Who is funding the tech revolution?

Nvidia is also investing US$1.5bn in the project’s developer, SB Energy. The datacentre will require enormous quantities of computing equipment, much of which will ultimately contain Nvidia technology.

Interconnected ecosystems

This is an unusually large example of a much broader phenomenon. Capital is moving around the AI ecosystem while products and services move between many of the same companies. The important question is therefore not whether these transactions are legitimate. It is whether the financial indicators we normally use to judge demand tell us everything we think they do.

Consider CoreWeave, one of the companies that has emerged to provide the enormous computing capacity required by AI developers. Nvidia supplies the graphics processing units (GPUs) underpinning CoreWeave’s infrastructure; in January 2026, it invested US$2bn directly in the company.

CoreWeave, meanwhile, is heavily dependent on Nvidia technology. Its regulatory filings state that all GPUs currently used in its infrastructure are Nvidia GPUs, while Nvidia  accounted for 17% of its purchases from suppliers in 2025. 

The relationships extend further. Microsoft accounted for approximately 67% of CoreWeave’s revenue in 2025. OpenAI has committed to purchase up to US$6.5bn of services until 2031 under one order form, while CoreWeave’s total contracted commitments with Meta have since grown to approximately US$35bn.

Circular financing

The result is an increasingly interconnected system. A chipmaker can simultaneously be a supplier and investor. A cloud provider can be a customer, investor and infrastructure partner elsewhere in the ecosystem. AI developers can make enormous long-term computing commitments while themselves depending upon outside capital. This is what is increasingly being described as circular financing.

The term can sound more sinister than it necessarily is. Strategic suppliers frequently invest in promising customers, while infrastructure projects routinely use long-term purchasing commitments to secure financing; an electricity producer might build a power station because a major customer has guaranteed future demand, for example.

AI presents a similar economic challenge. Enormous amounts of infrastructure must be built today to serve demand expected tomorrow. Financing relationships can help bridge that gap – but they also complicate what the resulting numbers mean.

Suppose an AI infrastructure company announces billions of dollars of contracted revenue. That is meaningful information. Yet finance professionals should increasingly ask a second question: where does the money at the end of the chain ultimately originate?

Consider two simplified scenarios. In the first, thousands of businesses purchase AI applications because they reduce costs, increase revenues or allow them to perform activities they previously could not. AI companies use that revenue to purchase computing capacity; cloud providers expand datacentres; and those datacentres purchase more processors. Money moves upstream because economic value is being created downstream.

In the second, investors provide billions to an AI developer. The developer commits billions to computing capacity. The infrastructure provider raises capital against anticipated demand and purchases processors. The processor manufacturer records growing sales while simultaneously investing elsewhere in the same ecosystem.

The transactions in both scenarios can be entirely real, with the main difference being the economic origin of the demand. The distinction is between demand generated by realised economic value and demand supported by expectations of future economic value. And it is that distinction, rather than the existence of circular relationships themselves, that matters when assessing the sustainability of the AI boom.

Bursting bubble

Variations of vendor financing and strategic investment have supported capital-intensive industries for decades. But it does make one apparently simple question surprisingly difficult to answer: how much demand would exist if the supplier had not helped finance the customer?

Enormous infrastructure investments can initially appear excessive and subsequently enable genuine economic transformation. Railways, electricity networks, telecommunications infrastructure and the internet all required investors to commit capital before the full range of profitable applications became apparent.

Equally, financing arrangements can make expected demand appear more secure than it ultimately proves to be. Around the turn of the millennium, telecommunications equipment manufacturer Nortel provided around US$5.17bn of financing that helped customers build networks and Nortel sell equipment.

But when telecommunications markets deteriorated, the relationship worked in reverse. Some customers entered bankruptcy or experienced financial difficulties, while Nortel reported that its ability to transfer customer financing to third-party lenders had substantially diminished. By the end of 2001, the company had recorded US$887m in provisions against its drawn customer-financing exposure. 

Value added

The important takeaway is that financial interdependence can indeed amplify success, but it can also transmit weakness. Ultimately, circular financing cannot remain circular forever. Somewhere, money must enter the system from organisations and consumers willing to pay for AI because it creates value for them.

That may already be happening. Cloud growth remains strong, companies are deploying AI across activities ranging from software development to customer service, and computing capacity remains constrained in parts of the market. Today’s investment may therefore prove prescient: infrastructure is being built ahead of demand because companies expect that demand to continue expanding. But the extraordinary infrastructure now being constructed implicitly assumes much more demand still to come.

That changes the questions we should ask when evaluating AI companies – or our own organisations’ AI strategies.

For example, how much revenue comes from a small number of customers? How dependent are those customers on external financing? What long-term purchasing commitments sit behind infrastructure expansion? Who ultimately carries the risk if expected utilisation does not materialise? How quickly are AI assets depreciating economically as new generations of hardware arrive?

And, crucially, can the organisations using AI demonstrate financial benefits sufficient to justify what they are spending? These questions do not tell us whether AI is a bubble. Instead, they can offer an insight into whether economic value is beginning to catch up with financial expectation.

The AI industry may ultimately justify every billion currently being invested in it. Today’s extraordinary infrastructure build-out could eventually look like the foundations of a new industrial age. But the bigger the numbers become, the less useful those numbers are on their own. For those of us trying to understand the AI economy, following the money is no longer enough. Increasingly, we also need to ask where the money came from in the first place.