By Titik Puspa 
In a significant escalation of their burgeoning partnership, Amazon and Nvidia have announced a substantially expanded collaboration, including a monumental commitment to integrate an additional two million Nvidia Graphics Processing Unit (GPU) chips into Amazon Web Services (AWS) data centers. This latest pact, unveiled during Nvidia’s latest quarterly earnings call, signals a deepening reliance on Nvidia’s cutting-edge hardware to fuel the insatiable demand for artificial intelligence computation, even as Amazon aggressively pursues its own custom silicon development. The deal is poised to inject tens of billions of dollars into Nvidia’s coffers, underscoring the immense value placed on its AI processing capabilities.
The sheer scale and accelerated timeline of this agreement are noteworthy. The new commitment will see Nvidia’s next-generation GPUs, including the highly anticipated Blackwell Ultra, Rubin, and Rubin Ultra architectures, deployed across AWS infrastructure in 2027 and 2028. This move follows closely on the heels of a prior agreement made just five months ago, where Amazon pledged to deploy over one million Nvidia GPUs across its AWS network starting this year. According to Nvidia, the demand generated since that initial agreement has "exceeded those expectations," necessitating this substantial increase. While financial specifics remain undisclosed, industry analysts estimate the value of these GPU shipments to be in the tens of billions of dollars, reflecting the premium commanded by Nvidia’s advanced AI accelerators.
Beyond the sheer volume of chips, this expanded partnership signifies a more comprehensive integration of Nvidia’s technology within the AWS ecosystem. Nvidia confirmed that its high-performance networking hardware, crucial for interconnecting vast clusters of GPUs, along with its open-source AI models, central processing units (CPUs), data processing software, and robotics platforms, will also be incorporated across AWS. This holistic approach aims to provide AWS customers with a more seamless and powerful environment for developing and deploying AI applications. The companies attribute this intensified collaboration to "surging demand" originating from a diverse range of clients, including nimble startups, established enterprises, cutting-edge AI research labs, and even governmental organizations.
This strategic deepening of ties between Amazon and Nvidia unfolds against the backdrop of Amazon’s ambitious internal chip development initiatives. The e-commerce giant has been investing heavily in its own custom silicon, particularly in the realm of CPUs, the foundational processors for servers. This strategic push is driven by a desire to reduce its dependence on external chip suppliers like Nvidia and potentially emerge as a competitor in the AI chip market itself. Peter DeSantis, Amazon’s AI chief, has openly discussed the company’s efforts to commercialize its in-house AI chips, such as the Trainium processors. These chips are designed as direct alternatives to Nvidia’s flagship H100 and upcoming Blackwell offerings for demanding deep learning workloads, with Amazon exploring selling them to other organizations for data center deployment. Furthermore, Amazon’s Graviton CPUs, built on the Arm architecture, are increasingly viewed as formidable contenders to traditional server chips from industry stalwarts like Intel and AMD.
Amazon’s custom silicon business has demonstrated significant growth, recently reporting an annualized revenue run rate exceeding $25 billion. This impressive figure is bolstered by a staggering $225 billion in total commitments from prominent AI labs, including industry leaders like Anthropic and OpenAI, for the utilization of Amazon’s custom chips. This substantial customer base and revenue stream highlight Amazon’s growing influence and capability in the custom silicon arena.
Despite Amazon’s internal advancements, Nvidia continues to maintain its dominant position in the AI chip landscape, often referred to as the "GOAT" (Greatest Of All Time) in this domain. The current agreement extends beyond just GPUs, with Nvidia planning to supply an unspecified number of its Vera CPUs to AWS. These Vera CPUs will be delivered in configurations either integrated with the forthcoming Rubin GPUs or as standalone units, according to Nvidia CFO Colette Kress. Nvidia CEO Jensen Huang has expressed significant optimism about the Vera CPU, previously projecting a "brand new $200 billion TAM" (Total Addressable Market) for the company’s offerings in this segment. Kress further elaborated that Nvidia anticipates Vera CPUs to be adopted by "every major hyperscaler, neocloud, AI lab, and system OEM," with initial shipments already underway to key partners like Oracle and SpaceX AI.
The expanded partnership also extends into Amazon’s operational and enterprise-level solutions. Notably, Amazon plans to integrate Nvidia’s comprehensive physical AI stack to power its vast fleet of warehouse robots. This integration includes Nvidia’s Omniverse platform for simulation and digital twin creation, Cosmos for world modeling, Isaac for robotics development, and Jetson for embedded AI computing. This move signals Amazon’s commitment to leveraging advanced AI for optimizing its logistics and supply chain operations. Nvidia recently introduced a new iteration of its Jetson platform, the Jetson Orin Nano 2, designed to make edge AI more accessible for entry-level robotics applications.
On the enterprise front, AWS will offer Nvidia’s Nemotron family of open AI models through Amazon Bedrock, AWS’s managed service for foundation models, and Amazon SageMaker, its cloud-based machine learning platform. This integration will provide AWS customers with direct access to powerful, pre-trained AI models, simplifying the process of building and deploying sophisticated AI applications.
Nvidia’s robust performance continues to be a driving force behind these strategic alliances. The company reported exceptional financial results for its second quarter, with sales reaching $96.2 billion, significantly surpassing analyst expectations. The data center segment was the primary contributor to this revenue, accounting for an astounding $89 billion, representing a remarkable 117% increase year-over-year. Looking ahead, Nvidia projects third-quarter revenue to reach $108 billion, with a portion of this growth anticipated from its next-generation Rubin GPUs, for which production shipments have commenced this quarter. Investors are keenly observing the initial sales figures of the Rubin GPUs for indicators of sustained demand for Nvidia’s upcoming hardware generations.
To meet the escalating demand and secure its future supply chain, Nvidia has made substantial commitments to bolster its manufacturing capacity. The chipmaker has allocated a staggering $279 billion towards securing supply and manufacturing capabilities for current and future data center projects, a significant increase from the $119 billion commitment reported in the previous quarter. This extensive investment includes approximately $92 billion earmarked for the remainder of the current fiscal year and an additional $87 billion projected for fiscal year 2028, reflecting a long-term strategy to ensure adequate production to satisfy the voracious appetite of the AI industry.
Nvidia CEO Jensen Huang emphasized the pivotal moment the industry is experiencing, stating, "The thing that matters for the industry is that AI is now doing productive and useful work. AI is generating profitable tokens. If we had more compute, we could generate more profitable tokens, which results in more profit for all of the services. This is the exact phase where we’re at, which is the reason why everybody’s leaning in." This sentiment underscores the transformative impact of AI, where increased computational power directly translates into tangible economic value and profitability for businesses and services leveraging these technologies. As AI companies continue to pour hundreds of billions of dollars into infrastructure, investors will be closely scrutinizing whether this substantial investment in compute indeed leads to commensurate increases in profits across the rapidly evolving AI landscape.
