--- title: "Amazon's cost-effective AI chip products are favored by enterprises" type: "News" locale: "en" url: "https://longbridge.com/en/news/290193071.md" description: "Amazon's self-developed AI chips Inferentia and Trainium are favored by enterprises due to their price advantage. Analysts point out that their computing cost is 80% lower than NVIDIA's H100. Although local deployment is limited, more and more companies with self-built data centers are considering adopting this solution as a substitute for NVIDIA chips, thanks to its high cost-performance ratio" datetime: "2026-06-18T11:49:34.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/290193071.md) - [en](https://longbridge.com/en/news/290193071.md) - [zh-HK](https://longbridge.com/zh-HK/news/290193071.md) --- # Amazon's cost-effective AI chip products are favored by enterprises For companies selecting AI server chips for their own data centers, Amazon has a significant core advantage over NVIDIA: price. Irish consulting firm Co Driver Labs focuses on providing AI infrastructure implementation consulting services for enterprises. Analyst Carol Piatek stated that under the same computing load, using Amazon's second-generation **Inferentia inference chip** and **Trainium training chip** to deploy existing AI models for inference tasks can reduce computing costs by 80% compared to NVIDIA's H100 chip. More and more companies with self-built data centers are looking for cost-effective and more stable alternative chip solutions. In recent months, Amazon has been actively promoting its two self-developed AI-specific chips to these customers: Trainium for large model training and Inferentia for business inference, and an increasing number of companies are beginning to recognize this solution. A client in the insurance industry, due to compliance and regulatory requirements, must run AI operations for customers in its own data center and cannot use public cloud services. This company had previously been purchasing NVIDIA chips and is now looking for alternatives; after initial testing, Amazon's chips have shown impressive results due to their price advantage and stable computing capacity. Based on the good testing results, this insurance company immediately consulted with an Amazon Web Services (AWS) account manager about the possibility of deploying Inferentia inference chips in AWS Outposts local integrated machines. However, Amazon responded that this chip is currently not available for Outposts scenario testing. Piatek stated, "As long as the Inferentia chip can be deployed in the company's local data center, this insurance company will implement it immediately. Even if the cost of local deployment is twice that of renting cloud services, the company will still choose to deploy this chip provided by AWS in its own data center." He noted that while the cost of deploying chips in the company's local data center is likely higher than directly renting AWS cloud servers, the overall computing expenditure will still be significantly reduced compared to NVIDIA chips. An NVIDIA spokesperson responded by citing a recent blog post from the company: comparing chip procurement prices alone is of limited significance; the key is the actual computing output of the chips. If a low-cost chip can handle far fewer AI computing tasks in the same duration than a high-cost chip, then from the perspective of unit computing cost, the high-cost chip is actually more cost-effective. NVIDIA wrote in the blog post: "Our chips achieve the industry's lowest **token unit cost** and the highest token processing throughput." Sriram Kumaraesan, head of cloud and infrastructure services at the information technology consulting company Cognizant, stated that there is currently a tight supply of global computing power, and many clients are evaluating and piloting Amazon's self-developed chips, including scenarios for deployment in corporate local data centers. Kumaraesan mentioned in a written statement: "We have observed that the AI computing power demands of clients have exceeded the currently available supply, which has led enterprises to no longer rely solely on NVIDIA GPUs and to actively seek diversified computing solutions. Today, AWS's Trainium and Inferentia chips have become reliable production-grade alternatives in the industry, no longer just emergency options when GPUs are out of stock." However, Amazon's chips do not have an easy path. Chris Daniluk, CEO of cloud architecture consulting firm Rhythmic Technologies, stated that while many clients are exploring chip alternatives, they are currently only in the early research phase, and companies are generally concerned that adapting to these self-developed chips will require a completely new software system, posing technical implementation risks. At last December's AWS annual technology conference re:Invent, Amazon officially launched a new service that supports clients in deploying Amazon's self-developed AI chips outside of the AWS public cloud, while also releasing the AWS AI Factories solution, allowing enterprises to mix and deploy