AnswersPublished20 sources
AWS and NVIDIA’s 2 Million-GPU Expansion: What It Means for AI Infrastructure
AWS and NVIDIA plan to deploy 2 million additional NVIDIA GPUs across AWS’s global infrastructure in 2027–2028, following a previous plan for more than 1 million GPUs from 2026. The planned fleet includes NVIDIA Blackwell Ultra, Rubin and Rubin Ultra GPUs, alongside Vera CPU based systems, networking and rack scale...
Published byEdited with GPT-5.6 LunaImages generated with GPT Image 1.5
Research answer
Key takeaways
- AWS and NVIDIA plan to deploy 2 million additional NVIDIA GPUs across AWS’s global infrastructure in 2027–2028, following a previous plan for more than 1 million GPUs from 2026.
- The planned fleet includes NVIDIA Blackwell Ultra, Rubin and Rubin Ultra GPUs, alongside Vera CPU based systems, networking and rack scale infrastructure for agentic and physical AI.
- AWS is presenting NVIDIA hardware as complementary to its own Trainium accelerators, using expanded NVLink Fusion and custom high bandwidth memory work to offer customers more infrastructure choices.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Amazon Web Services and NVIDIA announce on August 27, 2026, regarding AWS’s planned deployment of 2 million additional NVIDIA GPUs. Article summary: AWS and NVIDIA announced a major expansion of their AI-infrastructure partnership: AWS plans to deploy 2 million additional NVIDIA GPUs globally in 2027–28, on top of more than 1 million NVIDIA GPUs it had already planne. Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fak
AWS and NVIDIA are expanding their long-running AI-infrastructure partnership with plans for AWS to deploy 2 million additional NVIDIA GPUs across its global infrastructure in 2027 and 2028. The commitment follows AWS’s earlier plan to add more than 1 million NVIDIA GPUs beginning in 2026, as demand for AI training and inference continues to rise. 3
6
9
The announcement is significant not only because of the GPU count. It outlines a broader, multigeneration infrastructure strategy spanning accelerators, CPUs, networking, AI models, data processing, robotics and secure government workloads.
The headline: 2 million more NVIDIA GPUs
The additional GPUs are expected to include NVIDIA’s Blackwell Ultra, Rubin and Rubin Ultra platforms. That makes the plan a forward-looking capacity commitment across several generations of NVIDIA hardware rather than a purchase centered on one accelerator family. 1
6
8
AWS and NVIDIA have not disclosed the financial value of the agreement. The 2 million figure should therefore be read as a planned deployment across AWS infrastructure, not as a confirmed number of GPUs already installed or immediately available to customers. 14
The scale also shows how quickly cloud providers are planning for AI workloads. The announcement came only months after AWS had announced plans for more than 1 million NVIDIA GPUs starting in 2026; reporting on the expansion linked the new commitment to customer demand exceeding earlier expectations. 3
9
More than a GPU supply agreement
AWS and NVIDIA described the expanded collaboration as an effort to build integrated AI infrastructure. In addition to GPUs, the partnership covers NVIDIA Vera CPU-based systems, networking, open models, data processing, robotics and AI factories. 1
6
That matters because large AI deployments depend on more than accelerator chips. CPUs feed and coordinate workloads, high-speed networking connects processors, and storage and data-processing systems help move information through training and inference pipelines. The companies are positioning their work as a coordinated platform for scaling AI rather than simply a larger catalog of cloud GPU instances.
The target workloads include agentic AI, in which software systems perform multistep tasks, and physical AI, including robotics-related applications. The sources describe these areas as central reasons for expanding the infrastructure collaboration. 6
8
NVIDIA Vera extends the partnership beyond accelerators
The planned addition of NVIDIA Vera CPU-based infrastructure gives AWS another NVIDIA component for agentic-AI systems. Vera is presented as part of the broader infrastructure stack, alongside NVIDIA’s GPUs and networking technologies. 10
11
For customers, the practical implication is that future AI instances may be designed around complete CPU-GPU systems rather than treating the GPU as an isolated component. That approach can be useful for workloads that require substantial orchestration, data movement and interaction between models and external tools.
