The scope of the announcement
AWS and NVIDIA plan to add two million Blackwell Ultra, Rubin and Rubin Ultra GPUs across AWS infrastructure during 2027 and 2028. The companies also intend to bring NVIDIA Vera CPU systems to AWS and deepen integrations with Nitro and Elastic Fabric Adapter.
A separate government effort includes plans for 100,000 GPUs on secure AWS infrastructure for federal and national-security workloads. NVIDIA also says Nemotron open models will remain available through Bedrock and SageMaker, while CUDA libraries will accelerate data processing and vector indexing.
Why the stack is widening
Large AI clusters are constrained by memory bandwidth, networking, data movement, power and operations as much as accelerator count. The collaboration adds NVLink Fusion and custom high-bandwidth memory support to Amazon's Trainium roadmap, creating a heterogeneous rack architecture rather than a single-chip story.
For customers, deeper integration can simplify deployment but may also increase dependency on a combined platform. Portability tests, data-transfer costs, quota commitments and fallback capacity belong in any procurement evaluation.
What to verify next
The two-million figure is a future plan. Readers should watch delivery milestones, regional instance availability, power and water disclosures, realized utilization, security accreditation and pricing for the actual instance families.
Follow the AINewsInu homepage and Industry hub for infrastructure updates. Capacity headlines are meaningful, but useful compute is the product of delivered systems, reliable software, energy availability and workloads that produce measurable value.
Sources & further reading
Social-media activity is treated as a signal of attention, not proof. Product claims are attributed to the linked publisher or announcement.