Cloud Computing Watch: AWS's Expanded Partnership with Nvidia & Anthropic's $45 billion, six-year compute agreement with Nscale
AWS Just Committed to 2 Million More Nvidia GPUs — the Same Day Nvidia's Own Earnings Explained Why
Two Numbers From the Same Day Explain the Entire Cloud Capacity Story
AWS and Nvidia announced plans to deploy an additional 2 million Nvidia GPUs across 2027-2028 — Blackwell Ultra, Rubin, and Rubin Ultra systems — building on a prior commitment to add more than 1 million GPUs starting in 2026. Hours earlier, Nvidia reported why that demand keeps materializing: $96.2 billion in quarterly revenue, data-center sales up 117% year-over-year to $89 billion, and a first-ever forward guidance of roughly 70% growth for the coming fiscal year. CEO Jensen Huang's standing claim — that demand remains constrained by supply, not customer appetite — got a very concrete data point today in the form of AWS's own multi-million-GPU order book.
The same day also brought evidence that AI labs are diversifying their compute sources even as they keep buying more from the traditional hyperscalers: Anthropic committed $45 billion over six years to British infrastructure provider Nscale — a company founded only in 2024 — for 460 megawatts of dedicated capacity. This edition covers what AWS's GPU commitment and Nvidia's guidance together say about cloud capacity planning, and what Anthropic's Nscale deal adds to the specialist-infrastructure-provider trend this series has tracked since CoreWeave's own backlog growth.
01 — The Cloud Lead
AWS's 2 Million-GPU Commitment Is the Clearest Capacity Signal Yet
AWS's expanded partnership with Nvidia covers far more than raw chip volume: the companies are extending collaboration across networking, data processing, open models, robotics, and dedicated "AI factories," including a secure 100,000-GPU system specifically for U.S. federal and national-security workloads. Amazon Robotics will separately adopt Nvidia's physical-AI technology for warehouse automation — meaning this deal reaches well beyond cloud data centers into Amazon's own operational infrastructure.
The scale genuinely complicates a narrative this series has referenced before: that Amazon's custom Trainium silicon competes with Nvidia GPUs for the same workloads. AWS is simultaneously investing in its own chip program and committing to millions of additional Nvidia units, because customers are demanding multiple compute options rather than converging on one architecture. That "both, not either" dynamic is likely to hold across every hyperscaler with a custom-silicon program, not just AWS.
02 — The Nvidia Demand Signal Behind Every Cloud Capacity Decision
Nvidia's first-ever year-ahead growth forecast — roughly 70% for its next fiscal year, on top of data-center revenue already up 117% year-over-year to $89 billion this quarter — is the demand backdrop justifying AWS's GPU commitment and every other cloud capacity announcement this series tracks. Nvidia's own $160 billion commitment toward memory supply, disclosed alongside the earnings, connects directly to the memory-scarcity story covered in recent editions: rising memory costs are now expected to pressure even Nvidia's gross margins, not just downstream cloud providers.
The more interesting wrinkle for cloud economics specifically: Nvidia is increasingly using its own balance sheet — investments, financing guarantees, partnerships — to support the infrastructure projects that then buy its chips. That's the same circular-financing dynamic this series flagged around Nvidia's OpenAI Ohio guarantee and its $500 billion third-party financing push. AWS's 2-million-GPU order doesn't appear to be financed through one of those Nvidia-backed structures, but the pattern raises a fair question worth tracking going forward: how much of the demand Nvidia reports is organic hyperscaler capacity planning versus demand Nvidia's own financing is helping create.
03 — Specialist Infrastructure Providers Keep Landing Mega-Deals
Anthropic's $45 billion, six-year compute agreement with Nscale — a British infrastructure company founded only in 2024 — extends the same trend this series covered with CoreWeave's backlog growth: specialized AI-compute providers are becoming legitimate, large-scale counterparts to the traditional hyperscalers, not niche alternatives. The deal covers roughly 460 megawatts at a West Virginia site and will run on Nvidia's next-generation Vera Rubin systems starting in late 2027 — a multi-year forward commitment that treats compute access the way an industrial company treats securing raw materials, locking in capacity years before it's needed. For Anthropic specifically, spreading commitments across Amazon, Google, Microsoft, and now Nscale continues the multi-provider hedging pattern this series has noted in Anthropic's compute strategy previously.
Three Cloud Signals
- Cloud capacity planning has shifted from cautious to aggressive. AWS committing to 2 million additional GPUs on top of a prior 1 million-plus commitment, alongside Nvidia's first-ever 70% forward growth guidance, shows the industry planning for sustained multi-year demand growth rather than near-term plateau.
- Custom silicon and merchant GPU purchasing are complementary, not competitive, strategies. AWS building Trainium while simultaneously ordering millions more Nvidia GPUs confirms hyperscalers are hedging across architectures to meet diverse customer preferences, not picking a single winning approach.
- Specialist compute providers are becoming durable, not transitional, infrastructure counterparts. Nscale landing a $45 billion commitment just two years after founding mirrors CoreWeave's trajectory, suggesting AI labs increasingly view purpose-built compute providers as permanent parts of their infrastructure strategy rather than stopgaps.
The CODEW Take
Today's news is the clearest evidence yet that neither hyperscalers nor AI labs are hedging their capacity bets downward. AWS locking in 2 million additional GPUs on the same day Nvidia posted 117% data-center revenue growth, and its first-ever 70% forward guidance removes any ambiguity about whether the largest cloud provider believes this demand is durable — it's placing an order that says yes, unambiguously, years in advance. Anthropic's parallel $45 billion Nscale commitment shows AI labs making the identical bet through a different channel, diversifying compute sources rather than consolidating around fewer providers.
The open question this series will keep tracking is how much of this reinforcing cycle — Nvidia's financing supporting infrastructure, infrastructure buying Nvidia chips, chip revenue funding more financing — represents genuine organic demand versus demand the financing itself is helping manufacture. What to watch next: whether Nvidia's 70% growth forecast survives contact with actual quarterly results over the next two reporting cycles, and whether Nscale can execute construction and power procurement at the scale its new $45 billion commitment requires.
Editorial Note: Cloud Computing Watch is The CODEW's recurring intelligence series tracking cloud infrastructure, hyperscaler economics, and the specialist providers reshaping how AI compute gets bought and sold.
Reviewed by Erwin Castro
on
Friday, August 28, 2026
Rating:
