Amazon hopes to challenge Nvidia more directly by selling its AI chips
Amazon Web Services (AWS) is exploring selling its proprietary AI chips, like Trainium, to third-party companies, aiming to directly challenge Nvidia's dominance in the AI chip market. This strategic move could establish a standalone $50 billion annual business for Amazon.
Intelligence analysis by Gemini 2.5 Flash
AWS is considering a significant shift by offering its homegrown AI chips, such as Trainium, for direct sale to other data center operators. Driven by high internal demand and CEO Andy Jassy's vision, this initiative positions Amazon to compete more aggressively with Nvidia, despite current capacity constraints and potential impacts on its cloud service revenue model.
Imagine Amazon, which usually lets you use its special computer brains (called AI chips) through its online services, is now thinking about selling those brains directly to other companies. It's like a toy company that makes amazing building blocks for its own games suddenly decides to sell those blocks to other toy makers, hoping to compete with the biggest block maker in the world.
Analysis
Amazon's Strategic Pivot into Chip Sales
Amazon Web Services (AWS) is signaling a significant strategic shift by exploring the direct sale of its proprietary AI chips, such as Trainium, to external companies for use in their own data centers. This move, confirmed by AWS AI chief Peter DeSantis and hinted at by CEO Andy Jassy in his annual shareholder letter, marks a departure from AWS's traditional model of offering these chips exclusively as part of its cloud services. Jassy's letter highlighted the immense internal demand for Amazon's homegrown AI chips, suggesting that if the chip business operated independently, it could achieve an annual run rate of approximately $50 billion. This potential revenue stream underscores Amazon's ambition to capitalize on its hardware innovation beyond its cloud ecosystem.
The decision to sell chips directly is driven by the overwhelming demand AWS has experienced for its AI hardware. Jassy noted that current Trainium chip capacity, and even the capacity for the upcoming Trainium4, has sold out almost instantly, often more than a year in advance of availability. This robust demand suggests a strong market appetite for alternatives to existing AI chip solutions. By opening up direct sales, Amazon aims to tap into a broader market, potentially positioning itself as a key hardware provider in the rapidly expanding AI infrastructure landscape.
Challenging Nvidia's Hegemony
This strategic pivot positions Amazon as a more direct and formidable challenger to Nvidia's long-standing dominance in the AI chip market. While a $50 billion annual run rate for Amazon's chip business would still be considerably smaller than Nvidia's current $326 billion revenue run rate, it represents a substantial entry into the high-stakes hardware arena. The comparison to Intel's annual revenues further emphasizes the scale of Amazon's ambition. Nvidia, recognizing the evolving market, has also been expanding its own territory, with CEO Jensen Huang recently declaring a new $200 billion market for Nvidia in selling CPUs for AI, moving into areas traditionally held by Intel and AMD. Amazon's counter-move with its AI chips intensifies this competitive dynamic, creating a multi-front battle for control over the foundational components of AI.
The direct competition with Nvidia is not merely about market share but also about influencing the future direction of AI development. By offering its own chips, Amazon provides developers and enterprises with more choices, potentially fostering a more diverse and competitive ecosystem. This could lead to specialized hardware optimized for different AI workloads, driving innovation and potentially reducing costs across the industry. The move reflects a broader trend among tech giants to vertically integrate and control key components of their AI stacks, from software models to underlying silicon.
Capacity and Cloud Conundrums
Despite the ambitious outlook, Amazon faces significant challenges, particularly regarding manufacturing capacity. The article notes that AWS has historically struggled to produce enough chips to meet its own internal and cloud customer demand, with capacity selling out rapidly. To sell chips to third parties, Amazon would need to dramatically increase its manufacturing output, likely through partners like TSMC. However, TSMC has recently become Nvidia's largest customer, supplanting Apple, which means Amazon would have to "miraculously elbow Nvidia out of the way" to secure sufficient foundry resources. This bottleneck could severely limit Amazon's ability to scale its direct chip sales business.
Furthermore, the decision to sell chips directly introduces a potential conflict with AWS's existing cloud revenue model. Currently, the money AWS makes from its chips is a "waterfall effect," where customers pay not just for AI token processing but also for a suite of associated cloud services like storage, security, networking, and monitoring. Selling chips outright could reduce the incentive for companies to use AWS's full cloud stack, potentially cannibalizing these lucrative ancillary services. Amazon will need to carefully balance the direct revenue from chip sales against the broader ecosystem benefits and revenue streams generated by keeping its chips exclusive to its cloud platform. This strategic tension highlights the complex trade-offs involved in Amazon's push for greater hardware market penetration.
Key points
- AWS is considering selling its proprietary AI chips, like Trainium, to third parties.
- Amazon CEO Andy Jassy estimates this could become a $50 billion annual business.
- This move aims to directly challenge Nvidia's dominance in the AI chip market.
- AWS currently sells out its chip capacity instantly, including for future models.
- Selling chips directly could impact AWS's cloud service revenue model and manufacturing partnerships.
If Amazon successfully sells its AI chips, it could foster greater competition in the AI hardware market, potentially leading to more innovation and lower costs for companies developing AI. This diversification could also reduce reliance on a single dominant supplier, benefiting the broader AI ecosystem with more specialized hardware options.
Amazon faces significant hurdles, including existing capacity shortages for its chips and the challenge of securing manufacturing priority with TSMC, which is heavily committed to Nvidia. Selling chips directly could also cannibalize AWS's lucrative cloud services revenue, where the chips currently drive additional service sales.
