Key Takeaways
- Amazon’s recently appointed AI chief, Peter DeSantis, is emphasizing cost efficiency over breakthrough AI performance capabilities.
- The e-commerce giant plans to use custom-built Trainium and Inferentia chips to create more economical AI solutions than rivals.
- Amazon’s current Nova model lags behind competitors in benchmark tests, though the company claims Nova 2 shows better performance.
- AMZN stock has fallen roughly 8% in January as investors worry about the company’s $200 billion capital spending roadmap.
- David Luan, former head of Amazon’s AGI Lab, announced his exit from the company this week.
Amazon is pursuing an unconventional artificial intelligence strategy: competing on price rather than cutting-edge capabilities.
Peter DeSantis, who recently took charge of Amazon’s AI division, is driving a cost-focused strategy. His core argument is simple — AI technology is too expensive, and Amazon can fix that.
“AI has a cost problem,” DeSantis stated. “If we ultimately want AI to transform everything, the costs have to be different.”
DeSantis took over AI leadership last December after previous chief AI scientist Rohit Prasad departed. With 28 years at the company, he was instrumental in building both AWS and Amazon’s chip-making business.
Amazon’s stock has dropped roughly 8% since January began. Investors are increasingly nervous about the company’s plan to spend $200 billion on capital expenditures this year — mostly for AI infrastructure — with Wall Street analysts projecting Amazon will burn through about $9 billion in cash in Q1 alone.
The pressure on DeSantis is immense.
Custom Silicon at the Core
The approach centers on Amazon’s custom-designed chips: Trainium for training AI models and Inferentia for running them. Amazon claims these chips offer up to 50% savings compared to similar products from rivals.
“If we can build our models on our chips, we can build them at a fraction of the cost of a pure-play AI model provider,” DeSantis stated.
This pricing advantage is already attracting customers. Nimbus Therapeutics, a Boston-based drug discovery firm, found Amazon’s Nova model performed as accurately as Anthropic’s Claude while costing only 10% as much.
Amazon also offers Nova Forge, allowing corporate customers to build custom AI models rather than paying for premium services like ChatGPT or Gemini.
Amazon’s flagship Nova product has lagged rivals in independent benchmark testing. The company claims Nova 2 performs significantly better, though independent third-party verification isn’t yet publicly available.
Amazon was also late to recognize the generative AI wave. After ChatGPT launched in late 2022, Amazon scrambled to develop a response, holding emergency planning meetings.
“Amazon was slower to realize the importance of generative AI,” said Lloyd Walmsley, senior analyst at Mizuho.
Talent Exodus and Competitive Pressures
Amazon faces serious talent retention challenges. Salaries for software engineers and research scientists lag behind Meta, OpenAI, Apple, and Anthropic, according to Levels.fyi compensation data. The company also cut approximately 30,000 corporate jobs across two layoff rounds.
This Tuesday, David Luan, who ran Amazon’s AGI Lab, announced he was leaving. The lab will continue operating under DeSantis’s direction.
DeSantis says he isn’t chasing splashy model launches like OpenAI and Anthropic. He described frequent releases as “kind of how you stay in the news” but stressed they don’t necessarily create real customer value.
Amazon says Nova model variants now handle more than 70% of Alexa requests. The company’s Rufus shopping assistant chatbot reached over 300 million users during 2025.
DeSantis acknowledged investor concerns about spending levels but defended the approach, pointing to similar skepticism Amazon faced during its early physical retail expansion and later AWS data center buildout.
Amazon’s AGI Lab, focused on building AI agents, continues under DeSantis’s management after Luan’s departure.
