Amazon Web Services (AWS) is dramatically expanding its use of Nvidia artificial intelligence chips, with the cloud giant planning to deploy 2 million additional Nvidia GPUs across its global infrastructure in 2027 and 2028. The expansion comes only about five months after Amazon announced plans to deploy more than 1 million Nvidia GPUs, meaning the latest commitment effectively takes the planned deployment to more than 3 million GPUs. Nvidia said customer demand has exceeded its earlier expectations.
The expansion underscores the scale of the AI infrastructure race as startups, enterprises, AI laboratories and governments increase spending on computing capacity. The latest AWS-Nvidia agreement also goes beyond GPUs, covering Nvidia Vera CPUs, advanced networking, open AI models, data-processing software and robotics technologies. The companies said the expanded partnership is designed to meet what they described as surging demand for AI infrastructure.
Amazon-Nvidia GPU Deal At A Glance
Amazon’s latest commitment is notable because AWS is simultaneously developing its own AI accelerators, including Trainium. Despite that push toward custom silicon, Amazon is substantially increasing its reliance on Nvidia’s accelerator ecosystem.
Key Details
| Particular | Details |
|---|---|
| Additional Nvidia GPUs | 2 million |
| Previous Nvidia GPU commitment | 1 million+ |
| Planned Nvidia GPU deployments | 3 million+ |
| Deployment period | 2027–2028 |
| Cloud platform | Amazon Web Services |
| GPU generations | Blackwell Ultra, Rubin, Rubin Ultra |
| Additional Nvidia technology | Vera CPUs, networking, software |
| Main demand drivers | Startups, enterprises, AI labs, governments |
| Financial terms | Not disclosed |
The two companies announced the expanded collaboration on August 26, 2026, during Nvidia’s quarterly earnings period.
Amazon Has More Than Tripled Its Planned Nvidia GPU Deployment
Amazon’s latest announcement is significant because of the speed at which its Nvidia commitment has expanded.
In April, Amazon announced plans to deploy more than 1 million Nvidia GPUs across AWS infrastructure beginning in 2026. Only a few months later, the companies announced another 2 million GPUs for 2027 and 2028.
If the earlier commitment is treated as approximately 1 million GPUs, Amazon’s planned Nvidia deployment has effectively increased to more than three times that level.
Earlier AWS Plan
1M+ Nvidia GPUs
↓
New Additional Commitment
2M Nvidia GPUs
↓
Planned Deployment
3M+ Nvidia GPUsNvidia said demand since the earlier agreement exceeded expectations, prompting the expansion.
Why Amazon Is Ordering So Many Nvidia Chips
The primary reason is the rapid increase in AI workloads running on AWS.
Customers are moving from AI experiments and pilot projects toward production deployments, which require significantly more computing power.
The companies highlighted demand from:
- AI startups
- Global enterprises
- Frontier AI laboratories
- Governments
- Agentic AI applications
- Scientific computing
- Robotics
- Physical AI
AWS said customers need infrastructure capable of scaling as their AI workloads become larger and more sophisticated.
AI Demand Cycle
More AI Applications
↓
More AI Users
↓
More Training + Inference
↓
Higher Compute Demand
↓
More Nvidia GPUs
↓
Larger AWS AI InfrastructureThe expansion therefore reflects not just Amazon’s own AI ambitions, but the computing requirements of AWS customers.
AWS AI Business Has Crossed $25 Billion Run Rate
Amazon’s own financial results provide context for the GPU expansion.
In its second-quarter 2026 results, Amazon said its AWS AI business had surpassed a $25 billion annual revenue run rate, growing at a triple-digit percentage rate year over year. Its chips business also exceeded a $25 billion annual revenue run rate.
AWS itself reported $42.2 billion in second-quarter sales, up 37% year over year.
Amazon’s AI And AWS Metrics
| Metric | Latest Figure |
|---|---|
| AWS Q2 2026 revenue | $42.2 Bn |
| AWS YoY growth | 37% |
| AWS AI annualized revenue run rate | >$25 Bn |
| Amazon chips annualized revenue run rate | >$25 Bn |
| Additional Nvidia GPUs | 2 million |
| Planned Nvidia GPU deployment | 3M+ |
The numbers show why Amazon is investing aggressively in AI infrastructure even while developing its own chips.
Nvidia Will Supply Multiple Generations Of GPUs
The expanded agreement covers several Nvidia architectures, including Blackwell Ultra, Rubin and Rubin Ultra GPUs.
This is important because Amazon is not simply buying one generation of hardware.
It is building a multi-year infrastructure roadmap that incorporates Nvidia’s current and next-generation systems.
Nvidia Hardware Roadmap For AWS
| Nvidia Technology | Role |
|---|---|
| Blackwell Ultra | Advanced AI computing |
| Rubin | Next-generation AI computing |
| Rubin Ultra | Future high-performance AI workloads |
| Vera CPU | AI server processing |
| Nvidia networking | Large-scale GPU cluster connectivity |
| CUDA-X software | AI and data processing |
The multi-generation approach should allow AWS to continue expanding AI capacity as Nvidia’s technology evolves.
