Amazon just tripled its order of Nvidia chips over ‘surging demand’
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Amazon is adding 2 million Nvidia GPU chips (Blackwell Ultra, Rubin, and Rubin Ultra) to AWS data centers for delivery in 2027-2028, tripling the 1 million GPU deal it made five months ago after Nvidia said initial "demand has exceeded those expectations."
- Nvidia plans to send unspecified numbers of Vera CPUs to AWS alongside the GPUs (some integrated with Rubin, others standalone), and expects Vera to ship to "every major hyperscaler, neocloud, AI lab, and system OEM," including Oracle and SpaceXAI.
- Amazon is integrating Nvidia's full physical AI stack — Omniverse, Cosmos, Isaac, and Jetson — across its warehouse robot fleet, and will serve Nvidia's Nemotron open models on Bedrock and SageMaker.
- Amazon simultaneously competes with Nvidia via its own Trainium AI chips and Graviton CPUs, with its custom chip business hitting a $25 billion annualized run rate on $225 billion in total commitments from AI labs including Anthropic and OpenAI.
- Nvidia reported $96.2 billion in Q2 sales (data center revenue $89B, up 117% YoY), beating analyst estimates, and guided $108 billion for Q3 — its first quarter of Rubin production shipments.
- Nvidia committed $279 billion to secure supply and manufacturing capacity, up from $119 billion last quarter, including $92 billion in projected spending for the rest of the fiscal year and another $87 billion in fiscal 2028.
Why it matters: The tripled GPU order is worth tens of billions in unit costs alone, deepening Amazon's reliance on Nvidia even as its in-house Trainium and Graviton chip business crosses a $25 billion annualized run rate. Nvidia's $279 billion in supply commitments — more than doubling quarter-over-quarter — shows the chipmaker is now fronting capital at unprecedented scale to keep Rubin shipments flowing.
Ask SkimNews