Trainium remains part of AWS’s strategy
The expansion does not mean AWS is abandoning its in-house AI chips. AWS continues to develop and offer Trainium, while also expanding its work with NVIDIA. The companies said their collaboration would extend NVLink Fusion and include custom NVIDIA high-bandwidth memory for Trainium-based systems. 11
13
This points to a coexistence strategy: AWS can offer NVIDIA-based infrastructure for customers seeking NVIDIA’s software and hardware ecosystem while continuing to develop Trainium for workloads optimized around AWS silicon. The partnership is therefore better understood as a choice-oriented model than as an either-or decision between NVIDIA GPUs and Trainium.
For customers, that could mean more options when balancing performance, software compatibility, availability and infrastructure cost. The available announcements establish the direction of the strategy, but they do not provide enough detail to compare the eventual economics or performance of every configuration.
Secure infrastructure for government workloads
AWS and NVIDIA also described secure AI-factory infrastructure for U.S. government and national-security workloads. Coverage of the announcement characterizes these deployments as aimed at secure federal use cases, in addition to commercial AI infrastructure. 8
10
The government focus broadens the partnership’s market beyond public cloud customers building consumer or enterprise applications. It suggests that the companies are also targeting workloads where security, access controls and deployment requirements are central considerations. The provided reporting does not establish all implementation details for these environments, so the announcement should not be treated as a complete description of their eventual technical or accreditation status.
Why the expansion matters
The strategic logic is straightforward. NVIDIA gains a major cloud distribution channel for multiple generations of its AI platforms, while AWS gains additional capacity and a wider set of NVIDIA-integrated systems to offer customers. 2
6
The deal also gives AWS a way to present itself as a flexible AI-infrastructure provider. Customers can use NVIDIA-optimized systems, AWS Trainium-based systems or configurations that connect elements of both ecosystems. That flexibility may become more important as AI workloads diversify from large-model training into inference, autonomous agents, data processing and robotics.
Still, the announcement is a plan, not proof that all 2 million GPUs will be online by the end of 2028. Availability will depend on hardware production, data-center construction, power, networking and customer demand. The sources provided confirm the companies’ intended scale and scope, but do not quantify the rollout by region, product or individual AWS service.
The bottom line
AWS and NVIDIA announced plans for 2 million additional NVIDIA GPUs in 2027–2028, building on AWS’s previously announced 2026 deployment of more than 1 million GPUs. The expanded partnership includes Blackwell Ultra, Rubin and Rubin Ultra, as well as Vera CPUs, networking, AI factories, data processing and robotics infrastructure. 1
3
6
The larger message is that cloud AI competition is moving toward complete, rack-scale systems. AWS is combining NVIDIA’s broad platform with its own Trainium roadmap, while NVIDIA is extending its reach across the infrastructure layers needed to run agentic and physical AI at scale.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "AWS and NVIDIA’s 2 Million-GPU Expansion: What It Means for AI Infrastructure"?
AWS and NVIDIA plan to deploy 2 million additional NVIDIA GPUs across AWS’s global infrastructure in 2027–2028, following a previous plan for more than 1 million GPUs from 2026.
What are the key points to validate first?
AWS and NVIDIA plan to deploy 2 million additional NVIDIA GPUs across AWS’s global infrastructure in 2027–2028, following a previous plan for more than 1 million GPUs from 2026. The planned fleet includes NVIDIA Blackwell Ultra, Rubin and Rubin Ultra GPUs, alongside Vera CPU based systems, networking and rack scale infrastructure for agentic and physical AI.
What should I do next in practice?
AWS is presenting NVIDIA hardware as complementary to its own Trainium accelerators, using expanded NVLink Fusion and custom high bandwidth memory work to offer customers more infrastructure choices.
Sources
techcrunch.comAmazon just tripled its order of Nvidia chips over 'surging ...
nvidianews.nvidia.comAWS and NVIDIA to Deliver 2 Million Additional GPUs ...
fortuneindia.comAWS to deploy 2 million more Nvidia GPUs as AI demand runs ahead of expectations
finance.yahoo.comAmazon and Nvidia expand AI partnership with 2 million GPUs
