon hired the founders of robotics startup Covariant, will lead the frontier model research team within Amazon’s AGI organization, while continuing his work with the company’s robotics team, according to Jassy’s memo.

News of Prasad’s departure comes two weeks after Amazon unveiled its Nova 2 models at its annual re:Invent conference. The company is attempting to close the gap with AI rivals including OpenAI and Google in the race to develop increasingly capable AI systems.

Prasad joined Amazon in 2013 during the early days of Alexa and was named senior vice president and head scientist for artificial general intelligence in mid-2023 as part of a broader effort to recharge the company’s AI initiatives in the face of stiff competition.

In his memo, Jassy framed the reorganization as an effort to unify Amazon’s most important AI bets at an “inflection point” for the technologies. The new organization will bring together Amazon’s most expansive AI models, including Nova and AGI, with its custom silicon development group, which builds chips including Graviton, Trainium, and Nitro, as well as its quantum computing efforts.

DeSantis, who will report directly to Jassy, has led some of Amazon’s biggest technical initiatives. He launched Amazon EC2, the company’s core cloud computing infrastructure, oversaw the acquisition of chip designer Annapurna Labs in 2015, and most recently ran AWS Utility Computing, which includes compute, storage, database, and AI services.

Jassy called him a leader with “unusual technical depth” and a track record of “solving problems at the edge of what’s technically possible.”

Prasad’s departure was mentioned toward the end of Jassy’s memo, with the Amazon CEO saying that Prasad “has built a strong team, differentiated technology, growing customer momentum, and a culture of ambitious invention.” The memo described the departure as Prasad’s decision, calling him “missionary, passionate, and selfless” and thanking him for “everything he’s built here.”

It’s not yet clear what Prasad will do next.

In a separate memo, AWS CEO Matt Garman laid out a new internal structure consisting of seven AWS groups: Compute, Platform, and AI Services (led by Dave Brown); Storage and Analytics (Mai-Lan Tomsen Bukovec); Databases (G2 Krishnamoorthy); Security and Observability (Chet Kapoor); Agentic AI (Swami Sivasubramanian); Applied AI Solutions (Colleen Aubrey); and Infrastructure and Region Services (Prasad Kalyanaraman).

Amazon has positioned itself as a major player in enterprise AI through its Bedrock platform. Its Nova models are competitive on industry benchmarks. The Nova Forge service, launched at re:Invent, lets businesses and developers customize models using their own data.

Separately at re:Invent, the company unveiled a series of “frontier agents,” aiming to get ahead of the industry’s push toward autonomous AI systems for businesses.

But Amazon is still generally viewed as a fast follower in generative AI, trailing OpenAI, Google, and others in the perception of frontier model capabilities. Amazon has partnered closely with Claude maker Anthropic as a counterpunch to Microsoft’s partnership with OpenAI.

More recently, Amazon has also been in discussions to invest $10 billion or more in OpenAI, according to a report this week from The Information, citing three people familiar with the talks.

In an interview with GeekWire at re:Invent, Prasad gave no hint that he was preparing to leave. He described Nova Forge as “a game changer” and said the company was focused on proving that AI could deliver real value for its business customers.

Prasad took a pragmatic view of artificial general intelligence, pushing back on what he described as Silicon Valley’s tendency to portray AGI as “some kind of a god power.” He spoke instead of developing “generally intelligent systems that you can specialize for your purpose.”