Amazon's Trainium chips alongside NVIDIA chips in their own data centers. For a long time, AWS has allowed companies with self-built data centers to deploy a full set of Amazon software, hardware, storage, and network resources locally through its Outposts integrated service. According to insiders, in recent months, the AWS sales team has begun preliminary discussions with companies such as AT&T to explore leasing Amazon's self-developed AI chips through the Outposts service. Additionally, insiders have indicated that Amazon is also negotiating to sell its self-developed AI chips directly to government and enterprise clients. In April of this year, Amazon CEO Andy Jassy mentioned in the annual shareholder letter that the company is considering selling its self-developed AI chips directly to external clients, which has significantly increased the willingness of enterprise clients to cooperate. Jassy has estimated that if Amazon no longer rents chips solely through cloud services but sells them directly, the annual revenue from the chip business could reach $50 billion. Currently, including general-purpose processors like Graviton and cloud-deployed Trainium chips, Amazon's chip business has an annual revenue of $20 billion. Data analysis from Information magazine shows that despite NVIDIA facing numerous competitive chip products and various industries accelerating the layout of AI alternative computing solutions, NVIDIA's market share in the AI inference computing power market continues to rise. AWS is following in Google's footsteps. Google has been engaging with companies like Anthropic since last year to promote the deployment of its self-developed TPU tensor processing unit AI chips in their own data centers; last month, Google also established a joint venture with Blackstone Group to create a dedicated TPU cloud business, competing with professional cloud service providers that are currently renting NVIDIA computing power to AI manufacturers ### Related Stocks - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [AMZN.US](https://longbridge.com/en/quote/AMZN.US.md) - [NVD.US](https://longbridge.com/en/quote/NVD.US.md) - [AMZU.US](https://longbridge.com/en/quote/AMZU.US.md) - [AMZW.US](https://longbridge.com/en/quote/AMZW.US.md) - [AMZZ.US](https://longbridge.com/en/quote/AMZZ.US.md) - [NVDS.US](https://longbridge.com/en/quote/NVDS.US.md) - [NVYY.US](https://longbridge.com/en/quote/NVYY.US.md) - [DIPS.US](https://longbridge.com/en/quote/DIPS.US.md) - [NVDY.US](https://longbridge.com/en/quote/NVDY.US.md) - [NVDQ.US](https://longbridge.com/en/quote/NVDQ.US.md) - [AMZD.US](https://longbridge.com/en/quote/AMZD.US.md) - [07388.HK](https://longbridge.com/en/quote/07388.HK.md) - [NVDB.US](https://longbridge.com/en/quote/NVDB.US.md) - [NVDG.US](https://longbridge.com/en/quote/NVDG.US.md) - [NVDO.US](https://longbridge.com/en/quote/NVDO.US.md) - [NVDU.US](https://longbridge.com/en/quote/NVDU.US.md) - [NVDW.US](https://longbridge.com/en/quote/NVDW.US.md) - [NVDL.US](https://longbridge.com/en/quote/NVDL.US.md) - [NYYY.US](https://longbridge.com/en/quote/NYYY.US.md) - [07788.HK](https://longbridge.com/en/quote/07788.HK.md) - [NVDX.US](https://longbridge.com/en/quote/NVDX.US.md) - [NVDD.US](https://longbridge.com/en/quote/NVDD.US.md) - [09388.HK](https://longbridge.com/en/quote/09388.HK.md) - [MAGX.US](https://longbridge.com/en/quote/MAGX.US.md) - [AMZO.US](https://longbridge.com/en/quote/AMZO.US.md) - [AMZY.US](https://longbridge.com/en/quote/AMZY.US.md) - [AZYY.US](https://longbridge.com/en/quote/AZYY.US.md) - [AMZP.US](https://longbridge.com/en/quote/AMZP.US.md) - [SMH.US](https://longbridge.com/en/quote/SMH.US.md) - [SOXX.US](https://longbridge.com/en/quote/SOXX.US.md) - [SOXL.US](https://longbridge.com/en/quote/SOXL.US.md) - [SOXQ.US](https://longbridge.com/en/quote/SOXQ.US.md) - [XSD.US](https://longbridge.com/en/quote/XSD.US.md) - [PSI.US](https://longbridge.com/en/quote/PSI.US.md) - [FTXL.US](https://longbridge.com/en/quote/FTXL.US.md) - [CTSH.US](https://longbridge.com/en/quote/CTSH.US.md) - [T.US](https://longbridge.com/en/quote/T.US.md) - [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md) - [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md) - [BX.US](https://longbridge.com/en/quote/BX.US.md) - [NVD.DE](https://longbridge.com/en/quote/NVD.DE.md) - [T-A.US](https://longbridge.com/en/quote/T-A.US.md) - [TBB.US](https://longbridge.com/en/quote/TBB.US.md) - [T-C.US](https://longbridge.com/en/quote/T-C.US.md) ## Related News & Research - [Amazon to sell custom artificial intelligence chips](https://longbridge.com/en/news/290302856.md) - [Chip startup Tensordyne expects $200 million in orders for AI system to rival Nvidia](https://longbridge.com/en/news/289795764.md) - [Why it is time to stop buying chip stocks](https://longbridge.com/en/news/290086818.md) - [Kyndryl's Expanded Amazon AWS AI Deal Highlights Push Beyond AI Pilots](https://longbridge.com/en/news/290225754.md) - [Nvidia Supply Chain Gets A Boost As Coherent Expands Texas AI Manufacturing Facility](https://longbridge.com/en/news/290038332.md)