In the shadow of a global shortage of computing power, Meta and Anthropic are negotiating a lease arrangement worth as much as $10 billion — a deal that quietly redraws the map of who builds AI and who merely enables it. What was once a social media company's internal infrastructure is becoming a commodity that rivals are willing to pay handsomely to access. This moment reflects something deeper than a single transaction: the race to build artificial intelligence has concentrated so much value in raw computational capacity that the chips themselves have become the new oil, and those who hold t
Meta in Talks to Lease AI Computing Power to Anthropic in Potential $10B Deal
Computing power has become the most precious commodity in AI
Mark
Why does Meta have excess computing power in the first place? Isn't the whole industry supposed to be scrambling for capacity?
Mimi
Meta built massive data centers to run its own platforms—recommendation algorithms, content moderation, all of it. Those systems don't need to run at full capacity all the time. Meanwhile, AI development is a different beast entirely. It's constant, it's hungry, and it's growing. So Meta has pockets of unused infrastructure that Anthropic desperately needs.
Mark
But isn't Meta building its own AI models? Why would they lease power to a competitor?
Mimi
They are, absolutely. But Meta's AI work doesn't consume every bit of capacity the company has. And more importantly, this is money. Leasing that power generates revenue from something that's already paid for. It's not zero-sum—Meta can do both.
Mark
What changes if this deal actually closes?
Mimi
You'll start seeing other tech giants do the same thing. Google, Amazon, Microsoft—they all have excess capacity at different times. Suddenly, computing power becomes tradeable, like electricity. That reshapes how startups and smaller companies can compete. You don't need to own the infrastructure anymore; you just need to afford the lease.
Mark
Does that make the AI race more fair or less fair?
Mimi
Probably less, honestly. It still favors companies with capital. But it does lower the barrier to entry compared to building everything yourself. The real question is whether these leases come with strings attached—data access, partnership terms, things like that.
Der Puls
- The global shortage of high-end processors has created a genuine bottleneck in AI development, forcing even well-funded companies like Anthropic to scramble for computing resources they cannot easily build themselves.
- Meta, once purely a consumer platform, finds itself sitting on excess computing capacity that has suddenly become one of the most coveted assets in the technology industry.
- A potential $10 billion lease deal would blur the line between competition and cooperation, turning two AI rivals into landlord and tenant in an arrangement that would have seemed improbable just years ago.
- If the deal closes, it could signal a broader reshaping of the AI landscape — where infrastructure partnerships, not just research breakthroughs, determine who can stay competitive in the race to build powerful models.
In the shadow of a global shortage of computing power, Meta and Anthropic are negotiating a lease arrangement worth as much as $10 billion — a deal that quietly redraws the map of who builds AI and who merely enables it. What was once a social media company's internal infrastructure is becoming a commodity that rivals are willing to pay handsomely to access. This moment reflects something deeper than a single transaction: the race to build artificial intelligence has concentrated so much value in raw computational capacity that the chips themselves have become the new oil, and those who hold them hold leverage over the entire field.
The negotiations between Meta and Anthropic revolve around something that has quietly become the most precious resource in artificial intelligence: raw computing power. Meta is exploring a lease arrangement potentially worth $10 billion — a number that speaks to just how desperately AI developers need access to the chips and data centers required to train and run their systems.
For years, Meta built its computing infrastructure to serve its social media platforms — powering recommendation algorithms, processing billions of daily interactions. But the company has found itself with surplus capacity, and in a moment when every major AI developer is scrambling for GPUs, that surplus has taken on an entirely new kind of value.
Anthropic, one of the most well-funded AI startups in existence, faces the same constraint as everyone else: there simply aren't enough high-end processors to go around. The global shortage has become a structural bottleneck, with companies bidding against each other for Nvidia's latest chips and negotiating with cloud providers just to stay in the race. Leasing capacity from a competitor that already has it built is, under these conditions, a rational move.
What makes the deal significant beyond its dollar figure is what it signals about Meta's identity. The company is evolving into something like an infrastructure provider — a landlord of computing power to the very rivals it competes with in AI. For Anthropic, a lease frees up capital for research rather than construction. For Meta, it generates revenue from assets already paid for and maintained.
Yet the arrangement also raises quieter questions. When competitors begin leasing resources to one another, the boundaries between rivalry and interdependence dissolve. This is what extreme scarcity produces: business relationships that would have seemed unthinkable just a few years ago, forged not by shared vision but by the hard logic of who controls the infrastructure that makes the future possible.
The conversation between Meta and Anthropic, now in active negotiations, centers on something that has become the most precious commodity in artificial intelligence: raw computing power. Meta is exploring a lease arrangement that could be worth as much as $10 billion, a figure that reflects just how desperately the companies building the next generation of AI systems need access to the chips and infrastructure required to train and run them.
For years, Meta built its own computing capacity primarily to serve its social media platforms—to power the algorithms that show you content, to train recommendation systems, to process billions of user interactions daily. But the company has found itself with excess capacity, and in a moment when every major AI developer is scrambling to secure enough GPUs and data center resources to stay competitive, that surplus has become valuable in an entirely new way.
Anthropicis one of the most well-funded AI startups in existence, backed by billions in investment and focused on building large language models that can compete with OpenAI's offerings. Yet even with that financial firepower, the company faces the same constraint that every AI developer faces: there simply aren't enough chips and computing clusters to go around. The global shortage of high-end processors has become a genuine bottleneck in the race to build and improve AI systems. Companies are bidding against each other for access to Nvidia's latest GPUs, negotiating with cloud providers, and now, apparently, looking to lease capacity from competitors who happen to have it available.
What makes this potential deal significant is not just the dollar figure, though $10 billion is substantial. It represents a fundamental shift in how Meta sees itself and its assets. The company is no longer purely a consumer platform business. It is becoming, in effect, an infrastructure provider—a landlord of computing power to the very companies it competes with in the AI space. This is a new business line, one that could generate recurring revenue from assets that might otherwise sit underutilized.
The scarcity that makes this deal possible is real and structural. Training a state-of-the-art large language model requires weeks or months of continuous computation on thousands of high-end processors. The energy costs alone are staggering, and the upfront capital required to build the necessary data centers is in the billions. Not every company can or wants to build that infrastructure from scratch. Leasing it from someone who already has it built is a rational alternative—if the price is right and the terms are workable.
For Anthropic, securing computing power through a lease agreement would free up capital that might otherwise go toward building or buying infrastructure, allowing the company to focus on research and product development. For Meta, it opens a revenue stream from assets it already owns and is already paying to maintain. Both companies benefit, at least in theory, from an arrangement that lets each focus on what it does best.
But the deal also raises questions about the nature of competition in AI. When companies start leasing computing power to each other, the boundaries between cooperation and competition blur. Meta and Anthropic are not partners in any traditional sense—they are rivals in the race to build powerful AI systems. Yet they are also becoming interdependent, at least temporarily, in a way that reflects the extreme concentration of resources required to compete at the highest levels of AI development. This is what scarcity does: it forces strange bedfellows and creates business relationships that might have seemed unthinkable just a few years ago.
Möchten Sie die ganze Geschichte?
Das Original lesen bei The New York Times ↗
Technology