Amazon's 2 Million NVIDIA GPU Expansion Shows the AI Infrastructure Race Is Only Getting Bigger

Amazon Web Services is making another massive move in the AI infrastructure race. AWS and NVIDIA announced in late August 2026 that they plan to deploy 2 million additional NVIDIA GPUs across Amazon's global cloud infrastructure in 2027 and 2028 — an expansion designed to meet demand for AI computing that, in the words of NVIDIA CEO Jensen Huang, is running ahead of every forecast.

The announcement reinforces a clear reality: artificial intelligence is no longer just a software trend. It has become a physical infrastructure race built around data centers, high-performance chips, networking systems, energy capacity, and cloud platforms capable of supporting increasingly demanding AI models.

For businesses, developers, and enterprise technology teams, Amazon's latest GPU commitment is a signal worth understanding — because the compute being installed in hyperscale data centers eventually shapes the AI tools available to everyone else.

What AWS and NVIDIA Actually Announced

The headline number is 2 million GPUs, but the details matter. The new deployment covers NVIDIA's latest data center processors — Blackwell Ultra along with the upcoming Rubin and Rubin Ultra generations — scheduled to come online across AWS infrastructure in 2027 and 2028.

This comes on top of an earlier commitment. At NVIDIA GTC in early 2026, AWS pledged to deploy more than 1 million NVIDIA GPUs starting that year. Demand outpaced that plan, which is why the companies are now adding 2 million more — bringing planned deployments across both commitments to more than 3 million GPUs.

The partnership extends well beyond graphics processors. The announcement also includes:

• NVIDIA Vera CPUs on AWS — a new compute option aimed at agentic AI workloads.
• AI factories for the U.S. government — roughly 100,000 GPUs on secure AWS infrastructure supporting federal and national-security workloads.
• Physical AI and robotics — Amazon Robotics is adopting NVIDIA's Jetson, Omniverse, and Isaac platforms for simulation, synthetic training data, and route optimization.
• Open models and data processing — NVIDIA's Nemotron open models are coming to Amazon Bedrock and SageMaker, alongside GPU-accelerated data and search tooling.
• New EC2 instances — G7 instances featuring RTX PRO 4500 Blackwell Server Edition GPUs target inference and graphics workloads with significant generation-over-generation gains.

Why AI Workloads Need This Many GPUs

Modern AI systems require enormous amounts of computing power. Training and running advanced models involves processing massive datasets, handling complex parallel calculations, and serving millions of users or enterprise workflows at once.

NVIDIA GPUs became the backbone of the AI industry because they excel at exactly that kind of parallel processing — machine learning, deep learning, generative AI, simulation, and high-performance data work all map naturally onto GPU architectures.

The scale of that demand shows up in NVIDIA's own results. The company reported record data center revenue of $89 billion in its most recent quarter — up 117 percent from a year earlier and now representing over 90 percent of its total business. Cloud providers like AWS are the largest buyers.

And this is not just about chatbots. The expanded capacity supports enterprise AI agents, AI-assisted software development, robotics, scientific computing, healthcare research, industrial automation, generative media, cybersecurity, and large-scale model training and inference.

More Than Chips: CPUs, Networking, and Robotics

AI performance depends on more than the GPU alone. Large-scale AI infrastructure also requires fast networking between chips, high-bandwidth memory, efficient CPUs, optimized data pipelines, and reliable data center design. That is why the AWS-NVIDIA collaboration touches so many layers of the stack — from NVLink interconnects and Spectrum networking integrated with AWS's own Nitro and Elastic Fabric Adapter systems, up through open models and robotics platforms.

The robotics piece is easy to overlook but significant. Amazon operates one of the largest robot fleets in the world inside its fulfillment network. Standardizing on NVIDIA's physical AI platform — simulating warehouses in Omniverse, training robots on synthetic data, validating behavior before deployment — points to where AI is heading next: out of the chat window and into physical operations.

AI Factories Are Becoming the New Data Centers

One of the most important terms in the announcement is "AI factories." An AI factory is a large-scale computing environment designed specifically to produce intelligence rather than host websites. Instead of manufacturing physical products, these facilities ingest data, train models, run inference, and serve AI-powered applications around the clock.

Traditional data centers were built mainly for websites, databases, storage, and general cloud computing. AI factories are optimized for a different workload entirely: dense GPU clusters, advanced liquid cooling, extremely fast internal networking, and enormous energy budgets.

