Anthropic is making a very clear bet: the biggest risk to its growth may not be demand for Claude, but having enough computing power to keep up with that demand.
The company has agreed to spend roughly $45 billion over six years on computing capacity from Nscale, securing about 460 megawatts of power at a planned data center campus in West Virginia.
The deal is enormous. But the more interesting story is what it says about the AI infrastructure race, Anthropic’s IPO ambitions, and the growing importance of Nvidia, AMD and custom AI chips.
$45 billion just to secure computing power
The Nscale agreement covers the first building at its planned Monarch campus in West Virginia.
The key numbers:
- $45 billion in total spending over six years
- 460 MW of computing capacity
- Nvidia Vera Rubin systems
- Expected to come online in late 2027
- Roughly $7.5 billion a year on average
To put 460 MW into perspective, the capacity is enough to power roughly 345,000 US homes at once.
That gives you a sense of how different the economics of AI infrastructure have become.
The AI race is no longer simply about who has the best model. It is increasingly about who can secure enough chips, electricity, data centers and networking capacity to actually run those models at scale.
Google and Microsoft wanted the same capacity
There is another interesting layer to the deal.
Before Anthropic secured the Nscale capacity, Google and Microsoft were reportedly in talks with Nscale as well.
Microsoft eventually walked away during a review of its data center portfolio and power generation needs.
Google’s discussions also did not result in a deal after an internal review of its computing requirements and capital spending.
Anthropic ultimately took the capacity.
That tells you something about the intensity of the current compute scramble. The biggest technology companies are competing for the same infrastructure that AI labs need to keep scaling.
Anthropic is preparing for a massive IPO story
The timing is difficult to ignore.
Anthropic has reportedly filed confidentially for an IPO and is preparing to meet potential investors.
The company is expected by some reports to target a valuation of $2 trillion or more, although that valuation is far from established.
And this is where the Nscale deal becomes particularly important.
When Anthropic eventually pitches itself to public-market investors, one of the obvious questions will be:
Can Claude keep growing without Anthropic running out of computing capacity?
The company is effectively trying to answer that question in advance.
Instead of waiting until demand overwhelms its infrastructure, Anthropic is locking in capacity years ahead.
That comes with a huge cost, but it also gives the company something valuable in the AI race: visibility into future computing capacity.
Anthropic is not betting on just one chipmaker
One of the most interesting parts of this story is that Anthropic is spending heavily across multiple AI hardware platforms.
The company has commitments involving:
- Nvidia
- AMD
- Amazon’s Trainium
- SpaceX-linked capacity
- Nscale
Its latest Nscale deal will use Nvidia’s next-generation Vera Rubin systems.
But just weeks earlier, Anthropic had also agreed to a major AMD arrangement involving up to 2 gigawatts of Instinct MI450 capacity, beginning in 2027.
AMD also agreed to invest up to $5 billion in Anthropic as certain milestones are reached.
So the Nvidia versus AMD debate is not quite as simple as picking a winner.
Anthropic wants both.
And that makes sense.
For an AI company expecting enormous growth, depending entirely on one hardware supplier creates its own risk.
Having multiple platforms gives Anthropic more options on capacity, cost and performance.
The custom-chip angle could become even bigger
Anthropic is also exploring another way to reduce its dependence on external processors.
Reports indicate the company was in discussions to acquire MatX, an AI chip startup founded by former Google TPU engineers, in a deal that could have been worth almost $7 billion.
Those discussions reportedly shifted toward a possible partnership.
The significance goes beyond one potential acquisition.
Anthropic is already building an internal silicon team and hiring experienced chip engineers.
Why?
Because custom silicon could eventually give AI companies more control over cost, performance and infrastructure design.
That could become increasingly important as AI models become more expensive to train and operate.
The spending is happening because demand is enormous
Anthropic’s infrastructure commitments make more sense when you look at the growth numbers being reported.
The company reportedly reached an annualized revenue run rate of around $65 billion by July, up sharply from roughly $47 billion in May.
It also reportedly generated $11.6 billion in quarterly revenue in the second quarter, more than twice its first-quarter figure.
The company is projecting $190 billion to $200 billion in revenue for 2028.
Those are extraordinary growth expectations.
But extraordinary growth also requires extraordinary infrastructure.
Every new Claude user, enterprise customer and AI workload ultimately creates another demand for computing resources.
That is why Anthropic is spending so aggressively now.
It is essentially buying capacity before it needs all of it.
But $45 billion creates a very different problem
Securing compute solves one problem.
It creates another.
The economics have to work.
A six-year, $45 billion commitment is a massive financial obligation.
Anthropic will need revenue growth to remain strong enough to justify the infrastructure spending.
And the West Virginia facility is not expected to come online until late 2027.
That means Anthropic still has to manage its capacity needs with its existing infrastructure while waiting for this new supply to arrive.
The question for investors is therefore not simply:
Can Anthropic get enough compute?
It is:
Can Anthropic turn all that compute into profitable growth?
The $30 trillion market opportunity needs scrutiny
Anthropic is reportedly preparing to tell potential IPO investors that its total addressable market could exceed $30 trillion.
That is a staggering number.
But investors should separate market opportunity from actual revenue.
A huge theoretical market does not automatically translate into a huge business.
The company reportedly expects $190 billion to $200 billion of revenue in 2028. That would still represent less than 1% of a $30 trillion opportunity.
So the real test will be execution.
Can Anthropic convert Claude’s popularity into durable enterprise revenue?
Can it maintain rapid growth as competition intensifies?
And perhaps most importantly, can revenue grow faster than the cost of all the infrastructure required to generate it?
What this means for Nvidia and AMD
Anthropic’s spending is also another signal for AI infrastructure investors.
The Nscale deal is powered by Nvidia’s next-generation systems, while the AMD agreement represents a significant commitment to an alternative platform.
That creates an interesting dynamic.
Nvidia still has enormous demand for its latest infrastructure.
But AMD is increasingly getting meaningful hyperscale and AI-model workloads, rather than simply competing on benchmarks.
At the same time, companies such as Google and Anthropic are working on their own custom silicon strategies.
The AI chip market could therefore become much more competitive over time.
For now, though, the spending is so large that multiple chipmakers can benefit from the same AI infrastructure boom.
The bigger story is not Anthropic alone
Anthropic’s $45 billion commitment is a window into where the AI industry is heading.
The bottleneck is moving beyond simply building better models.
The next battle is over:
- Chips
- Electricity
- Data centers
- Networking
- Capital
- Long-term computing capacity
And companies are willing to commit tens of billions of dollars years before that infrastructure is fully operational.
That changes the investment story.
The AI boom is creating opportunities far beyond the model companies themselves. Chipmakers, data center operators, power providers and infrastructure companies are all becoming part of the same spending cycle.
The question investors should really be asking
Anthropic has shown that it can raise enormous amounts of capital and secure enormous amounts of compute.
The next challenge is turning that infrastructure into economics that make sense.
$45 billion can buy a lot of computing power. It cannot guarantee a $2 trillion valuation.
For investors watching Anthropic’s potential IPO, three numbers will matter enormously:
- How fast revenue continues to grow
- Whether profitability improves as scale increases
- Whether all this secured capacity translates into reliable, monetizable AI demand
The AI infrastructure race is getting bigger by the month.
And Anthropic’s latest deal makes one thing clear:
The companies building AI are no longer just competing to create the smartest models. They are competing to secure the electricity, chips and computing power needed to keep those models running.