Nvidia Vera CPUs Will Also Enter AWS
The partnership is expanding beyond GPUs.
Nvidia said AWS will also deploy its Vera CPUs, with some systems integrating Vera CPUs with Rubin and others using them separately.
The move gives Nvidia another opportunity to compete in the server CPU market, where Amazon has historically relied heavily on its own Graviton processors.
Nvidia GPUs
+
Nvidia Vera CPUs
+
Nvidia Networking
+
Nvidia Software
↓
Complete AI InfrastructureFor Nvidia, this represents an attempt to capture more value from every AI server rather than relying solely on accelerator sales.
Amazon Is Still Investing Heavily In Its Own AI Chips
The Nvidia expansion does not mean Amazon is abandoning its custom silicon strategy.
Amazon has been investing heavily in Trainium, its dedicated AI accelerator, as well as Graviton, its Arm-based CPU platform.
Amazon said its custom chip business has crossed a $25 billion annual revenue run rate and is growing at triple-digit rates.
The company has also said Trainium demand is strong, with Trainium2 fully subscribed and Trainium3 seeing significant demand.
Amazon’s Chip Strategy
| Chip Platform | Purpose |
|---|---|
| Nvidia GPUs | High-performance AI workloads |
| Trainium | Custom AI acceleration |
| Graviton | General-purpose cloud computing |
| Nitro | AWS infrastructure processing |
Amazon’s strategy is therefore becoming multi-platform rather than Nvidia-exclusive.
Why Amazon Needs Both Nvidia And Trainium
Different AI workloads have different cost and performance requirements.
Nvidia’s ecosystem offers broad compatibility with AI models and software, making its GPUs attractive for customers that need flexibility.
Amazon’s Trainium chips can offer AWS greater control over cost and infrastructure economics.
Customer Workload
↓
AWS Compute Options
↓
Nvidia GPUs ←→ Trainium
↓
Choose Hardware Based On
Performance + Cost + WorkloadProviding both options could help AWS compete against Microsoft Azure and Google Cloud, which are also investing heavily in custom AI silicon.
AWS And Nvidia Are Building More Than GPU Capacity
The latest partnership covers a much broader technology stack.
Nvidia said its networking technology, open models, CPUs, data-processing software and robotics platform will also be integrated across AWS.
This creates a more comprehensive relationship between the two companies.
Expanded AWS-Nvidia Partnership
| Technology | Planned Integration |
|---|---|
| Nvidia GPUs | 2M additional units |
| Vera CPUs | AWS infrastructure |
| NVLink Fusion | Advanced AI infrastructure |
| Networking | Large GPU clusters |
| Nemotron models | Amazon Bedrock and SageMaker |
| CUDA-X libraries | Data processing |
| Robotics platform | Amazon Robotics |
| Physical AI | Warehouse automation |
The agreement effectively brings Nvidia deeper into AWS’s broader AI infrastructure stack.
Nvidia Technology Will Power Amazon’s Robots
The partnership is also extending into physical AI.
Amazon Robotics plans to adopt Nvidia’s physical AI technologies, including:
- Omniverse
- Cosmos
- Isaac
- Jetson
These technologies are designed to support simulation, robotics development, world models and edge AI computing.
Amazon Robotics And Nvidia
Nvidia Omniverse
↓
Simulation
↓
Nvidia Cosmos
↓
World Models
↓
Nvidia Isaac
↓
Robot Development
↓
Jetson
↓
Physical RobotThis expands the partnership from data-center AI into Amazon’s physical operations.
Nvidia’s Open Models Will Come To AWS
AWS will also support Nvidia’s Nemotron family of open models through Amazon Bedrock and SageMaker.
This gives AWS customers another model option within Amazon’s managed AI services.
The move reflects the increasing importance of open and open-weight AI models alongside proprietary systems.
AWS AI Model Ecosystem
| Platform | Role |
|---|---|
| Amazon Bedrock | Managed foundation-model platform |
| SageMaker | AI development and deployment |
| Nvidia Nemotron | Open-model family |
| Amazon Nova | Amazon’s proprietary AI models |
| Anthropic Claude | Frontier AI models |
| Other third-party models | Multi-model choice |
The combination reinforces AWS’s strategy of allowing customers to select from a broad range of AI models and computing platforms.
Amazon’s AI Infrastructure Spending Is Already Huge
The Nvidia commitment comes as Amazon dramatically increases capital expenditure.
In its second-quarter results, Amazon said purchases of property and equipment had increased by $66.1 billion year over year, primarily reflecting investments in artificial intelligence. Its trailing-12-month free cash flow was a $7.6 billion outflow.
That demonstrates the financial cost of building AI infrastructure at scale.
Amazon AI Investment Indicators
| Metric | Figure |
|---|---|
| Year-over-year increase in property purchases | $66.1 Bn |
| Trailing-12-month free cash flow | -$7.6 Bn |
| AWS Q2 2026 revenue | $42.2 Bn |
| AWS Q2 growth | 37% |
| AWS AI revenue run rate | >$25 Bn |
Amazon is effectively spending heavily today to capture what it expects to be a much larger AI market in the future.
