Why AI Companies Are Now Renting Power From Rivals

Why AI Companies Are Now Renting Power From Rivals

Here is a strange picture of the modern AI industry: one company that builds AI models may soon pay billions to rent computing power from a direct competitor that also builds AI models. That is what is reportedly being discussed between Meta and Anthropic.

It sounds odd, but it reveals something important about how artificial intelligence actually works today, and about a bottleneck that affects every country with AI ambitions, including Pakistan. Here is what is being discussed, why it is happening, and what it means beyond Silicon Valley.

(Disclosure: Anthropic, mentioned in this article, is the company behind the Claude AI assistant. This piece is based entirely on independent reporting from Reuters, the New York Times, and CNBC.)

What Is Reportedly Being Discussed

The talks were first reported by the New York Times and confirmed by other outlets. Meta Platforms is in talks to lease computing power to Anthropic in a potential deal worth up to $10 billion over two years, according to a source familiar with the matter.

The structure is notable for its flexibility. Anthropic would pay Meta in monthly increments over the two-year period, and both companies would be able to exit any agreement early.

The most important caveat, and one worth emphasising, is that this may not happen at all. Anthropic proposed the deal in June, and Meta is considering it, but the discussions are in their early stages and may not result in a deal, according to the source. Both companies declined to comment. So this is a reported negotiation, not a signed agreement.

Why Would an AI Company Rent From a Rival?

This is the genuinely interesting question, and the answer explains a lot about today's AI industry.

Training and running advanced AI models requires enormous amounts of specialised computing power, referred to simply as "compute." This means data centres packed with expensive chips, mostly made by Nvidia, running continuously and consuming vast amounts of electricity.

The problem is that there is not enough of it to go around. Securing sufficient Nvidia chips continues to be a bottleneck for AI developers such as Anthropic, which has had to cap how much users can interact with its top-tier models. When demand outstrips available computing capacity, companies must limit what their own products can do.

Building your own data centres takes years and billions of dollars. So AI companies increasingly rent capacity wherever they can find it, even from competitors. As one report noted, the compute shortage has made such arrangements routine, with SpaceX selling computing capacity to multiple AI companies.

The Scale of Spending Is Staggering

The numbers involved put the AI infrastructure race in perspective.

Anthropic already has a much larger arrangement elsewhere. It reached a deal with Elon Musk's SpaceX in May calling for payments of roughly $1.25 billion each month, totalling about $45 billion over three years, for computing resources at SpaceX's Colossus 1 data centre. The reported Meta deal would be roughly a third that size.

Meta's own spending is extraordinary. The company could spend as much as $145 billion on capital expenditures in 2026, much of it on AI infrastructure, more than double the $72 billion it spent the previous year.

To put that in context for Pakistani readers: Meta's projected annual capital spending is many times the size of Pakistan's entire IT export earnings, which recently hit a record $4.5 billion. That gap is the reality of the global AI infrastructure landscape.

Why Meta Would Do This

For Meta, this is about turning a massive cost into a revenue stream. The company has spent enormous sums building AI infrastructure, and investors have questioned whether that spending will ever pay off.

Selling excess capacity is one answer. CEO Mark Zuckerberg said in May that entering cloud computing was "definitely on the table," noting that companies approach Meta seeking to purchase capacity at a premium to what Meta paid for it.

The company has been building toward this. Meta has hired Dave Brown, a long-serving former Amazon Web Services executive, to lead a new initiative reported as Meta Compute. That would put Meta in competition with established cloud providers.

There is an obvious tension, though. Meta builds its own AI models and competes with Anthropic's Claude, so a deal would make it simultaneously a competitor and an infrastructure supplier. Analysts have noted this unusual dynamic, though the compute shortage makes such arrangements increasingly normal.

The Pakistan Angle: Why Compute Is the Real Bottleneck

For readers in Pakistan, this story carries a genuinely important lesson.

Pakistan has ambitious AI plans, including a $1 billion investment commitment by 2030 and a stated goal of building "sovereign compute", national AI computing capacity owned and controlled domestically. This story shows exactly why that is both important and extremely difficult.

If companies worth hundreds of billions of dollars are struggling to secure enough computing power, and paying billions to rent it from rivals, the challenge facing a developing economy is obvious. Compute is expensive, chips are scarce, and the infrastructure requires enormous, reliable electricity, an area where Pakistan faces well-documented constraints.

This helps explain why Pakistan ranks eighth of seventeen in its region for AI readiness, held back significantly by infrastructure gaps. Ambition is not the limiting factor; physical infrastructure is.

The practical implication is that Pakistan's realistic near-term AI strategy is probably not competing to build frontier models, which requires compute at a scale far beyond current national capacity. It is more likely applying existing AI tools well, building strong talent, developing specialised local applications like Urdu-language AI, and gradually growing domestic infrastructure. That is a less headline-grabbing path, but a far more achievable one.

Industry Impact: What It Means Globally

The compute crunch is reshaping the technology industry in several ways.

For AI companies, access to computing power has become a primary competitive constraint, sometimes more limiting than talent or funding. Companies are signing multi-billion-dollar, multi-year commitments simply to secure capacity.

For infrastructure providers, this is an enormous opportunity, which is precisely why Meta is considering entering the market alongside established cloud companies.

For investors, the spending has raised real questions. Analysts have noted that investors increasingly question whether such extraordinary levels of spending can be justified by returns, a concern that has contributed to market volatility around AI stocks.

For users everywhere, including in Pakistan, compute scarcity is why AI services have usage limits and subscription tiers. The constraints you experience using AI tools trace back directly to this shortage.

Expert Insight: Infrastructure Is the New Oil

The broader pattern is that AI competition has shifted from being purely about algorithms and talent to being substantially about physical infrastructure: chips, data centres, and electricity.

This favours entities with enormous capital and existing infrastructure, and creates a meaningful divide between countries that can build AI capacity and those that primarily consume AI built elsewhere. It is precisely the "AI divide" that international bodies, including the newly formed WAICO that Pakistan recently joined as a founding member, say they aim to address.

For developing economies, the honest assessment is that closing this gap entirely is not realistic in the near term. The more practical strategies are building talent, applying AI effectively to local problems, and developing infrastructure incrementally rather than attempting to leapfrog.

Future Outlook

Watch whether these talks produce an actual agreement, given that early-stage negotiations frequently collapse. More broadly, expect the compute shortage to persist, more unusual partnerships between competitors, and continued massive infrastructure investment.

For Pakistan, the relevant question is whether its AI investments focus on realistic, high-impact areas rather than attempting to match global compute spending, which is not a race any single developing country can win alone.

Conclusion

The reported Meta-Anthropic talks are, on the surface, a story about two American technology companies. But the underlying reality matters everywhere: computing power has become the scarce resource shaping the entire AI industry, so scarce that direct competitors are becoming each other's suppliers. For Pakistan, the lesson is clarity rather than discouragement. Building frontier AI infrastructure requires capital and energy at a scale no developing economy can currently match. The winning strategy is to build talent, apply AI intelligently to local challenges, and grow infrastructure steadily. Understanding where the real bottleneck lies is the first step to navigating around it.

This article is for general informational purposes only and reflects reports available as of July 2026. The described negotiations are early-stage and may not result in an agreement. Disclosure: Anthropic, mentioned here, is the developer of the Claude AI assistant. This is not investment advice.