--- title: "Amazon initiates an AI cost revolution! Self-developed AI ASIC directly targets large model training, NVIDIA's computing power monopoly faces its strongest challenge" type: "News" locale: "en" url: "https://longbridge.com/en/news/277209673.md" description: "Amazon plans to extensively use its self-developed AI chips Trainium and Inferentia, aiming to reduce the development costs of large AI models. This move may weaken NVIDIA's market monopoly in the medium to long term, leading to marginal pressure. Amazon AWS will use self-developed AI ASIC computing clusters as its core computing system, marking significant progress in its AI computing industry chain. The new head of AI infrastructure, Peter DeSantis, stated that using self-developed chips to build models will significantly reduce costs" datetime: "2026-02-27T14:09:03.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/277209673.md) - [en](https://longbridge.com/en/news/277209673.md) - [zh-HK](https://longbridge.com/zh-HK/news/277209673.md) --- > Supported Languages: [简体中文](https://longbridge.com/zh-CN/news/277209673.md) | [繁體中文](https://longbridge.com/zh-HK/news/277209673.md) # Amazon initiates an AI cost revolution! Self-developed AI ASIC directly targets large model training, NVIDIA's computing power monopoly faces its strongest challenge According to Zhitong Finance APP, Amazon (AMZN.US), a leader in e-commerce and cloud computing in the United States, will conduct large-scale trials using its self-developed AI chips—specifically, the AI ASIC computing power cluster infrastructure named Trainium and Inferentia—to develop and iteratively update its proprietary artificial intelligence large models, thereby significantly reducing costs. This move by Amazon may bring "medium to long-term marginal pressure + weakened monopoly premium" to the expansion prospects of the AI GPU computing power system dominated by NVIDIA and AMD. Under the wave of AI inference and the trend of "micro-training" focused on embedding AI large models into enterprise operations, the more cost-effective AI ASIC technology route may launch the strongest impact to date on NVIDIA's nearly 90% market share monopoly in AI chips. From the perspective of the AI computing power industry chain and chip engineering, Amazon's cloud computing platform AWS uses self-developed AI chips to train AI large models, rather than focusing on the main AI inference computing power as before. This is a significant milestone for Amazon's self-developed AI ASIC computing power route, but not a starting milestone for AI ASICs—this milestone has already been proven by Google's TPU (part of the AI ASIC technology route). Now, Amazon AWS is fully upgrading its "self-developed AI ASIC computing power cluster infrastructure participating in AI" to "self-developed AI ASIC directly undertaking the core computing power system of its cutting-edge AI large models," which holds significant meaning for the industrial chain of self-developed AI chips among hyperscalers like Amazon, Google, and Microsoft. **Market concerns about NVIDIA's prospects are valid** Peter DeSantis, Amazon's new head of artificial intelligence infrastructure, stated in a media interview on Friday morning: "If we can build models on our self-developed AI chips, we can build these models at just a small fraction of the cost of pure AI large model providers." DeSantis also added: "There is indeed a cost issue when building ultra-large-scale AI data centers. If we ultimately want AI to change everything, the costs must be different." The market generally believes that NVIDIA (NVDA.US), the "superpower" in AI chips, currently still holds the majority market share in the core area of AI computing power infrastructure—the artificial intelligence chip market. The chip giant led by Jensen Huang has just announced significantly better-than-expected results for the fourth quarter of fiscal year 2026 and guidance for the next fiscal quarter, but its stock price fell sharply by 5% on Thursday, mainly due to increasing market concerns about the recent announcements from hyperscalers to launch more cost-effective AI ASIC chips based on self-developed models, which increasingly show signs of risk to NVIDIA's long-term absolute dominance in the core area of global AI infrastructure—the AI chip field. Undoubtedly, as Amazon announces its attempt to use Trainium and Inferentia to develop AI large models, the market's concerns are valid Earlier this month, Amazon's management stated that capital expenditures for 2026 will reach approximately $200 billion, far exceeding Wall Street's expectations. Amazon CEO Andy Jassy mentioned that part of this expenditure will be used for the development and upgrade of self-developed AI ASIC computing infrastructure. Jassy stated, "Given the strong demand for e-commerce services, traditional cloud computing services, and AI computing from our existing business, as well as the groundbreaking growth opportunities presented by large AI models, humanoid robots, and low Earth orbit satellites, we expect Amazon to invest approximately $200 billion in capital expenditures in 2026, which we anticipate will yield strong long-term returns on invested capital." **The wave of AI inference is coming, and NVIDIA may not still be the "biggest winner in AI."