Amazon.com, Inc. (AMZN.US) launches an AI cost revolution! Self-developed AI ASIC accelerates large model training, challenging NVIDIA Corporation's computing monopoly.

The leading e-commerce and cloud computing leader in the United States, Amazon.com, Inc. (AMZN.US), will massivel get to try its self-developed AI chip the AI ASIC cluster infrastructure named Trainium and Inferentia to develop and update its proprietary large AI models, in order to significantly reduce costs. For the prospects of the AI GPU computing system dominated by NVIDIA Corporation and AMD, this move by Amazon.com, Inc. may bring about a "medium-term margin suppression + weakening of monopoly premium" effect. In the wave of AI inference and the trend of embedding large AI models into enterprise operations through "micro-training," a more cost-effective AI ASIC technology route may launch the most powerful challenge to NVIDIA Corporation's nearly 90% market share dominance in AI chips to date. From the perspective of the AI computing industry and chip engineering, Amazon.com, Inc.'s cloud computing platform AWS uses its own AI chips to train AI large models, rather than focusing on AI inference computing power as before, which marks an important milestone in Amazon.com, Inc.'s self-developed AI ASIC computing route, but not the starting point milestone that the Alphabet Inc. Class C TPU (belonging to the AI ASIC technology route) has already proven. Now, Amazon.com, Inc.'s AWS is upgrading from "using self-developed AI ASIC cluster infrastructure for AI" to "directly using self-developed AI ASIC to handle the most core computing power systems for their cutting-edge AI large models," which is of significant importance for Amazon.com, Inc., Alphabet Inc. Class C, and Microsoft Corporation, such mega-scale cloud computing giants that have their self-developed AI chip routes in the industry chain. Concerns about the future of NVIDIA Corporation are justified. Peter DeSantis, the new head of artificial intelligence infrastructure at Amazon.com, Inc., said in a media interview on Friday morning, "If we can build models on our own AI chips, we can build these models at only a small fraction of the cost of a pure AI large model provider." DeSantis also added, "There is indeed a certain cost issue with building ultra-large AI data centers. If we ultimately want AI to change everything, the cost must be different." It is widely believed in the market that "AI chip superpower" NVIDIA Corporation (NVDA.US) still holds the vast majority of the market share in the core area of AI computing infrastructure the artificial intelligence chip market. This chip giant led by Jensen Huang has just announced significantly better-than-expected performance for the fourth quarter of the 2026 fiscal year and guidance for the next quarter, but its stock fell by 5% on Thursday, mainly due to growing concerns in the market about hyperscalers' recent intensive announcements of launching more cost-effective AI ASIC chips based on self-developed models, indicating increasing signs of risk to NVIDIA Corporation's long-term absolute dominance in the core area of global AI infrastructure the AI chip sector. There is no doubt that with Amazon.com, Inc.'s announcement of its intention to use Trainium and Inferentia to develop AI large models, concerns in the market are justified. Earlier this month, Amazon.com, Inc.'s management said that capital expenditures in 2026 will reach approximately $200 billion, far exceeding Wall Street's expectations. Amazon.com, Inc. CEO Andy Jassy said that part of this spending will be used for developing and upgrading its self-developed AI ASIC computing infrastructure. Jassy stated, "Given the strong demand for our existing e-commerce services, traditional cloud computing services, and AI computing power, as well as the huge growth opportunities presented by large AI models, humanoid robots, and low earth orbit satellites, we expect Amazon.com, Inc. to invest around $200 billion in capital expenditures in 2026 and expect to bring strong long-term return on investment." As the wave of AI inference approaches, NVIDIA Corporation may not necessarily remain the "biggest winner in AI." The truly innovative aspect of Amazon.com, Inc.'s latest plan lies not in "whether self-developed AI ASICs can train large models," but in preparing to push its self-developed AI chips from optional AI computing in the cloud to the core path of its own model development. NVIDIA Corporation's AI GPUs need greater universality of AI computing cluster and rapid iteration capability of the entire computing system on the training side, while the inference side values unit cost, latency, and energy efficiency more after the large-scale landing of cutting-edge AI technology. For example, Alphabet Inc. Class C explicitly positions Ironwood as a TPU generation "born for the era of AI inference," emphasizing performance/efficiency/computing cluster cost and scalability. However, Amazon.com, Inc.'s latest move demonstrates the strong potential of AI ASIC in training large models. The AI ASIC computing system will undoubtedly continue to erode NVIDIA Corporation's monopoly premium and some market share in the medium to long term, rather than replacing the GPU system linearly. The fundamental underlying reason is that the core competition in the inference era is no longer just "peak computing power," but also total cost of ownership involving cost per token, power consumption, memory bandwidth utilization, interconnection efficiency, and the overall cost after software and hardware coordination. In these metrics, ASICs tailored for specific workloads, compilers, and connections are naturally more cost-effective than general GPUs. However, for NVIDIA Corporation and AMD, this largely signifies the existence of marginal suppression, but it is more likely to manifest as a decline in bargaining power, a share of the market being eaten away, and compressed valuation premiums, rather than an absolute collapse in demand. Under the AI inference super wave, AI ASIC will continue to impact NVIDIA Corporation's dominant position in the GPU monopoly, but the impact is more like reshaping the industry profit pool and customer procurement structure, rather than making the GPU expansion logic invalid. AWS specifically positions Trainium/Inferentia as accelerators dedicated to generative AI training and inference, with Trainium2 providing approximately 30%-40% better price performance compared to its AI GPU cloud instances. Alphabet Inc. Class C recently announced that Gemini 2.0's training and inference run 100% on TPU. This indicates that "mega cloud computing companies using self-developed ASICs for core model training/inference" is no longer a concept validation, but has entered a replicable industrialization stage. But extrapolating it further to mean that "GPU systems will be rapidly undermined" seems too extreme. NVIDIA Corporation's true moat lies not only in the chip itself but also in CUDA, the development toolchain, model adaptability breadth, and ecosystem inertia; Bloomberg analysts pointed out last year that there are already over 4 million developers worldwide relying on CUDA, which means that a large number of cutting-edge training, complex hybrid workloads, and new models requiring rapid iteration still fit better on GPUs in the short term. Even AWS is advancing its self-developed AI chips while still introducing GPU systems in future chips, and continuing to provide AI computing infrastructure based on NVIDIA Corporation; this precisely shows that the real strategy of hyperscalers is not "deprioritizing GPUs" but retaining GPUs in the high-end training layer and increasing ASIC share in large-scale inference and their own model stacks. Therefore, from an engineering standpoint, the future is more likely to be "GPU + ASIC coexistence in layered fashion" rather than the victory of a single route.