Anthropic has agreed to spend $45 billion over six years to rent artificial intelligence computing capacity from Nscale’s planned data center campus in West Virginia, according to reports citing people familiar with the agreement. The deal covers approximately 460 megawatts of power capacity and is designed to give the Claude developer additional infrastructure as demand for its AI services, including Claude Code, continues to rise.
Nscale plans to deploy Nvidia’s next-generation Vera Rubin systems to support Anthropic’s workloads, with the capacity expected to begin coming online in late 2027. The agreement is another example of the enormous infrastructure commitments now being made by frontier AI companies, which are racing to secure GPUs, data-center capacity and electricity well ahead of anticipated demand.
Anthropic-Nscale Deal At A Glance
The six-year agreement represents one of Anthropic’s largest disclosed commitments to AI computing infrastructure.
Importantly, Anthropic is renting computing capacity rather than buying Nscale’s data center outright. The arrangement gives the AI company access to infrastructure while allowing Nscale to build and operate the facility.
Key Deal Details
| Particular | Details |
|---|---|
| AI company | Anthropic |
| Infrastructure provider | Nscale |
| Reported commitment | $45 billion |
| Contract duration | 6 years |
| Power capacity | ~460 MW |
| Location | West Virginia, U.S. |
| AI chips | Nvidia Vera Rubin |
| Expected deployment | Late 2027 |
| Main demand driver | Claude and Claude Code |
| Deal structure | Rented AI computing capacity |
The companies had not publicly confirmed all commercial terms when the reports emerged, with Reuters reporting that Anthropic declined to comment on the agreement.
$45 Billion Compute Commitment Explained
At the headline level, the agreement works out to an average of approximately $7.5 billion per year over six years.
That does not necessarily mean Anthropic will pay exactly $7.5 billion every year because the actual payment schedule can vary according to deployment and capacity.
Approximate Contract Economics
| Calculation | Amount |
|---|---|
| Total reported commitment | $45 billion |
| Duration | 6 years |
| Simple annual average | $7.5 billion |
| Simple monthly average | ~$625 million |
| Power capacity | ~460 MW |
The scale demonstrates how quickly AI infrastructure spending is moving beyond traditional cloud-computing budgets.
Why Anthropic Needs So Much Computing Power
Anthropic is expanding the capacity available for both AI model training and inference.
Training requires large clusters of accelerators to develop increasingly capable models. Inference requires computing resources every time users interact with those models.
As Claude becomes more widely used, the amount of inference required can rise rapidly.
More Claude Users
↓
More AI Queries
↓
More Inference
↓
Higher Compute Demand
↓
More Data Center CapacityClaude Code is particularly important because agentic coding workloads can involve multiple steps, reasoning processes and interactions with large codebases.
Claude Code Is A Major Demand Driver
Anthropic has been aggressively expanding its computing footprint as it anticipates increasing demand for products such as Claude Code, its AI-powered programming tool.
Unlike a basic chatbot interaction, coding agents can perform extended tasks that require repeated model calls.
That means greater adoption can translate into significantly higher computing requirements.
Claude Code Growth Cycle
| Stage | Impact |
|---|---|
| More developers adopt Claude Code | Higher usage |
| Longer coding sessions | More inference |
| More complex tasks | Higher compute per task |
| Enterprise adoption | Larger workloads |
| Greater usage | More infrastructure required |
This helps explain why Anthropic is securing computing capacity years before the new Nscale facility is expected to become operational.
Nscale’s West Virginia Campus
The computing capacity will come from Nscale’s major data center development in Mason County, West Virginia.
The broader campus is planned to reach approximately 1.35 gigawatts of capacity, making the 460 MW Anthropic commitment a substantial portion of the initial development.
The site is being developed specifically around the requirements of high-density AI computing.
Nscale Campus
| Feature | Details |
|---|---|
| Location | Mason County, West Virginia |
| Planned campus capacity | ~1.35 GW |
| Anthropic allocation | ~460 MW |
| Anthropic share of planned capacity | ~34% |
| Planned AI hardware | Nvidia Vera Rubin |
| Expected Anthropic deployment | Late 2027 |
The size of the campus reflects the growing physical requirements of frontier AI.
Nvidia Vera Rubin Chips Will Power The Deal
Nscale plans to use Nvidia Vera Rubin systems for Anthropic’s computing requirements.
Vera Rubin represents Nvidia’s next-generation AI infrastructure platform and is designed for high-performance AI training and inference.
The deployment timeline is significant because Anthropic is effectively reserving access to future-generation computing capacity before the hardware is broadly deployed.
