Is circular financing in AI a problem?

There are lots of ways in which circular financing among AI companies can be organised. A new report from a team of Bank for International Settlements researchers lists three.

First, there are those tech firms who receive both financing and business from another tech firm; so, maybe including Amazon pouring billions into Anthropic while also supplying the Trainium chips used to train Claude.

Second, suppliers could be financing customers to stimulate demand for their products, a bit like old-fashioned vendor financing. Nvidia’s pivot to providing ‘balance sheet as a service’ comes to mind — along with numerous reports of the giant chipmaker pumping cash in the direction of its financially more precarious customers so that they can buy more chips.

Or the flow could run the other way: customers financing suppliers in order to secure critical inputs. And finance can take many forms, from straight-up loans and equity injections through to forward compute contracts. So maybe the $300bn of business with which OpenAI is backstopping Oracle’s data centre build-out counts, even if OpenAI doesn’t actually have the $300bn and is relying on current or future investors to supply it.

Third comes everything else: reciprocal commercial relationships where goods or services flow both ways.

Looking at a universe of 1,246 AI firms split across compute, infrastructure, data tools, models and applications, the report’s authors — Jon Frost, Rudraksh Kansal, Kumar Rishabh, Vatsala Shreeti and Leanne Si Ying Zhang — found, perhaps unsurprisingly, a lot of circular relationships.

Totting up all the incoming investment flowing to AI firms between 2021 and 2025, they reckon more than half came from other AI firms. And looking just at this lump of AI-to-AI deals, almost half of these were between firms that shared some commercial relationship, suggesting “that a significant portion of financing and commercial relationships in AI are self-referencing.”

But is it a problem?

... circular relationships make reported demand partly endogenous to firms’ own financing decisions. For example, when a supplier finances a customer, part of the supplier’s revenue growth reflects its own capital investment, rather than organic final demand. This makes it harder for investors, lenders and supervisors to gauge what part of the current AI boom is based on organic demand.

... circular relationships make reported demand partly endogenous to firms’ own financing decisions. For example, when a supplier finances a customer, part of the supplier’s revenue growth reflects its own capital investment, rather than organic final demand. This makes it harder for investors, lenders and supervisors to gauge what part of the current AI boom is based on organic demand.

Beyond making analysts’ lives a bit tricky, are there real-life problems for the rest of us?

The parallel with the telecommunications boom of the late 1990s is instructive: upstream equipment vendors such as Lucent and Nortel financed network operators so that the operators could buy the vendors’ equipment. This meant that part of the equipment vendors’ reported sales was being funded by the vendors themselves. For a time, as operators expanded their networks, equipment orders also expanded and vendors booked both the sales and loans as assets. However, when operators’ own revenues failed to materialise or slowed, they could neither repay the loans nor sustain the equipment purchases. Equipment vendors then sustained both financial losses and a loss of sales. Such dynamics may also play out in AI if revenue growth and end user demand fall short of firms’ expectations.

The parallel with the telecommunications boom of the late 1990s is instructive: upstream equipment vendors such as Lucent and Nortel financed network operators so that the operators could buy the vendors’ equipment. This meant that part of the equipment vendors’ reported sales was being funded by the vendors themselves. For a time, as operators expanded their networks, equipment orders also expanded and vendors booked both the sales and loans as assets. However, when operators’ own revenues failed to materialise or slowed, they could neither repay the loans nor sustain the equipment purchases. Equipment vendors then sustained both financial losses and a loss of sales. Such dynamics may also play out in AI if revenue growth and end user demand fall short of firms’ expectations.

We couldn’t find a market cap series for Lucent, but the Nortel market cap chart suggests circular financing can be a problem. But is it a problem this time?

...these circular arrangements tend to be opaque, making monitoring and supervision difficult. Many of the firms involved in AI are private and may disclose limited information. Even where firms are publicly listed, deal terms can often be complex, mixing cash investments with long-term purchase commitments and guarantees on the value of the underlying assets. Residual value guarantees (RVGs) are one example, where the guarantor pledges to cover any shortfall in an asset’s worth (such as chips or data centre equipment) after a fixed period. These contingent commitments sit off balance sheet and only materialise during a downturn – when a guarantor is least able to absorb them. As such, reported deal values may not capture the whole picture, and headline figures can differ substantially from disbursed amounts. Moreover, many of these firms reside in different sectors and jurisdictions, making it difficult for a single supervisor or regulator to monitor these risks.

...these circular arrangements tend to be opaque, making monitoring and supervision difficult. Many of the firms involved in AI are private and may disclose limited information. Even where firms are publicly listed, deal terms can often be complex, mixing cash investments with long-term purchase commitments and guarantees on the value of the underlying assets. Residual value guarantees (RVGs) are one example, where the guarantor pledges to cover any shortfall in an asset’s worth (such as chips or data centre equipment) after a fixed period. These contingent commitments sit off balance sheet and only materialise during a downturn – when a guarantor is least able to absorb them. As such, reported deal values may not capture the whole picture, and headline figures can differ substantially from disbursed amounts. Moreover, many of these firms reside in different sectors and jurisdictions, making it difficult for a single supervisor or regulator to monitor these risks.

Hmm. Without the necessary disclosures to pore through, even the hardest-working supervisor or investor is going to struggle to know whether it’s a problem until after the fact.

But there is a silver lining. If it does turn out to be a problem, the BIS reckon that at least we’ll all know about it. Not only through a public equity market swoon, and upending of macroeconomically significant investment. But also because they reckon risks will be magnified by hidden leverage and interconnected exposures across private credit and special purpose vehicles used to finance AI infrastructure. 😬

Further reading:
— Joining the dots between big AI (FTAV)
— Nvidia’s $200bn ‘balance sheet-as-a-service’ (FTAV)
— The US economy is running hot (FTAV)