The GPU Deal Could Be Worth Tens Of Billions
Neither Amazon nor Nvidia disclosed financial terms for the additional 2 million GPUs.
However, given the volume of chips and associated infrastructure, the deal is expected to be worth tens of billions of dollars, according to TechCrunch.
The actual value could vary considerably depending on the GPU mix, networking equipment, system configuration, deployment schedule and pricing.
Therefore, the 2 million-unit figure should not be converted into a precise deal value without official pricing information.
Nvidia Is Benefiting From A Supply-Constrained Market
Nvidia’s latest earnings commentary indicates that AI demand remains stronger than the company’s ability to supply all potential customers.
The company expects approximately 70% revenue growth in fiscal 2028, describing the outlook as supply constrained.
That makes large long-term commitments from hyperscalers particularly important.
Amazon’s expanded order helps Nvidia secure a major customer commitment while Amazon gets greater visibility into future GPU availability.
Nvidia Has Committed $279 Billion To Secure Supply
Nvidia is also responding to the supply challenge by securing future manufacturing and component capacity.
The company has committed approximately $279 billion to secure supply and manufacturing capacity for current and future data-center projects, according to reporting following its latest earnings results.
That figure includes projected spending across the remainder of the current fiscal year and fiscal 2028.
The strategy reflects the industry’s broader shift toward long-term supply commitments.
Amazon’s Own Chips Still Have Strong Demand
Amazon’s custom silicon strategy is progressing rapidly despite the Nvidia expansion.
Amazon previously said Trainium2 was fully subscribed, with 1.4 million chips landed, while Trainium3 was seeing strong demand and nearly all supply expected to be committed by mid-2026.
Amazon also said OpenAI had committed to consume approximately 2 GW of Trainium capacity beginning in 2027, while Anthropic would secure up to 5 GW of current and future Trainium capacity.
This means Amazon is simultaneously becoming:
- A major Nvidia customer.
- A major custom AI chip developer.
- A large AI cloud provider.
- A major infrastructure provider for frontier AI companies.
Amazon Is Building A Multi-Chip AI Cloud
The combination of Nvidia and Amazon’s own accelerators could become a competitive advantage for AWS.
Customers increasingly want different hardware options depending on workload economics.
AWS can potentially offer:
Nvidia
↓
Maximum Ecosystem Compatibility
Trainium
↓
AWS-Optimized AI Economics
Graviton
↓
General-Purpose Computing
Cerebras / Other Accelerators
↓
Specialized InferenceAmazon has already expanded its AI infrastructure ecosystem to include multiple accelerator options.
Why The Deal Matters For Nvidia
For Nvidia, Amazon’s expansion is a strong validation of continued demand for its accelerators.
AWS is one of the world’s largest cloud providers and operates infrastructure at enormous scale.
If Amazon is increasing its Nvidia GPU commitment despite investing heavily in competing chips, it suggests customers continue to value Nvidia’s performance, software ecosystem and compatibility.
Why The Deal Matters For Amazon
For Amazon, Nvidia provides access to an established AI computing ecosystem that can be deployed rapidly across AWS.
Its own chips can help optimize cost and performance, but Nvidia GPUs remain important for customers that want broad software compatibility and access to the company’s mature AI ecosystem.
The expanded relationship gives AWS greater flexibility while maintaining Nvidia as a critical component of its infrastructure.
The Bigger Picture
Amazon’s decision to add 2 million Nvidia GPUs to AWS infrastructure is one of the clearest signs yet that demand for AI computing continues to exceed earlier expectations. The commitment comes only months after Amazon announced plans for more than 1 million Nvidia GPUs, effectively taking its planned Nvidia deployment to more than 3 million units. The chips will include Nvidia’s Blackwell Ultra, Rubin and Rubin Ultra architectures and are expected to be deployed across AWS infrastructure during 2027 and 2028.
The more important story, however, is that Amazon is increasing Nvidia spending while simultaneously expanding its own Trainium and Graviton chip businesses. AWS’s AI revenue run rate has surpassed $25 billion, while its custom chip business has also crossed $25 billion. Amazon therefore appears to be pursuing a dual strategy: use Nvidia’s ecosystem to meet enormous near-term AI demand while developing proprietary silicon that could improve AWS’s long-term cost and performance economics.
Looking Ahead
The next phase of the AWS-Nvidia relationship will depend on how quickly Amazon can deploy the additional 2 million GPUs and how efficiently customers monetize the resulting computing capacity. Nvidia’s Vera Rubin generation should become increasingly important during this period, while Amazon’s Trainium platform will continue competing for workloads where custom silicon can provide better economics. The coexistence of both platforms suggests AWS is unlikely to pursue a single-chip strategy.
For Nvidia, the Amazon expansion provides another major source of long-term demand at a time when the company says AI infrastructure supply remains constrained. For Amazon, the deal strengthens AWS’s ability to serve frontier AI labs, enterprises and startups while giving customers access to Nvidia’s latest computing architecture. With AWS AI revenue already above a $25 billion annual run rate and AI workloads moving rapidly from experimentation into production, the 2 million-GPU expansion could become an important component of Amazon’s next stage of AI growth
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