Anthropic
↓
Nscale Data Center
↓
Nvidia Vera Rubin
↓
AI Compute
↓
Claude Training + InferenceThe agreement therefore connects three major parts of the AI infrastructure chain: the model developer, the specialized cloud provider and the accelerator supplier.
AI Infrastructure Is Becoming A Race For Capacity
Anthropic’s deal reflects a broader shift in the AI industry.
The competition is no longer limited to developing better AI models. Companies increasingly need to secure the physical infrastructure required to train and operate those models.
That includes:
- AI accelerators
- High-bandwidth memory
- Data centers
- Electricity
- Cooling systems
- Networking
- Storage
- Cloud capacity
The New AI Infrastructure Race
Better AI Models
↓
Higher User Demand
↓
More Compute
↓
More GPUs
↓
More Data Centers
↓
More Electricity
↓
Long-Term Infrastructure ContractsAnthropic’s $45 billion agreement represents this shift toward long-term capacity planning.
Anthropic Is Building A Multi-Provider Compute Network
Nscale is only one part of Anthropic’s expanding infrastructure strategy.
The company has also entered major arrangements involving Amazon, Google, SpaceX, AMD and other infrastructure providers. Reuters reported that Anthropic has been spending heavily on GPUs, computing capacity, model training, inference and hiring as it prepares for further growth.
This multi-provider strategy gives Anthropic access to different types of computing infrastructure.
Anthropic’s Expanding Compute Ecosystem
| Provider | Role |
|---|---|
| Nscale | AI computing capacity |
| Amazon | Cloud and AI infrastructure |
| AI accelerators and cloud capacity | |
| SpaceX | Large-scale AI data-center capacity |
| AMD | AI servers and accelerator infrastructure |
| Nvidia | GPUs and Vera Rubin systems |
The exact terms and values of these arrangements differ and should not be treated as directly comparable.
SpaceX Deal Adds To Anthropic’s Compute Commitments
Anthropic has already committed substantial resources to SpaceX’s AI infrastructure.
According to SpaceX’s IPO filing, Anthropic agreed to pay $1.25 billion per month through May 2029 for computing capacity that includes SpaceX’s Colossus and Colossus II AI training clusters.
Anthropic also agreed in May to rent the computing power of SpaceX’s Colossus 1 facility in Memphis, which houses more than 220,000 Nvidia processors and represents approximately 300 MW of new capacity.
This shows that the Nscale contract is part of a much larger effort to secure future computing supply.
Anthropic’s Revenue Expectations Are Also Rising
The infrastructure spending comes as Anthropic projects extraordinary revenue growth.
Reuters reported that Anthropic is projecting 2028 revenue of approximately $190 billion to $200 billion, compared with a current revenue run rate of around $47 billion.
Anthropic Revenue Outlook
| Metric | Figure |
|---|---|
| Current revenue run rate | ~$47 billion |
| 2028 projected revenue | $190–200 billion |
| Midpoint | ~$195 billion |
| 2028 revenue vs. current run rate | ~4.1x |
The company would need a dramatic expansion in customer adoption and AI usage to reach those projections.
The need for additional compute is therefore directly connected to Anthropic’s growth expectations.
AI Compute Is Becoming A Strategic Asset
For companies such as Anthropic, computing capacity is increasingly similar to a strategic raw material.
A company may have strong models and customers but still face growth constraints if it cannot access enough GPUs and data-center capacity.
This makes long-term contracts valuable.
AI Model Capability
+
Customer Demand
+
Compute Availability
↓
AI Business GrowthWithout sufficient compute, even strong demand can translate into slower service availability or capacity constraints.
Why Nscale Benefits From The Deal
The agreement is also significant for Nscale.
A six-year commitment from a major AI company provides long-term visibility into demand for its data-center capacity.
It can also strengthen the economics of building a highly capital-intensive AI facility.
Potential Benefits For Nscale
| Benefit | Impact |
|---|---|
| Six-year contract | Revenue visibility |
| 460 MW commitment | Large capacity utilization |
| Anthropic as customer | Industry validation |
| Nvidia Vera Rubin deployment | Advanced infrastructure |
| Long-term demand | Supports financing and expansion |
Nscale has been expanding rapidly as demand grows for specialized AI cloud infrastructure.
The Data Center Power Challenge
The 460 MW figure is important because AI infrastructure is increasingly constrained by electricity availability.
A large AI data center requires power not only for GPUs but also for networking, cooling and other infrastructure.
As AI models become larger and inference workloads increase, data-center operators need access to reliable, high-capacity power.