Amazon's expansion shows that cloud providers are redesigning infrastructure around AI as a core workload rather than an add-on service — and the inclusion of dedicated, classified-capable AI factories for the U.S. government underlines how strategic this capacity has become.

What the Expansion Means for Businesses and Developers

Most companies that want to use AI have no interest in buying and managing their own GPU clusters. Building private AI infrastructure is expensive, technically complex, and hard to scale. Cloud platforms offer the alternative: rent GPU-powered compute on demand and scale usage with the workload.

For developers, more available capacity means less friction experimenting with model training, fine-tuning, and inference without owning hardware. For enterprises, it means the compute required for serious AI deployment — custom assistants, internal automation, high-performance analytics — keeps getting easier to access.

The downstream effects reach small businesses too. As cloud AI capacity grows, the tools built on top of it get cheaper and more capable: customer support automation, inventory forecasting, product content generation, personalized recommendations, and business intelligence that once required large engineering budgets. The real opportunity is not using AI because it is popular — it is using AI to automate repetitive work, improve customer experience, and make better decisions from business data. Teams planning hardware upgrades to take advantage of these tools can start with the fundamentals covered in the business tech refresh guide.

The Power and Infrastructure Challenge

The scale of this expansion also raises hard questions. AI infrastructure requires enormous investment in power, cooling, land, networking, and hardware supply chains. As cloud providers race to add capacity, data center growth has become a major topic for utilities, regulators, local communities, and environmental groups.

Millions of additional GPUs mean substantial new electricity demand, and grid capacity is already a constraint in several major data center markets. The providers that solve energy sourcing, cooling efficiency, and siting most effectively may end up with a lasting advantage that has little to do with chips.

Where the Cloud AI Race Goes Next

Amazon is not alone. Microsoft, Google, Oracle, Meta, and a wave of specialist AI cloud providers are all investing heavily in GPUs and custom silicon — AWS itself continues to develop its own Trainium chips alongside its NVIDIA fleet. Businesses choosing a cloud provider increasingly weigh GPU availability, model support, cost efficiency, and networking performance alongside traditional factors.

The size of this commitment suggests both companies expect AI demand to keep climbing as organizations move from experimentation to production. Not every AI investment will pay off, and many businesses are still working out where AI delivers measurable value. But the infrastructure bet is unambiguous: the next phase of AI will be shaped not only by the models, but by the data centers, chips, networks, and energy strategy needed to run them.

Build Your Own Workstation for the AI Era

Hyperscale GPU clusters handle the training, but most day-to-day AI work still happens on a capable local machine — running creative tools, managing cloud workloads, editing content, and building software. A well-specced desktop or laptop with a solid GPU, fast storage, and enough memory goes a long way.

• Desktop computers and laptops for development and creative work
• High-resolution monitors for multitasking across tools and dashboards
• Keyboards, mice, and docks that round out a productive workstation
• Networking gear for fast, reliable connections to cloud services

Browse the Computer Electronics department to build a workstation ready for AI-assisted work, or start with the first laptop buying guide if a portable setup fits better.

Frequently Asked Questions

How many NVIDIA GPUs is Amazon adding to AWS?

AWS and NVIDIA announced plans to deploy 2 million additional NVIDIA GPUs — including Blackwell Ultra, Rubin, and Rubin Ultra models — across AWS global infrastructure in 2027 and 2028. Combined with a prior commitment of more than 1 million GPUs, planned deployments total over 3 million.

Why is Amazon buying so many NVIDIA GPUs?

Demand for AI computing on AWS has outpaced earlier forecasts. The expanded capacity supports generative AI, enterprise AI agents, robotics, data processing, analytics, and large-scale machine learning workloads for cloud customers.

What is an AI factory?

An AI factory is a large-scale computing environment designed specifically to produce and run AI systems rather than host general workloads. It combines dense GPU clusters, high-speed networking, CPUs, software, and specialized cooling and power infrastructure.

Will this expansion help AWS customers?

More GPU capacity generally means better access to scalable compute for AI development, model training, inference, and analytics. It can also make AI-powered business tools cheaper and more widely available over time.

Is AWS the only cloud provider investing in AI infrastructure?

No. Microsoft, Google, Oracle, Meta, and specialist AI cloud providers are all investing heavily in GPUs, custom chips, and data center expansion, making AI capacity a central battleground in cloud computing.