Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteAmazon’s Project Rainier is operational in northern Indiana, where a roughly $11 billion data-center investment supports Anthropic’s Claude AI models with Amazon-designed Trainium2 chips. The headline describes an Indiana campus, but Rainier is a broader, multi-site U.S. compute program—not one building holding every chip associated with it. The project is a significant investment in AI infrastructure, but it does not mean Anthropic has stopped using Nvidia or that the full planned campus is already consuming power.
What Amazon opened in Indiana
Project Rainier’s Indiana campus is in or near New Carlisle in St. Joseph County. The project has been reported as an approximately $11 billion investment on a site of about 1,200 acres. Amazon said Rainier was operational on October 29, 2025, and that Anthropic was using it to train and deploy Claude models. Amazon’s Project Rainier announcement describes a distributed cluster spanning multiple U.S. data centers; the Indiana campus is part of that wider system.
That distinction matters when reading chip-count claims. Amazon has described nearly half a million Trainium2 chips in connection with Project Rainier, but that does not establish that all those chips are physically installed in Indiana. Amazon has also said Anthropic is expected to use more than one million Trainium2 chips across its workloads. That is a broader deployment statement, not a count for this campus.
Why Anthropic is the anchor customer
Rainier is designed around a substantial customer relationship, not merely an equipment demonstration. Anthropic uses AWS as its primary cloud and training partner, and the companies work together to optimize Anthropic’s workloads for Trainium. Amazon’s cloud services, custom chips, and investment in Anthropic reinforce one another: AWS gains a prominent large-scale customer for its silicon, while Anthropic gains access to compute built for its Claude training and inference needs.
Free tools Windows power users keep installed
One-click scans. No signup required.
#1 Best Overall
The investment figures are separate from the Indiana construction figure. Amazon’s cumulative investment in Anthropic reached $8 billion by November 2024. Amazon later announced an additional $5 billion investment, with the possibility of up to $20 billion more under specified conditions. Those are investments in Anthropic, not the cost of the Indiana campus. Anthropic’s 2024 partnership announcement describes the AWS collaboration and $8 billion total at that time; Amazon’s later announcement describes the additional investment.
What Trainium2 does—and what the specifications tell you
Trainium2 is Amazon’s accelerator for AI training and inference. Rather than being sold as a typical retail component, it is accessed through AWS infrastructure. AWS pairs its Trainium instances with the Neuron software stack, which helps compile and optimize workloads, and Elastic Fabric Adapter (EFA) networking for communication across distributed systems.
A public AWS Trn2.48xlarge instance has 16 Trainium2 chips, 1.5 TB of accelerator memory, 192 vCPUs, 2 TB of system memory, and up to 3.2 Tbps of network bandwidth, according to AWS’s EC2 accelerated-computing specifications. These are specifications for that public instance type, not confirmation of the exact server configuration at Rainier.
For a large AI cluster, a chip count alone does not determine useful performance. Memory, networking, software support, and the ability to keep accelerators working efficiently all matter. A model may run on Trainium but still need engineering work or perform differently from the same workload on another architecture.
Why Amazon is building its own AI chips
- More control over supply: Custom silicon gives AWS another source of accelerator capacity instead of relying entirely on Nvidia hardware.
- Workload-specific economics: Amazon can design systems around workloads it understands and optimize chip, server, network, and software together. Whether that lowers costs depends on the workload and how effectively the system is used.
- A differentiated cloud offering: Trainium gives AWS an alternative to simply offering third-party accelerators, potentially attracting customers prepared to optimize for AWS.
- Deeper customer integration: A large deployment can tie a customer’s software, training processes, and infrastructure more closely to AWS. That can be useful, but it can also make moving workloads more demanding.
Amazon reported that Trainium and Graviton together exceeded a $10 billion annual revenue run rate and that Trainium2 was fully subscribed. Those are company-reported figures, not independent measurements of market share or proof that Trainium is cheaper or faster for every workload. Amazon’s 2026 results also discussed production use of Trainium3. Amazon’s investor-relations results provide the company’s reported figures and update.
