AWS orders 1M Nvidia GPUs
Amazon Web Services has committed to buying 1 million Nvidia GPUs through 2027, locking hyperscalers to next‑gen Blackwell/Rubin silicon and stretching availability for on‑prem buyers. That deal strengthens cloud providers’ priority access to cutting‑edge AI compute and raises the bar for hybrid models that must juggle cloud burst capacity versus on‑prem determinism. (investing.com)
AWS said it will add more than one million NVIDIA GPUs — explicitly naming Blackwell and Rubin architectures — across its global cloud regions beginning in 2026. (aws.amazon.com) Nvidia vice‑president Ian Buck told Reuters that shipments will start in 2026 and run through 2027, and that the agreement covers GPUs plus Spectrum networking chips and ConnectX interconnects for AWS data centers. (thestar.com.my) Datacenter Dynamics reported on March 16, 2026 that AWS plans to deploy those Nvidia GPUs within the next 12 months, signaling an accelerated, region‑wide rollout timetable. (datacenterdynamics.com) Nvidia CEO Jensen Huang has framed Blackwell and Rubin as a greater‑than $1 trillion revenue opportunity through the end of 2027, providing the commercial context for large multi‑year hyperscaler buys. (money.usnews.com) Market tracking shows hyperscalers already control roughly 44% of global data‑center capacity today, with projections to reach about 61% by 2030, concentrating purchasing leverage for next‑gen accelerators. (datacenterfrontier.com) Procurement analysis warns that hyperscalers’ priority allocations are creating tighter component availability, longer lead times and upward price pressure for enterprise on‑prem customers. (fusionww.com) Buck told Reuters AWS will use a mix of seven Nvidia chips — including the newly introduced Groq family — for inference workloads and will deploy ConnectX and Spectrum X networking gear in AWS sites, shifting both accelerator and NIC supply toward cloud‑scale customers. (thestar.com.my)
