BIS figures put supply-chain partners behind about a quarter of the money AI firms raise

The sector funding itself and suppliers funding their own customers are two different claims. The BIS numbers, as Crypto Briefing reports them, back the first far more than the second. Take the 55.2% of incoming money that came from other AI firms ^[1] and multiply it by the 46.4% of in-sector deal value tied to commercial supply-chain relationships ^[5]. The product is about 25.6% of all investment into AI firms, assuming both shares sit on the same pool of deals ^[1]. About three-quarters of the capital has no supplier-customer link ^[6], and 44.8% came from outside the sector entirely ^[2].

That quarter is concentrated. Supply-chain deals were 16.1% of in-sector deals by count ^[6], so the average one was about 4.5 times the size of the average in-sector deal without a supply link (46.4 divided by 16.1, against 53.6 divided by 83.9) ^[3]. The authors, Jon Frost, Rudraksh Kansal, Kumar Rishabh, Vatsala Shreeti and Leanne Si Ying Zhang ^[11], say larger investment deals are a significant driver of the circular structures ^[7].

The flows are also lopsided, or rather, they come out of balance sheets that invest more than they raise. The 28.7% of AI firms' outgoing deal value that went to other AI firms ^[4] is the same money as the 55.2% they received from each other ^[1]. If both shares are measured on the same deals, the firms in the sample deployed about 1.9 dollars of investment for every dollar they took in ^[5]. The BIS puts hyperscalers at the centre, spending on capital expenditure while investing in AI companies and locking in supply commitments for hardware and services ^[10]. A cloud provider that funds a model developer and then sells it compute gets part of its own cash back as revenue. I think that revenue deserves a lower multiple than a sale to a customer paying with money raised elsewhere.

That view can fail in three ways. The BIS says an investor that also supplies critical inputs can reduce information asymmetries ^[8], so its price may be better informed than a generalist fund's. Guaranteed access to hardware and computing capacity, both bottlenecks, is a real asset to the customer however it is financed ^[8]. The third cuts the other way. The bulletin flags greater opacity, contagion during financial stress and macroeconomic instability ^[9]. A discount sized for calm markets would be too small if a few large investors pulled back at once. Earlier BIS research had already raised warning signs about possible overvaluation and pullbacks in AI investment ^[13].

The test is what share of a supplier's revenue comes from customers it funded, and whether those customers' outside sales cover their bills. If funded customers turn out to be a small slice of supplier revenue, or pay mostly from outside sales, the discount shrinks toward zero. Crypto Briefing, summarising the bulletin, says analysts may want to separate organic demand from demand the supplier effectively financed itself ^[12]. The study as reported measures investment flows, and the coverage does not say how much of the funded firms' spending goes back to their investors.