AI Data Center Requirements
| Infrastructure | Why It Matters |
|---|---|
| Electricity | Powers AI accelerators |
| Cooling | Manages heat from high-density systems |
| Networking | Connects large GPU clusters |
| Storage | Supports AI datasets and models |
| Data center space | Houses computing equipment |
| Power generation | Ensures reliable operations |
The West Virginia project is therefore an infrastructure investment as much as a technology project.
Nscale Is Building Around Nvidia Hardware
The use of Vera Rubin systems also highlights how specialized AI cloud providers are increasingly building infrastructure around particular accelerator generations.
For Anthropic, this could provide access to high-performance Nvidia systems without having to own and operate the entire data center.
For Nscale, deploying the latest Nvidia architecture can make its capacity attractive to frontier AI customers.
The Deal Comes Ahead Of Anthropic’s IPO Plans
The timing is another important part of the story.
Anthropic is preparing for a potential initial public offering, according to Reuters and other reports.
Securing large amounts of future compute could help demonstrate to prospective investors that the company is preparing for substantial future demand.
At the same time, these agreements represent major financial commitments.
IPO Perspective
| Positive | Risk |
|---|---|
| Secured future compute | Large fixed commitments |
| Capacity for growth | Utilization risk |
| Supports Claude expansion | Higher infrastructure costs |
| Strong demand signal | AI pricing could decline |
| Long-term infrastructure | Technology can evolve quickly |
Investors will therefore need to evaluate both sides of the equation.
Anthropic Faces A Capital-Intensive Growth Model
Traditional software companies can often increase revenue without proportionally increasing physical infrastructure.
Frontier AI is different.
As usage increases, companies need additional computing capacity.
That creates a business model where growth can require enormous infrastructure spending.
Revenue Growth
↓
More AI Usage
↓
More Compute
↓
Higher Infrastructure Costs
↓
Need For More Capital
↓
More Compute Capacity
↓
Higher Potential RevenueThe challenge is ensuring that revenue grows faster than the cost of operating the infrastructure.
Anthropic Is Not Relying On One Chip Supplier
Anthropic’s infrastructure strategy is also notable for its hardware diversification.
The company has access to Nvidia GPUs, Google TPUs and Amazon Trainium, while its AMD relationship adds another source of AI server capacity.
This could reduce the risks associated with relying entirely on a single accelerator supplier.
At the same time, the Nscale agreement demonstrates that Nvidia remains a critical part of Anthropic’s long-term infrastructure strategy.
AI Companies Are Locking In Capacity Years Ahead
The Nscale deal illustrates how far ahead AI companies now have to plan.
The capacity is not expected to begin coming online until late 2027, yet Anthropic is committing to it now.
This is because data centers require years of planning, construction, power procurement and hardware deployment.
AI Capacity Planning Timeline
2026
Contract Signed
↓
2026–27
Data Center Construction
↓
Late 2027
Vera Rubin Systems Begin Deployment
↓
2028
Expanded AI Capacity
↓
Future
Claude ScalingLong lead times make early capacity reservations increasingly important.
The Bigger Picture
Anthropic’s reported $45 billion agreement with Nscale is a clear sign of how quickly the economics of frontier AI are changing. The six-year contract covers approximately 460 MW of computing capacity at Nscale’s West Virginia campus and will use Nvidia Vera Rubin systems. The capacity is expected to begin coming online in late 2027, giving Anthropic additional infrastructure to support Claude, Claude Code and other AI workloads.
The deal also illustrates the broader AI infrastructure race. Anthropic is simultaneously securing capacity from multiple providers as it anticipates substantial growth in demand. Reuters has reported that the company is targeting roughly $190–200 billion in revenue in 2028, compared with a current revenue run rate of around $47 billion. Reaching that scale would require not only stronger customer adoption but also enormous amounts of computing capacity, making long-term infrastructure agreements strategically important.
Looking Ahead
The next major milestone will be the deployment of Nvidia Vera Rubin systems at Nscale’s West Virginia facility. Anthropic’s decision to reserve approximately 460 MW years ahead of full deployment suggests that the company expects AI compute demand to remain constrained for an extended period. If Claude and Claude Code adoption continues to accelerate, the additional capacity could help Anthropic expand without facing the infrastructure shortages that have affected AI services during periods of heavy demand.
The larger question is whether the economics of AI can support infrastructure commitments of this scale. Anthropic must generate enough revenue from Claude and its enterprise and developer products to justify billions of dollars in annual computing costs. For Nscale, meanwhile, the agreement provides a major long-term customer as it builds one of the world’s increasingly important AI infrastructure businesses. Together, the deal shows that the AI race is becoming a competition not only for better models, but also for GPUs, data centers, power and long-term access to computing capacity
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
