Recently, Amazon and NVIDIA announced further deepening of their strategic cooperation. Following Amazon's announcement a few months ago about deploying over 1 million NVIDIA GPUs, the order volume has surged three times, with plans to introduce an additional 2 million NVIDIA GPU chips into its data centers. This collaboration between two global giants not only confirms the explosive growth in current AI demand but also brings the massive capital competition behind computing power to the forefront.
According to the details disclosed by the two companies, the newly added 2 million GPU chips include NVIDIA's latest Blackwell Ultra, Rubin, and Rubin Ultra architectures. These are expected to be officially deployed into AWS data centers between 2027 and 2028. Although the specific financial terms have not been disclosed, considering the high unit price of advanced GPUs, the transaction is undoubtedly worth hundreds of billions of dollars. NVIDIA stated that due to the continuous demand for computing power from startups, large enterprises, top AI laboratories, and even governments far exceeding expectations, this major deal was quickly finalized.
Notably, the collaboration between the two companies has gone beyond simple chip purchases. NVIDIA announced that its entire technology stack—including network hardware that connects thousands of GPUs into super clusters, open-source models, CPUs, data processing software, and robot platforms—will be fully integrated into AWS. Among them, NVIDIA's Vera CPU will also be delivered to AWS, with some integrated with the Rubin architecture and others used as standalone processors. In addition, Amazon plans to fully adopt NVIDIA's physical AI technology stack in its warehouse robots and enterprise-level operations, including the Omniverse digital twin platform, Cosmos world model platform, Isaac robot development platform, and Jetson edge computing hardware. On the enterprise-level AI service side, AWS's Amazon Bedrock and SageMaker platforms will officially introduce NVIDIA's Nemotron series of open-source large models.
At the same time as this epic collaboration unfolds, Amazon itself is actively promoting its own chip strategy, aiming to reduce reliance on NVIDIA. Currently, Amazon's self-developed chip business is growing rapidly, with its Arm-based Graviton series CPU challenging traditional server chips, while the Trainium chip designed specifically for deep learning workloads is seen by Amazon as a direct competitor to NVIDIA's H100 or Blackwell chips. Amazon executives previously mentioned that due to the total commitment of 22.5 billion USD from AI labs such as Anthropic and OpenAI, the annual recurring revenue run rate of its self-developed chip business has already exceeded 25 billion USD.
From NVIDIA's own financial data, its strong performance remains unshakable. In its recent quarterly report, NVIDIA reported total revenue of 9.62 billion USD, surpassing analysts' expectations. Among which, the data center business contributed 8.9 billion USD in sales, a 117% increase compared to the same period last year. Looking ahead to the third quarter, NVIDIA expects revenue to further rise to 10.8 billion USD, which will include the initial mass production shipments of the next-generation Rubin GPU.
To meet the continued surge in AI hardware demand in the coming years, NVIDIA is aggressively stockpiling supply chain and manufacturing capacity. The report shows that NVIDIA's commitments to supply chain and manufacturing capabilities for current and future data center projects have significantly increased to 27.9 billion USD, up from 11.9 billion USD in the previous quarter. This includes 9.2 billion USD in planned expenditures for the remainder of the fiscal year and 8.7 billion USD in projected expenditures for the 2028 fiscal year.
As NVIDIA CEO Huang Renxun emphasized during the earnings call, the core logic of the entire industry has shifted: AI is now generating actual and useful productivity, continuously creating "tokens" that bring substantial profits. As long as there is more computing power, more high-value profits can be created, which is the fundamental reason why all tech giants are willing to invest heavily and fully bet on AI infrastructure development.