Nvidia inks AWS deal to supply one million AI GPUs through 2027

Nvidia will sell 1 million GPUs and a mix of other chips to Amazon Web Services (AWS) by the end of 2027, with shipments starting this year and running through 2027, an Nvidia executive told Reuters.

The deal covers Nvidia networking gear, such as Connect X and Spectrum X, and includes Groq chips and six other chips from Nvidia that AWS plans to use for AI inference workloads, the executive said.

Nvidia and AWS did not disclose the financial terms of the agreement.

Jensen Huang, Nvidia CEO, has previously said he sees a US$1 trillion sales opportunity for its Rubin and Blackwell chip families through 2027.

Ian Buck, Nvidia’s vice president of hyperscale and high-performance computing, said the companies will collaborate to deploy Connect X and Spectrum X in AWS data centres for major AI customers and workloads.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

The AWS-Nvidia deal runs both ways

The agreement makes Amazon Web Services (AWS) a leading Nvidia buyer, yet AWS also competes by building Trainium accelerators, custom chips for AI workloads 1.
AWS says these chips deliver 30 to 40% better price-performance than similar GPU-based instances on AWS, with large deployments that include a cluster of more than 500,000 Trainium2 chips for AI company Anthropic 1.
The work goes beyond shipping hardware, since both teams are weaving their systems together.
AWS plans to use Nvidia’s NVLink Fusion platform, technology for linking chips together at high speed, to connect next-generation Trainium4 chips, which adds a long-term technical tie instead of a simple vendor setup 2.

This huge GPU order may widen AWS’s AI compute lead

Executives at large cloud providers say demand still outstrips supply, with capacity sold as quickly as it comes online under ongoing constraints 3.
Locking in a million GPUs through 2027 strengthens AWS’s position in GPU-as-a-Service, a market projected to reach $26.6 billion by 2030 4.
That scale can squeeze smaller cloud firms that need scarce high-end chips to win enterprise AI work.
Even hyperscalers, giant cloud providers, with mature custom chip efforts still rely on Nvidia GPUs to keep up with customer needs, which supports a mixed chip market for years 1.