Peter DeSantis Takes Over AI, Chips, and Quantum Computing at AWS

Peter DeSantis is not a name you encounter in mainstream media. Yet if you use AWS, you use his work. For 27 years, this engineer has built a significant portion of the infrastructure that runs about one-third of the Internet. He led the launch of Amazon EC2 in 2006, the service that popularized cloud computing as we know it today. He drove the acquisition of Annapurna Labs in 2015, the Israeli lab that designs AWS custom chips. He oversaw AWS global infrastructure for 8 years, those thousands of servers distributed across 21 regions and 66 availability zones.

His annual technical presentations at re:Invent have become legendary among engineers. No marketing, no colorful slides. Just architecture diagrams, performance graphs, and explanations of how AWS solved a particular latency problem or optimized a specific hardware component. On January 7, 2026, Andy Jassy announced that DeSantis was taking over a new organization that brings together three of Amazon’s most ambitious technology bets: Nova artificial intelligence models, custom chips, and quantum computing.

A reorganization that reveals AWS strategy

This new organization reports directly to Andy Jassy, Amazon’s CEO, not to the AWS CEO. This organizational detail is not trivial. It signals that Amazon considers these technologies strategic at the group level, not just for AWS. Nova models already power Alexa, Amazon advertising, and e-commerce recommendation systems. Graviton chips equip a growing portion of Amazon’s internal infrastructure. Quantum remains a very long-term bet, but Amazon is investing heavily in it.

The decision to group these three areas under one leadership reveals AWS’s thesis: these technologies will reinforce each other. AI models are becoming larger and more expensive to train. Custom chips optimize the performance-per-dollar ratio. Advances in chip design accelerate quantum processor development. And AI models help solve certain calibration problems in quantum systems.

Peter DeSantis explains it bluntly in the interview published by Amazon: building increasingly capable AI models requires massive computing investments. Having chips optimized for these workloads provides a competitive advantage. The teams developing Nova know where AI research will go in the next two years. This knowledge influences the chip roadmap, which takes several years to design. Getting these two teams to work together, in a coordinated but not rigid way, creates leverage.

What’s coming to your hands in 2026

The three products DeSantis cites as immediate priorities are Graviton 5, Trainium3, and Nova Forge. Graviton 5 was announced at re:Invent in December 2025. It is the most powerful general-purpose processor AWS has ever produced. Previous Graviton generations worked well for certain workloads but not all. Graviton 5 changes the game: almost any application running on AWS should now perform better on Graviton than on traditional x86 processors, at lower cost.

Trainium3 is the AI accelerator announced at the same event. It targets large-scale AI model training. AWS claims it offers better performance than competing GPUs at lower cost. Early customers are already testing the hardware. If you train models or use intensive inference services, Trainium3 should reduce your bill.

Nova Forge is the capability that allows you to take an existing Nova model and customize it with your own data to create what AWS calls a Novella. This is not classic fine-tuning. It is deep customization that adapts the model to your specific business domain. A model optimized for e-commerce does not work well for Alexa. A model designed for advertising has different needs than a model that drives robots in a warehouse. Nova Forge allows each Amazon division, and each AWS customer, to build their own frontier model without starting from scratch.

What this changes for your business

If you use AWS to host your applications, Graviton 5 should be part of your architecture discussions for 2026. Migration from x86 instances to Graviton is generally straightforward for containerized applications or interpreted languages. Cost savings can reach 20 to 30% depending on workloads. We have supported several clients in Alsace through this migration and the results are consistently positive.

If you use AI models in production, two developments deserve your attention. The first is the arrival of Trainium3 in AWS inference services. You do not directly manage the chips, but services like Amazon Bedrock that use these chips should become more performant and less expensive. The second is Nova Forge. Until now, customizing a frontier model required rare skills and substantial budgets. Nova Forge industrializes this process. If you have specific business data and use cases that do not work well with generic models, this capability changes the game.

Quantum computing remains out of reach for most businesses in 2026. AWS is developing the Ocelot chip and already offers Amazon Braket to experiment with quantum computers. But Peter DeSantis is clear: it will take several more years before quantum solves concrete business problems. It is a long-term investment that should not influence your architecture decisions today.

AWS’s long-term bet

This reorganization reveals a strong conviction at AWS: AI will not be a market where one player buys GPUs from a supplier and rents compute time to customers. It will be a market where vertical integration between chips, models, and applications creates a lasting advantage. Amazon already controls the entire chain: it designs its chips, trains its models, and deploys these models in applications that reach hundreds of millions of users.

For you, an AWS customer, this means two things. The first is that your investments in the AWS ecosystem should continue to benefit from performance improvements and cost reductions over the years. AWS does not depend on an external GPU supplier who could raise prices or ration supply. The second is that AWS AI services should become increasingly differentiated from competing offerings, thanks to this integration between hardware and software.

Peter DeSantis has spent 27 years building systems that work at Internet scale. He does not make marketing promises. He builds infrastructures that meet their availability and performance commitments year after year. The fact that Amazon entrusts him with its three most ambitious technology bets says a lot about the importance the company places on these areas. And about the willingness to make them work together rather than letting them evolve in silos.