** The real novelty of Amazon's latest plan lies not in whether "self-developed AI ASICs can train large models," but in the intention to push self-developed AI chips from optional AI computing in the cloud to the core path of its own foundational model development. The AI training side, which is almost monopolized by NVIDIA's AI GPUs, requires more powerful AI computing clusters and rapid iteration capabilities across the entire computing system, while the AI inference side places greater emphasis on unit token costs, latency, and energy efficiency after the large-scale implementation of cutting-edge AI technologies. For example, Google has clearly positioned Ironwood as a TPU generation "born for the AI inference era," emphasizing performance/energy efficiency/cost-effectiveness of computing clusters and scalability. However, Amazon's latest actions demonstrate that AI ASICs may possess strong potential for training large models. The AI ASIC computing system will undoubtedly continue to weaken NVIDIA's monopoly premium and some market share in the medium to long term, rather than linearly replacing the GPU system. The fundamental underlying reason is that the core competition in the inference era is no longer just "peak computing power," but rather the cost per token, power consumption, memory bandwidth utilization, interconnect efficiency, and total cost of ownership after hardware-software collaboration. In terms of these metrics, ASICs customized for specific workloads with tailored data flows, compilers, and interconnects are naturally more capable of achieving high cost-effectiveness than general-purpose GPUs. However, for NVIDIA and AMD, this largely means that marginal pressure is real, but it is more likely to manifest as a decline in bargaining power, a loss of market share, and a compression of valuation premiums, rather than an absolute collapse in demand. AI ASICs will undoubtedly continue to impact NVIDIA's dominant GPU monopoly under the super wave of AI inference, but the impact is more about reshaping the industry's profit pool and customer procurement structure rather than rendering the GPU expansion logic ineffective. AWS has clearly positioned Trainium/Inferentia as dedicated accelerators for generative AI training and inference, with Trainium2 offering approximately 30%-40% better price-performance compared to its AI GPU cloud instances; meanwhile, Google recently publicly stated that Gemini 2.0's training and inference run 100% on TPUs. This indicates that "large cloud computing vendors using self-developed ASICs to undertake core model training/inference" is no longer a proof of concept but is entering a replicable industrialization phase However, if it is further extrapolated to say that "the GPU system will be quickly overwhelmed," it seems excessive. NVIDIA's real moat lies not only in the chips themselves but also in CUDA, the development toolchain, the breadth of model adaptation, and ecological inertia; Bloomberg analysts pointed out last year that over 4 million developers worldwide rely on CUDA, which means that a large number of cutting-edge training, complex mixed workloads, and new models that require rapid iteration are still more suitable to run on GPUs in the short term. Even as AWS promotes its self-developed AI chips, it is still introducing GPU systems into future chips and continuing to provide AI infrastructure based on NVIDIA's computing power; this precisely indicates that the true strategy of hyperscalers is not "de-GPU-ization," but rather to retain GPUs at the high-end training layer while increasing the proportion of ASICs in large-scale inference and their own model stack. Therefore, from an engineering reality perspective, the future is more likely to be "GPU + ASIC coexistence and layering," rather than a single route winning out ### Related Stocks - [Amazon.com, Inc. (AMZN.US)](https://longbridge.com/en/quote/AMZN.US.md) - [NVIDIA Corporation (NVDA.US)](https://longbridge.com/en/quote/NVDA.US.md) - [GraniteShares 2x Long AMZN Daily ETF (AMZZ.US)](https://longbridge.com/en/quote/AMZZ.US.md) - [T-REX 2X Long NVIDIA Daily Target ETF (NVDX.US)](https://longbridge.com/en/quote/NVDX.US.md) - [Direxion Daily AMZN Bull 2X Shares (AMZU.US)](https://longbridge.com/en/quote/AMZU.US.md) - [GraniteShares 2x Long NVDA Daily ETF (NVDL.US)](https://longbridge.com/en/quote/NVDL.US.md) - [Direxion Daily Semicondct Bull 3X ETF (SOXL.US)](https://longbridge.com/en/quote/SOXL.US.md) - [Direxion Daily NVDA Bull 2X Shares (NVDU.US)](https://longbridge.com/en/quote/NVDU.US.md) - [YieldMax NVDA Option Income Strategy ETF (NVDY.US)](https://longbridge.com/en/quote/NVDY.US.md) - [Roundhill AMZN WeeklyPay ETF (AMZW.US)](https://longbridge.com/en/quote/AMZW.US.md) ## Related News & Research - [SpaceX Doubles Down on Nvidia (NVDA) Chips Amid Musk's AI Expansion Plans](https://longbridge.com/en/news/279809619.md) - [Nvidia to sell 1 million chips to Amazon by end of 2027 in cloud deal](https://longbridge.com/en/news/279850825.md) - [BREAKINGVIEWS-Nvidia $1 trln goal creates more noise than signal](https://longbridge.com/en/news/279640359.md) - [Planet to build world’s first GPU-native AI engine for planetary intelligence with Nvidia](https://longbridge.com/en/news/279330399.md) - [L'Oreal and NVIDIA to expand their collaboration on using AI](https://longbridge.com/en/news/279470980.md)