American tech giant Amazon (AMZN) and artificial intelligence chip startup Cerebras Systems recently announced a cooperation agreement aimed at integrating their computing chips to jointly launch a new service that accelerates AI applications. This move is seen as an important step in challenging the current industry leader, Nvidia, within the growing AI computing power market.
Cerebras, valued at approximately $23.1 billion, is a startup dedicated to developing new types of AI chips. Its chip design differs significantly from Nvidia’s flagship products, intending to capture a share of the market. Earlier this year, the company signed a $10 billion chip supply agreement with OpenAI to provide computing power for its chatbot.
According to the cooperation plan, Cerebras’ chips will be deployed within Amazon Web Services (AWS) data centers, connected with Amazon’s self-developed Trainium3 AI chips, and interconnected via Amazon’s customized networking technology. Cerebras CEO Andrew Feldman stated that this collaboration will enable AWS customers globally, from individual developers to large banks, to conveniently access Cerebras’ computing power. However, neither party disclosed the specific scale of this partnership.
The core of this collaboration is to jointly address the “inference” stage in AI applications—the process where a trained AI model receives user requests and generates answers. The two companies have broken this process down into two steps: first, the “pre-fill” stage, where the user’s natural language is converted into “tokens” that the AI can understand; second, the “decoding” stage, where the AI generates the answer based on these tokens.
Amazon’s Trainium3 chips will handle the “pre-fill” stage, while Cerebras’ chips will focus on the “decoding” stage. This model of divided labor and collaboration highlights the trend of the AI industry shifting from the model training phase to the inference phase. With the proliferation of AI tools and the surge in user numbers, the market’s demand for inference computing power that offers higher response speeds and better cost-performance ratios is growing. Companies are beginning to seek diversified chip sources rather than relying entirely on traditional GPUs.
This strategy aligns with Nvidia’s movements. It is reported that Nvidia will announce a similar plan next week, intending to integrate its GPUs with products from another AI chip startup, Groq.
It is worth noting that over the past five years, Amazon has performed the worst among the “Magnificent Seven” tech stocks, with a five-year annualized return of only 6.9%, even underperforming the S&P 500 Index, which had an annualized return of 11.7% over the same period. Entering 2026, its stock price has accumulated a decline of 7%, second only to the declines of Tesla and Microsoft.
Market concerns about Amazon primarily focus on its declining market share in cloud computing, the impact of tariffs on its e-commerce business, and investor skepticism regarding its massive investments in the artificial intelligence field. Earlier this year, Amazon announced a capital expenditure plan of up to $200 billion, a historic high, with most of it directed towards AI infrastructure construction. This has raised concerns among some investors about excessive spending and uncertain returns.
However, some analysts believe that the current negative sentiment actually creates a buying opportunity. Amazon’s current price-to-earnings ratio is approximately 29 times, on par with the average level of the S&P 500 Index, and its valuation is near a two-decade low. Analysts point out that Amazon’s increased investment in AI is aimed at overcoming capacity constraints and solidifying its leading position in the cloud computing market. This $200 billion capital expenditure is precisely for building necessary data centers and processing its backlog to meet future demand.
