Something unusual is happening in the AI race. The world’s biggest technology companies are spending staggering amounts building AI data centres packed with the most advanced computing hardware on the planet.
The obvious assumption is that all this infrastructure is being built to power their own AI products.
But what if that’s only half the opportunity?
Because AI computing power is becoming so scarce – and so valuable – that the companies building the biggest pools of it may be able to make billions renting their spare capacity to the very competitors they’re trying to beat.
And that changes the investment case for some of the biggest spenders in the AI boom.
Imagine building the most powerful factory in the world.
You spend $145 billion constructing it. The machines inside are the fastest ever made. The capacity is extraordinary – more than your own products currently need.
Then your fiercest competitor calls and asks if they can use it.
Most people’s instinct would be to say no.
You’re in a race. Why hand weapons to the enemy?
But what if renting that capacity to your competitor turns an enormous capital expense into an entirely new source of revenue?
That’s the question sitting at the centre of one of the most interesting business stories in technology right now.
And it could change how investors think about the hundreds of billions of dollars being poured into AI infrastructure.
The $10 billion computing deal that surprised everyone
Anthropic – the AI research company behind Claude, one of the most capable and fastest-growing AI assistants in the world – recently approached Meta Platforms to lease its AI data centre capacity.
The potential agreement could be worth as much as $10 billion over two years.
Let that sink in for a moment.
Anthropic and Meta are direct competitors.
Meta is spending heavily to build AI models that compete directly with Claude.
And yet Anthropic, which has reportedly already signed a three-year, $45 billion agreement with SpaceX’s xAI division at $1.25 billion per month, is now in talks to hand Meta $10 billion for the privilege of using Meta’s computers.
The financial media reported this as a sign that Meta is falling behind.
That it’s building more computing capacity than it can use and preparing to hand valuable resources to a rival.
But there is another way investors should look at it.
What if Meta hasn’t built too much AI infrastructure?
What if demand for AI computing power is now so enormous that any surplus capacity can itself become a highly valuable product?
That’s a very different investment story.
Why this is really about scarcity
Training large AI models requires extraordinary amounts of computing power.
Running those models for millions of users requires even more.
And right now, the most capable AI systems appear to be consuming computing capacity faster than some of the world’s richest companies can build it.
Anthropic’s Claude Code – its enterprise coding assistant and currently its fastest-scaling product – has driven a surge in demand so sharp that Anthropic is seeking computing infrastructure from any reliable supplier it can find.
It has already committed $45 billion to SpaceX’s data centre capacity.
Now it’s in talks with Meta.
That tells investors something important.
AI compute has become one of the scarcest and most valuable resources in the global technology economy.
So scarce that competitors may be willing to pay each other billions to access it.
And that creates a potentially powerful second source of returns for the companies spending enormous sums building AI infrastructure.
They don’t necessarily have to earn a return solely from their own AI products.
They may also be able to earn a return by selling access to the infrastructure underneath them.
Meta recognised this possibility before many investors did.
Management explicitly acknowledged that the company may build more capacity than its own AI products need.
That acknowledgement was widely reported as a concern – excess capacity, wasted investment and potentially deteriorating returns.
But look at it another way.
A company that builds more capacity than it needs and then leases the surplus to competitors at attractive prices hasn’t necessarily made a mistake.
It’s turned infrastructure into a product.
And if AI compute remains scarce, that could transform some of today’s enormous AI capital expenditure from a potential liability into a valuable revenue-generating asset.
The circular economy of AI computing
The first phase of AI’s commercial development had a specific financial logic.
Big technology companies such as Microsoft, Alphabet and Amazon invested billions into AI startups like Anthropic and OpenAI.
Those startups then spent much of that capital on cloud computing and chips supplied by the same big technology companies.
The money went out and came back.
It created a circular economy that benefited the largest cloud providers while simultaneously funding the development of frontier AI.
Now we may be entering a second phase.
And it’s potentially even more interesting.
AI competitors are starting to recycle one another’s computing power.
Anthropic uses Meta’s servers.
Meta uses the revenue to build more servers.
Those servers can then be leased to other AI companies chasing capacity they can’t build fast enough themselves.
The circle expands.
And every participant can potentially become both a customer and a supplier.
For investors, this matters because it broadens the AI investment thesis.
The winners may no longer simply be the companies with the best AI models.
Owning vast amounts of computing infrastructure could itself become an enormously valuable business.
That means the billions being spent on data centres today shouldn’t automatically be viewed simply as the cost of competing in AI.
For companies able to monetise surplus capacity, those data centres could become revenue-producing assets serving the entire AI industry.
The complications?
There are genuine risks before drawing investment conclusions.
First, regulatory scrutiny is increasing.
A world where AI competitors share computing infrastructure creates concentrated chokepoints that regulators in the US, EU and UK are already watching carefully. A deal of this size and nature between major AI competitors would inevitably attract attention.
Second, the pricing power of excess compute depends on scarcity.
If the AI infrastructure buildout continues at its current pace, the shortage that makes surplus computing capacity so valuable today may not persist indefinitely.
More supply eventually means lower prices.
And third, there are limits to the strategic logic of selling computing power to a direct competitor.
Meta can lease Anthropic its servers today.
But every improvement Anthropic achieves using those servers potentially makes Claude more competitive against Meta’s own AI products.
There is therefore a point – and only Meta’s management can judge where it lies – at which the revenue from the lease becomes worth less than the competitive advantage being handed to the tenant.
What this means for investors
The AI investment story is evolving faster than most portfolio strategies are keeping up with.
The first-generation AI trade was relatively simple.
Buy the chip makers.
Buy the cloud providers.
Buy the companies supplying the picks and shovels of the enormous AI infrastructure buildout.
That trade has worked extraordinarily well.
But the next phase could be different.
Investors now need to ask not only who is building the most AI infrastructure – but who can make the most money from it.
Because if AI computing power remains scarce, the companies sitting on the biggest pools of capacity could have something enormously valuable.
They can use that infrastructure to develop their own AI products.
They can use it to serve their existing customers.
And when they have capacity left over, they may be able to rent it to AI companies desperate for more computing power.
That means the same billion-dollar data centre could potentially generate returns from several different directions at once.
And that is the part of the AI infrastructure boom I think investors should be watching very closely.
Not a subscriber to Money Morning?
You can get free daily recommendations like these with Money Morning eletter. Just sign up here.
