Nvidia has committed to delivering 1 million graphics processing units (GPUs) to Amazon Web Services between 2026 and 2027, with shipments beginning this year. This isn't just a hardware transaction; it represents a fundamental realignment of how cloud giants build their AI infrastructure. The deal spans Nvidia's entire technology stack, including networking chips and newly acquired Groq inference processors, signaling that the era of cloud providers building everything in-house may be ending. What Makes This AWS-Nvidia Deal Different From Previous Cloud Partnerships? For years, Amazon Web Services has been fiercely independent when it comes to infrastructure. The company designed its own networking hardware, built custom chips, and controlled nearly every layer of its data center stack. This new agreement breaks that pattern dramatically. Ian Buck, Nvidia's vice president of hyperscale and high-performance computing, explained the scope of the collaboration: "Inference is hard. It's wickedly hard. To be the best at inference, it is not a one chip pony. We actually use all seven chips." The deal covers far more than just GPUs. AWS will also purchase Nvidia's Spectrum networking chips, Connect X networking equipment, and Groq chips, which Nvidia acquired through a $17 billion licensing agreement in late 2025. This multi-layered approach reflects a shift in how cloud providers think about AI workloads. Rather than optimizing for general-purpose computing, they're now building specialized stacks tailored to specific AI tasks. How Does This Align With Nvidia CEO Jensen Huang's Broader Vision? The timing of this AWS partnership is no accident. The delivery schedule lines up precisely with CEO Jensen Huang's projection that Nvidia is pursuing a $1 trillion sales opportunity for its Rubin and Blackwell chip platforms through 2027. That figure excludes CPUs, networking hardware, and Groq-based products, meaning the actual revenue potential could exceed $1 trillion when those categories are included. Huang has suggested that incorporating Groq technology could generate $300 billion in annual revenue per gigawatt of computing power, with expectations that approximately 25% of GPU workloads will be paired with Groq chips. The combined Nvidia-Groq configuration, called LPX, is currently positioned as an optional component of Nvidia's Vera Rubin platform. This AWS deal represents one of the first major deployments of this integrated approach at scale. Steps to Understanding the Infrastructure Implications of This Deal - GPU Supply Commitment: Nvidia will deliver 1 million units to AWS between 2026 and 2027, with shipments starting this year and continuing through 2027, establishing a clear multi-year supply chain. - Inference Optimization Stack: AWS will deploy Groq chips alongside six additional Nvidia chip types specifically designed to fine-tune performance for AI inference tasks, where AI models generate outputs in response to input data. - Networking Integration: Nvidia's Connect X and Spectrum X networking products will be installed in AWS data centers, marking a notable shift away from AWS's historically proprietary networking solutions. - Financial Secrecy: Neither Nvidia nor AWS has disclosed the financial terms of the agreement, leaving investors to estimate the deal's value based on GPU volumes and market pricing. Buck noted that AWS will continue developing its own networking solutions but is now collaborating with Nvidia on deploying Connect X and Spectrum X for "important workloads and biggest customers across AI". This hybrid approach suggests that AWS sees value in Nvidia's specialized networking hardware for the most demanding AI applications, even as it maintains its custom infrastructure for other workloads. What Should Investors Watch as This Deal Unfolds? Several critical metrics will determine whether this partnership delivers on its promise. First, how quickly Nvidia executes on the 1 million unit commitment and how those shipments are distributed across different GPU generations will reveal whether the company can sustain production at scale. Second, the pace at which AWS moves from pilot programs to broader deployment of Groq-based inference chips will signal whether the LPX configuration gains real traction in production environments. Third, the degree to which AWS actually incorporates Nvidia's Connect X and Spectrum X networking gear relative to its existing custom-built solutions will show whether Nvidia can penetrate the networking layer of cloud infrastructure, historically a stronghold for proprietary solutions. Finally, Nvidia's fiscal 2027 results, expected in early 2027, will offer concrete evidence of how much revenue this AWS agreement contributes to the company's bottom line. The competitive backdrop matters too. AMD and other semiconductor competitors will be watching closely to see how Nvidia deepens its integration within AWS's infrastructure stack. If Nvidia successfully becomes the default choice for AWS's most demanding AI workloads, it could lock out competitors for years. Conversely, if AWS finds ways to reduce its dependence on Nvidia's specialized chips, it could open doors for alternative suppliers. This deal represents more than a simple supply agreement; it's a bet on how cloud computing will evolve over the next two years. By committing to 1 million GPUs and an integrated technology stack, AWS is signaling that AI inference, not just training, will be central to its competitive strategy. For Nvidia, it's validation that Jensen Huang's $1 trillion opportunity thesis has real backing from one of the world's largest cloud providers.
