\n\n
The most recent wave of GPU cloud announcements centers on a massive scaling effort by Amazon Web Services and NVIDIA. The two companies said they will jointly deliver two million extra GPUs and next‑generation infrastructure aimed at "Agentic and Physical AI" ("AWS and NVIDIA to deliver 2 million additional GPUs and next‑generation infrastructure for Agentic and Physical AI" – Aug 26 2026). This push will deepen the capacity of cloud providers to run the most demanding AI models and will likely tighten competition among other providers.
Key developments:
- Scale: Adding two million GPUs represents a roughly 30‑40 % increase in available high‑end AI compute across AWS’s data centers, positioning the platform to handle the growing demand for large‑scale foundation models and real‑time inference workloads.
- Pricing landscape: Current on‑demand rates for NVIDIA H100 GPUs show AWS’s p5.48xlarge instance costing $6.88 per GPU‑hour, the lowest among major U.S. cloud providers ("AWS | p5.48xlarge | $6.88 | Normalized from an 8‑GPU node"). Competing options range up to $10.98 on Google Cloud, highlighting AWS’s price advantage as it expands capacity.
- CoreWeave partnership: CoreWeave is already integrating NVIDIA’s Vera Rubin NVL72 GPUs and Vera CPUs, with its "Cognition" customer running production agentic AI workloads on the new platform ("Cognition is the first CoreWeave customer running production agentic AI workloads on Vera Rubin"). CoreWeave’s own pricing for an 8‑GPU H100 node sits at $6.16 per GPU‑hour, close to AWS’s rates, indicating a tight market for high‑performance GPU rentals.
Why it matters:
- Performance for next‑gen AI: The influx of GPUs supports more extensive training runs for emerging agentic AI, such as NVIDIA’s Vera Rubin platform, and enables faster iteration cycles for developers.
- Cost competitiveness: With AWS offering the cheapest listed H100 rates, enterprises can access top‑tier AI hardware without prohibitive cost, encouraging broader adoption of large‑scale models.
- Strategic positioning: By coupling GPU volume with next‑generation infrastructure, AWS and NVIDIA aim to lock in a dominant share of the AI cloud market, pressuring rivals like Azure, Google Cloud, and specialized providers (e.g., CoreWeave) to accelerate their own hardware upgrades.
What to watch next:
- Deployment timelines: How quickly AWS can provision the promised two million GPUs and whether any regional bottlenecks arise.
- Pricing adjustments: Potential discounts or new instance types as supply expands and competition intensifies.
- Ecosystem growth: Additional partnerships (e.g., with CoreWeave’s Vera Rubin rollout) and the rollout of NVIDIA’s upcoming GPU architectures, which could further reshape the cloud GPU landscape. [[1, 2]]