Key Points:
- Amazon’s AI leader, Peter DeSantis, acknowledged that the company’s proprietary models lag behind industry leaders OpenAI and Anthropic for major workloads.
- Despite trailing on the software side, Amazon has deeply hedged its position by investing $50 billion in OpenAI and up to $33 billion in Anthropic.
- Amazon’s catch-up strategy relies on custom-designed silicon like Trainium chips and leveraging its unparalleled ocean of retail and logistics data.
- While developing its flagship Nova2 model, Amazon generates significant revenue by acting as a neutral cloud host for its main competitors on AWS.
Amazon’s top artificial intelligence executive has delivered a remarkably candid assessment of the company’s position in the generative technology race, acknowledging that its in-house software has not yet reached the industry’s absolute cutting edge. Peter DeSantis, the senior vice president overseeing Amazon’s AI models, custom chips, and quantum computing divisions, admitted that the tech giant’s proprietary models lag behind industry frontrunners like OpenAI and Anthropic for the largest, most demanding enterprise workloads. However, DeSantis expressed strong confidence that Amazon’s rapid infrastructure scaling, custom silicon development, and vast proprietary data stores will put the company in the conversation about leading models within the coming year.
The executive’s remarks shed light on Amazon’s unique, dual-pronged strategy in the artificial intelligence sector. While the company actively develops its own native systems, it also generates massive revenue by acting as a neutral infrastructure provider for other developers. Through its Amazon Bedrock platform, the company operates a model marketplace that allows enterprise cloud customers to access leading systems from OpenAI, Anthropic, Meta, and Mistral through a single, secure cloud interface. This diversified marketplace model ensures that Amazon profits from the AI boom regardless of which individual developer ultimately wins the technological race.
On the native development side, Amazon has struggled to build models that match the highly advanced capabilities of OpenAI’s GPT-5.5 or Anthropic’s Claude. The company launched its flagship in-house model series, Nova, in late 2025. The latest iteration, Nova2, has successfully attracted roughly 50,000 enterprise customers who utilize it for specialized business workflows. Despite this solid commercial adoption, Nova2 has not demonstrated the raw reasoning power, coding performance, or deep multi-step coordination required for the highly complex, frontier-level workloads that keep researchers and top-tier corporations flocking to its major rivals.
To hedge its bets, Amazon has forged deep financial and strategic alliances with the very competitors that outpace its native software. Earlier this year, Amazon invested a massive $50 billion in OpenAI as part of a record-breaking $110 billion funding round that valued the ChatGPT creator at $730 billion. As part of this historic alliance, OpenAI expanded its existing distribution agreement with Amazon Web Services by an additional $100 billion over the next eight years. The deal designated AWS as the exclusive third-party cloud provider for OpenAI’s premier enterprise platform, Frontier, and placed advanced models like GPT-5.5 directly onto the Amazon Bedrock service.
In addition to its massive bet on OpenAI, Amazon has committed up to $33 billion in Anthropic since 2023. This alliance includes a landmark $25 billion agreement signed in April, which granted Anthropic access to up to five gigawatts of power on Amazon’s proprietary Trainium computing clusters. In return, the Claude developer pledged to spend more than $100 billion on AWS cloud infrastructure over the next decade. This highly structured deal ensures a highly profitable circular economy, allowing Amazon to benefit directly from Anthropic’s commercial success through both its equity ownership and massive cloud infrastructure usage fees.
To close the gap with its heavily funded partners, DeSantis is relying on a combination of custom silicon and proprietary data. Rather than relying entirely on Nvidia’s expensive graphics processing units, Amazon has invested heavily in its own in-house microprocessors, specifically its Trainium and Inferentia chip series. The company’s recently launched Trainium3 processors deliver a 30% to 40% improvement in price-performance compared to traditional off-the-shelf options. DeSantis believes that deploying proprietary models on custom-built chips gives Amazon a structural cost advantage that will make its in-house models highly competitive as they scale.
The second pillar of Amazon’s catch-up roadmap involves capitalizing on its vast proprietary datasets. While other artificial intelligence developers face mounting legal hurdles and data scarcity when training new models, Amazon owns an unparalleled ocean of consumer, logistics, financial, and retail data. By training future iterations of the Nova model series on this highly specialized e-commerce data, the tech giant can build highly customized tools designed to automate complex supply chains, retail logistics, and consumer support. This targeted utility could prove far more valuable to corporate buyers than a generic, all-purpose virtual chatbot.
However, Amazon’s dual role as both a primary backer and a direct competitor to Anthropic and OpenAI is creating visible industry friction. Recently, Amazon security researchers identified critical software vulnerabilities inside Anthropic’s Fable 5 model, demonstrating how easily users could bypass the model’s safety safeguards. When Amazon shared these findings with the United States government, it led to an export control directive that temporarily forced Fable 5 offline over national security concerns. While Anthropic argued that the vulnerabilities were simple and common across the entire industry, the incident highlighted the awkward and potentially hostile dynamics of the modern tech sector’s co-opetitive partnerships.
Ultimately, Amazon’s willingness to acknowledge its model limitations highlights the long-term, structural approach the company is taking to dominate the next era of computing. By owning the foundational cloud infrastructure, manufacturing the specialized microchips, and hosting the dominant enterprise AI marketplaces, Amazon has positioned itself to profit immensely even if its own native models remain a step behind. For Peter DeSantis and the rest of Amazon’s artificial general intelligence team, the next year will be a critical test of whether custom silicon and specialized data can finally turn the cloud giant into a true frontier model developer.
