Nvidia will supply as many as one million of its graphics processing unit chips to Amazon’s cloud division over the next several years, marking one of the more substantial hardware commitments in the expanding market for artificial intelligence infrastructure.

The arrangement, which runs through 2027, was outlined by an Nvidia executive in comments to Reuters on Thursday. While both companies had earlier confirmed the scale of the order, the timeline had not been publicly detailed.

Sales tied to the agreement are expected to begin this year and continue steadily over the period. The chips form part of a broader package that goes beyond GPUs, reflecting the increasingly complex demands of large-scale AI systems.

Ian Buck, a vice president at Nvidia overseeing hyperscale and high-performance computing, said the transaction includes additional components such as Spectrum networking chips and newly introduced Groq chips. These are intended to support different layers of computing workloads inside data centres.

The companies have not disclosed financial terms.

The deal comes at a time when Nvidia is positioning its newer chip families, Rubin and Blackwell, as central to what its chief executive has described as a long-term sales opportunity worth up to one trillion dollars over the same period.

Within Amazon Web Services, the hardware will be used to support inference tasks, where AI systems generate responses and perform actions based on trained models. According to Buck, this part of the process presents distinct technical challenges and cannot be handled efficiently by a single type of chip.

AWS is expected to deploy a combination of seven Nvidia chips to manage these workloads. Buck described inference as “wickedly hard,” noting that performance gains depend on how different components work together rather than on any one processor alone.

The agreement also extends into networking infrastructure. Nvidia’s ConnectX and Spectrum-X systems are set to be installed in AWS data centres, an area where Amazon has traditionally relied on its own custom-built technology.

Buck indicated that this would be a collaborative effort rather than a replacement of AWS’s existing systems, with Nvidia hardware being introduced for specific high-demand AI workloads and larger customers.

Xiaomi Poco X8 Pro Max Price Set As New Poco X8 Series Arrives In India

For now, the deal signals a deepening relationship between the two companies at a moment when demand for AI computing capacity continues to build, reshaping how cloud providers plan their infrastructure.