High-yield buyers steer AI credit toward borrowers with contracts and hard assets

The lenders quoted in the Reuters report, as summarized in Ukrainian by mezha.net, describe a return with a fixed ceiling ^[16]. Larry Holzenthaler, a senior portfolio manager for debt at Catalyst Funds, said high-yield investors want to know how much cash flow to expect, when it arrives and how likely it is never to arrive ^[5]. He said the higher pay reflects how little of the upside lenders share: if things go well they get par back, and if not they take the losses ^[5]. Lotfi Karoui, a credit strategist at PIMCO, called the risk and reward asymmetric ^[13]. Lender income is mostly coupons and principal, while the risks include high debt, project delays and fast technological change ^[13].

The pool of lenders is limited by rule. High-yield and leveraged-loan buyers have less room to add exposure than investment-grade investors, because portfolio rules limit how much risky paper they hold ^[12]. The biggest buyers of leveraged loans are CLO managers. They buy large loan portfolios and pay for them by issuing their own securities ^[14]. Rising debt, fund outflows or rating downgrades can hold that buying back ^[14]. Many AI companies need heavy spending before revenue stabilizes, and the report says that makes CLO financing harder ^[14]. Elizabeth Templeton, a senior product manager for fixed-income and multi-asset indexes at Morningstar, said CLO managers are growing more cautious about individual borrowers and checking them more thoroughly ^[15].

What does get funded fits a narrow profile. Buyers concentrate on BB-rated names, and demand is strongest for borrowers with stable revenue, long-term contracts, tangible assets and an established customer base ^[6]. Data centers account for much of that demand, and many of them meet the criteria ^[7]. Erin Brown, head of leveraged finance at BNP Paribas, said total high-yield bond issuance would have fallen noticeably without data-center borrowing ^[8]. She estimates it is roughly flat on last year ^[8]. BNP's year-to-date AI-infrastructure bond total is already about 3.3 times its figure for all of 2025 ^[2].

Price rises fast below that profile. Moving from BB+ to a lower rating adds roughly five percentage points a year ^[3]. SoftBank's three tranches show what maturity costs at BB+: four extra years added 1.125 points, from 8.625 percent to 9.75 percent ^[4]. Part of the rise is market-wide. Heavy borrowing by higher-rated issuers and a sell-off in US Treasuries are pushing market yields up ^[4].

The headline growth figure needs the same care. The $88 billion is a Goldman Sachs count of AI-related deals by low-rated companies, mostly loans to US issuers ^[1]. The $20 billion base is Neuberger Berman's figure for the leveraged-loan market over the first 11 months of 2025 ^[2]. Divided, that is a 4.4-fold rise ^[1]. The ratio holds only if both firms define an AI deal the same way, and the account does not say whether they do.

In my view, the lenders' checklist is what matters to a team buying compute or model access from a small vendor. Lenders now test projected revenue, collateral value and the ability to service debt ^[3]. A vendor with long contracts and hard assets is in the segment being funded. A vendor still spending ahead of revenue is where CLO managers are screening harder and coupons can reach 14 to 15 percent ^[10][15]. That evidence shows credit getting costlier and more selective for the second group. The mezha.net text breaks off before its worked example, and the portion available does not report a vendor default or a service cutoff.