Amazon has accumulated $225 billion in revenue commitments for its Trainium AI chips, making it one of the world's top three data center chip companies and fundamentally reshaping how enterprises access AI computing power. This massive backlog reflects a seismic shift in the semiconductor industry, where hyperscalers are no longer just buying chips from vendors like NVIDIA; they're building their own and selling them to competitors, creating an entirely new competitive dynamic.

The scale of Amazon's chip business is staggering. CEO Andy Jassy disclosed that Amazon's combined chip operation, which includes Trainium (for AI workloads), Graviton (for general computing), and Nitro (for security), currently runs at approximately $20 billion annualized revenue. If Amazon were to sell these chips on the open market as a standalone business, Jassy estimated the revenue could reach roughly $50 billion. To put this in perspective, that would make Amazon's chip division larger than many Fortune 500 companies.

What's driving this explosive demand? Amazon's latest Trainium3 chip reportedly offers 30 percent to 40 percent better price-to-performance compared to its previous generation Trainium2 chip. The Trainium2 itself delivers 30 percent higher price-to-performance than traditional graphics processing units (GPUs), which have long been the default choice for AI workloads. These performance advantages explain why Trainium3 is nearly fully sold out, and why customers are already reserving Trainium4 chips that won't launch for another year and a half.

Why Are Major Tech Companies Abandoning Traditional Chip Vendors?

The answer lies in cost and control. Major AI companies including Anthropic, OpenAI, and Meta Platforms are increasingly turning to Amazon's custom processors. In April 2026, Meta signed a deal to deploy millions of AWS Graviton chips, Amazon's ARM-based central processing unit (CPU), to power AI-related compute workloads like real-time reasoning, code generation, and agent coordination. This represents a significant shift because Meta had previously relied on Google Cloud and Microsoft Azure alongside AWS.

The timing of Meta's announcement was particularly notable. AWS announced the deal just as Google Cloud's Next conference wrapped up, effectively showcasing a major AI customer as validation for Amazon's homegrown chips. Google also makes custom AI chips and unveiled new versions at the same event, but Amazon's ability to secure Meta signals that custom silicon is becoming table stakes for hyperscalers.

Anthropic, the company behind Claude, took an even more dramatic step. The AI startup agreed to spend $100 billion over 10 years running its workloads on AWS, with a particular focus on Trainium chips. In return, Amazon committed an additional $5 billion in investment to Anthropic, bringing its total stake to $13 billion. This partnership underscores how tightly integrated custom chips have become with AI development strategy.

How Is Amazon Planning to Expand Its Chip Business?

Amazon is considering a major strategic shift that could reshape the entire semiconductor market. CEO Andy Jassy indicated that the company could begin selling physical racks of Trainium chips to external customers "over the next couple of years," moving beyond cloud-only access. This would represent a fundamental change in how Amazon monetizes its silicon, allowing enterprises to purchase dedicated hardware rather than renting cloud cycles.

Andy Jassy

The implications are significant for how organizations evaluate their AI infrastructure options:

• Supply Constraints: Trainium2 is effectively fully allocated, and Trainium3 reservations have consumed nearly all available capacity, creating intense scarcity across generations.
• Capital Investment: Amazon is planning approximately $200 billion in capital expenditures this year to support chip production and data center expansion, signaling massive confidence in demand.
• Competitive Positioning: Amazon's chips directly compete with NVIDIA's new Vera CPU, which is also ARM-based and designed for AI agentic workloads, but Amazon's advantage is vertical integration with its cloud platform.

The company's growth trajectory is remarkable. Amazon's semiconductor business grew nearly 40 percent quarter-over-quarter in the first quarter of 2026, with year-over-year growth in the triple digits. This performance has caught the attention of investors and analysts, who see Amazon's chip business as a major growth driver for the company and its partners.

What Does This Mean for the Semiconductor Industry?

Amazon's dominance in custom AI chips is creating a ripple effect throughout the semiconductor ecosystem. Marvell Technology, which designs custom AI processors and networking components for Amazon under a five-year partnership deepened in December 2024, has seen its stock double in 2026. The company's revenue grew 42 percent year-over-year to $8.2 billion in fiscal 2026, with earnings jumping 81 percent to $2.84 per share, largely driven by its Amazon relationship.

"Our chips business continues to grow rapidly and is larger than what a lot of folks thought. We saw nearly 40 percent quarter over quarter growth in Q1, and our annual revenue run rate is now over $20 billion and growing triple-digit percentages year over year," stated Andy Jassy, CEO of Amazon.

Andy Jassy, CEO at Amazon

Analysts project that the custom AI processor market will grow at a compound annual growth rate of 27 percent through 2033, according to Bloomberg data cited in the sources. This expansion suggests that Amazon's current $225 billion backlog may be just the beginning of a much larger market opportunity.

The broader implication is clear: the era of enterprises relying solely on NVIDIA and other traditional chip vendors for AI computing is ending. Hyperscalers are building their own silicon, selling it to competitors, and creating entirely new supply chains. Amazon's massive backlog and plans to sell physical racks represent a fundamental restructuring of how the world sources AI computing power, with profound consequences for chip makers, cloud providers, and enterprises planning their AI infrastructure for the next decade.