On Thursday, April 9, Amazon CEO Andy Jassy released his annual shareholder letter,Disclosed that the company is considering selling self-developed chips to third parties. Jassy stated in his annual letter to shareholders that Amazon's internal chip department currently has annual revenue of more than 20 billion US dollars. If it is an independent business and sells to AWS customers and external third parties, the annual revenue can reach 50 billion US dollars.
Jassy wrote in the letter:
The market demand for our chips is extremely strong.We will likely sell racks of chips to third parties in the future.
Amazon has been involved in chips for many years, but it used to have a completely different business model than Nvidia or Intel. Previously, Amazon did not directly "sell" chip entities to third parties, but "rented" the computing power of these chips to enterprises through its cloud service AWS.
The scale of self-developed chip business is exposed for the first time
Amazon's chip division mainly produces three categories of products: general computing chips, AI accelerators, and specialized chips that improve server operating efficiency.
Currently, the above-mentioned hardware resources are provided to AWS cloud computing customers on a lease basis and are not sold directly to the outside world.
Jassy compared the evolution of the current AI chip landscape to the previous structural migration in the CPU field.Amazon launched its self-developed CPU chip Graviton in 2018. Its price-performance ratio is up to 40% higher than that of x86 processors. It has been widely adopted by 98% of the top 1,000 EC2 customers.
Amazon EC2 (Elastic Compute Cloud) is one of the core cloud computing services provided by Amazon AWS, allowing users to rent virtual servers on demand in the cloud to run applications.
In the field of AI chips, Amazon is copying this path.The second-generation AI training chip Trainium2 has a price-performance ratio that is about 30% higher than similar GPUs, and is now basically sold out.
The performance of Trainium3, which will start shipping in early 2026, is 30% to 40% higher than that of Trainium2, and the subscription rate is close to full capacity. Trainium 4 is still about 18 months away from its official launch, and a considerable part of its production capacity has already been booked.
The shareholder letter specifically pointed out that Amazon Bedrock, the main inference service of AWS, has most of its inference workloads running on Trainium and the demand continues to be strong.Jassy also revealed that he would not rule out selling Trainium chip racks to third-party customers in the future.
AI computing power shortage drives demand spillover
The huge demand for computing power in AI model training has continued to put pressure on market supply and prompted companies to actively look for alternatives to NVIDIA GPUs.
It is against this background that the external sales potential of Amazon's self-developed chips has received more attention.
Jassy disclosed in the letter that two large AWS customers have proposed to purchase all Graviton instance production capacity in 2026, but they were unable to agree due to the need to take into account the needs of other customers. This detail reflects the current high tension in computing power supply.
Jassy admitted that AWS currently has capacity bottlenecks that prevent it from meeting some demands, and predicts that its total power capacity will double its current size by the end of 2027.
AWS will add 3.9 GW of new power capacity in 2025. Q4 2025 revenue will increase by 24% year-on-year, with annualized revenue reaching US$142 billion. He characterized the short-term free cash flow pressure as a necessary price, believed that AI was a "once-in-a-lifetime opportunity", and said that he "will not respond in a conservative manner."
Andy Jassy has been at the helm of Amazon for nearly five years. In this letter to shareholders, he elaborated on the growth potential of the chip business, reflecting Amazon's deeper layout intentions in the fields of cloud computing and AI infrastructure.
With the possibility of external sales raised, this previously behind-the-scenes business may become Amazon's next growth narrative for the capital market.