How Rainier changes the Nvidia story
Rainier is a major example of a cloud provider deploying its own AI accelerators at scale for a leading model developer. It increases competitive pressure on Nvidia and shows that a hyperscaler can coordinate chips, networking, software, and cloud delivery around a particular customer’s needs. It is not evidence that Nvidia has been displaced.
Anthropic has said that it uses AWS Trainium, Google TPUs, and Nvidia GPUs. Its multi-platform strategy means Rainier is one important source of compute rather than an exclusive commitment to Amazon’s architecture. For companies considering Trainium, the relevant comparison is not a chip specification in isolation: it is the performance and total cost of their own model, software, engineering effort, and required scale on the available systems. Anthropic’s compute partnership announcement describes its use of the three platforms.
What the scale figures do—and do not—mean
| Measure | What is reported | How to interpret it |
|---|---|---|
| Indiana investment | About $11 billion | The reported investment for the Project Rainier Indiana project; it is not Amazon’s investment in Anthropic. |
| Site area | About 1,200 acres | Reported area for the New Carlisle campus. |
| Project chips | Nearly half a million Trainium2 chips | Amazon’s figure for Rainier; it does not confirm that all chips are located in Indiana. |
| Anthropic chip use | More than one million Trainium2 chips | Amazon’s broader statement about Anthropic workloads, not the Indiana campus alone. |
| Power capacity | About 2.2 GW planned or ultimate capacity | A reported planning figure for the campus, not evidence of present electricity consumption. |
| Construction timing | Amazon said the broader Rainier infrastructure was deployed in less than one year after announcement | This describes the project’s rapid deployment, not proof that every planned campus phase was completed in that period. |
Amazon has also said Rainier provides more than five times the compute power Anthropic used for earlier models. That is Amazon’s comparison and should not be read as an independent, universal benchmark across AI systems or workloads. The available figures describe different scopes—site, broader cluster, customer usage, and planned power—so they should not be added together or treated as measurements of one Indiana building.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What Indiana gains, and what communities should scrutinize
A large data-center campus changes more than the local tax base or construction skyline. Converting farmland to industrial use can affect land patterns, roads, and nearby development. Construction activity can bring short-term employment and supplier demand; permanent staffing and the public cost of supporting infrastructure are separate questions that require local, project-specific evidence.
Electricity is a central issue for a campus planned at this scale. A reported 2.2 GW ultimate capacity is a planning figure, not the site’s current draw. Residents and policymakers should distinguish announced capacity from actual demand, ask how new generation and transmission are funded, and examine whether utility arrangements shift costs indirectly. Water use and cooling requirements, tax incentives, road improvements, and the number and duration of local jobs also merit scrutiny. The available project figures alone do not establish whether local economic benefits outweigh public costs.
Amazon later announced a separate $15 billion northern Indiana data-center expansion, estimated to add 2.4 GW of capacity. It is additional to—not part of—the original $11 billion Rainier investment. Amazon says its subsidiary will pay fees associated with use of existing power lines and cover costs for new generation, transmission, or related equipment needed to serve its facilities. That is Amazon’s announced arrangement; it does not by itself prove that residents face no indirect system costs. Amazon’s expansion announcement describes the separate investment and its stated power-cost commitments.
Quick Recap
What to watch next
- Whether Trainium generalizes: Rainier is a powerful reference deployment, but performance and cost for other customers will depend on software support, workload fit, and access to capacity.
- Whether Anthropic keeps diversifying: Its continued use of Google TPUs and Nvidia GPUs gives it options and reduces reliance on one accelerator platform.
- Whether planned Indiana capacity is built and used: Announced campus capacity should be tracked separately from completed facilities and actual electricity consumption.
- Whether custom silicon becomes easier to adopt: The Neuron toolchain and engineering requirements will influence whether customers beyond a deeply integrated partner can use Trainium effectively.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →
