Excess Spread is our weekly newsletter, covering trends, deals and more in structured credit and ABS — subscribe to this newsletter here.

When the chips aren’t down

The CoreWeave IPO is a bit of a Rorschach test for beliefs about AI. It’s either the pick-and-shovel seller perfectly positioned to benefit from the latest technological frontier, or over-levered, capital-hungry and in hock to Microsoft’s (62% of revenues) and OpenAI’s grandiose ambitions.

We’re not going to pretend to know much about the equity story; the book was reportedly oversusbcribed day one, but there’s clearly a lot of high conviction sceptics around, and as of Thursday (27 March), the size was revised down, and NVIDIA, CoreWeave’s largest supplier, is supplying a $250m anchor. Your ECM mileage may vary.

But we’re here to talk about the debt, which counts as among the largest ever asset-based credit facilities. It consists of two delayed draw transactions, for $2.3bn and $7.6bn, signed in 2023 and 2024 and referred to as DDTL 1.0 and 2.0 throughout the registration statement for the stock offering.

These are not just large, they’re structurally inventive.

Like a data centre ABS, to which they have some kinship, they are secured initially against the customer contracts, with the second of the two deals distinguishing between different customer categories.

Debt drawn to fund the infrastructure to fulfil “specified investment grade” customer contracts (probably blue chip clients like Microsoft) will pay 600bps over SOFR. For other IG customers, it’s 650bps, and for sub-IG its 1,300bps. There are also covenants around the percentage of sub-IG customers, debt coverage, and prepayments.

If the customer contracts fall away, then the second line of recourse is to the time on the GPU hardware itself; this can be resold, just as a commercial property can be relet. Some CoreWeave criticism since the IPO launch has discussed the rapid depreciation curve for cutting edge AI chips. It’s all very well selling picks and shovels but if you have to replace all the tools every 18 months it gets expensive.

But it’s not like this is some sort of secret. The depreciation curves for GPUs probably include fewer data points than used car sales but they’re more homogenous and subject to modelling, and the GPU-backed facilities are structured to a specified advance rate against the depreciated value of these assets. A sensible lender bakes this in from the beginning.

Reductions in the contract value or reductions in the depreciated value of the GPUs require prepayment of the facilities, so the transaction always stays fully collateralised from both directions.

The third line of recourse is a parent guarantee from CoreWeave, though in a world where nobody wants to buy high performance compute one would imagine this guarantee isn’t super-helpful, except insofar as it obtains a better seat at the table.

If it comes down to it, Magnetar Capital, which co-led the facilities with Blackstone (also a minority investor in the Magnetar GP and an LP in Magnetar funds) would already be sat on the other side of said table.

The Illinois-based hedge fund has accumulated, through direct investments, converts and warrants, a 34% stake in CoreWeave, slated to drop to 29% post-IPO. It has a board seat, was the earliest institutional investor in CoreWeave and is the largest institutional investor, though the dual-class stock means management retains the vast bulk of the voting power. The share position would be worth $5.9bn at the top end of the initial IPO range.

The GPU-backed facilities seem from this seat like very compelling trades (though do write in if you have a different view!).

If the primary risk is Microsoft’s performance on a committed contract, and you get paid SOFR+600bps for lending against that with added collateral…..that is an extremely nice risk-adjusted position.

I’ve no idea if it’s possible to lever up such an esoteric piece of lending in such large size, but if you can write triple-A backed loans at 600bps you should be doing that all day long regardless.

Magnetar and Blackstone are not the only lenders; the other participants include some of the biggest names in private ABF — Carlyle, CDPQ, DigitalBridge Credit, BlackRock, Eldridge Industries and Great Elm Capital Corp in DDTL 2.0. Other funds in DDTL 1.0 included Coatue, DigitalBridge Credit, BlackRock, PIMCO and Carlyle.

We’d love to have been a fly on the wall for the syndication process!

So there’s kind of a barbell strategy for Magnetar. It has huge potential option value from the public equity valuation associated with the buzziest of hype sectors in the economy, plus a piece of mostly investment grade financing at rates wildly above any investment grade position anywhere else in credit.

In a sense, the debt financing is a central part of the proposition (the Core!)….the ability of CoreWeave to perform its role in the AI ecosystem is dependent on its ability to raise large amounts of capital cheaply, and drilling into its corporate structure to pledge its most valuable assets (long term customer contracts and cashflows and the GPUs themselves) is the most efficient way to do that.

The spreads are tight but the water’s fine

We were fully immersed in Invisso’s SRT Symposium on Thursday, an event which keeps growing (and finally outgrew Clifford Chance’s airy, well-lit conference suite this year).

The tone was a little different to last year’s event, which was fizzing with excitement about the immense potential of the US market, which had just started to come on line, prompting vigorous capital raising and new allocations from multi-strategy funds.

The regulatory uncertainty hanging over the US market, though, has not gone away, and has become much more serious, as the new US administration’s intentions on bank capital are…obscure.

The CFPB has been gutted, the SEC is in the crosshairs, and regulation in general, especially regulation agreed by international organisations, is not very MAGA-friendly.

Maybe we get a 2am text wall on Truth Social starting: “For TOO LONG the FAILING BASEL COMMITEE supported by the WEF ELITES has attacked US BANKS.”

But what does that mean for calculating the P-factor and structuring a deal which can achieve regulatory capital relief and fit within existing interpretations of Regulation Q? Trump is unlikely to tweet actionable guidance on these topics.

So US supply has disappointed the market, but the money raised has to be deployed somehow, and thus we’ve seen intense spread tightening, eroding the excess returns available in the SRT market vs other comparable credit products.

Exactly what you benchmark SRTs against remains heavily debated. Credit indices, index tranches, CLOs (but which tranche?), other asset-backed markets, AT1?

There are problems with each as a comparison point but all have tightened, and even if some of the juice is gone from the SRT market there’s still a good argument that it remains attractive on a risk-adjusted basis. Or at least, lots of SRT PMs think so.

If SRT spreads are historically tight, what’s the issuer reaction function? Some markets (CLOs for example) have a sort of natural stabilisation built in. When CLO arbitrage is good, more CLOs are created and more CLO tranches issued, until supply weight erodes the arbitrage, and vice versa.

SRT supply is more strategic. Deals take a long time to put together, and issuers try to signal their plans early in the year. So there’s a limited ability for issuers to add more deals to their pipeline to take advantage of good conditions.

Somewhat easier is increasing the size of existing transactions, especially broadly distributed vanilla large corporate deals. These portfolios are enormous, relatively homogeneous and more reference obligations can easily be added in, while the relatively standardised deal terms mean bumping up the size and allocating to more investors isn’t too painful.

Regulation is always in the air when SRT is being discussed. The US is a mystery wrapped up in an enigma, but Canada has delayed the Basel Endgame, wanting to see what everyone else does.

In the EU, change is afoot, with the European market expecting the Commission to launch a major overhaul of the Securitisation Regulation this year, in response to the “Targeted Consultation”.

The bid-offer on regulatory expectations is wide, and there’s a real risk of disappointment. If insurers get, say, a 10% improvement in capital treatment for investing in ABS, that is not going to unlock a great wave of demand for the product, and if the regulation gives with one hand (lower capital) and takes with the other (still more onerous disclosure or reporting) that could net out worse than the current status quo.

One realistic aspiration is for a change to authorise Simple Transparent and Standardised (STS) treatment for unfunded (i.e. insurance liability-side) synthetics. This small tweak would have large effects. From the buyside’s perspective, the last thing they need is more competition for deals, but adding more insurers to the market could be a benefit, as they’re more likely to be focused on mezzanine tranches.

Tight spreads on thick tranches can be partly compensated for by structural leverage; perhaps the traditional SRT hedge fund can buyer thinner tranches with insurance taking up the slack.

That might be particularly important as scrutiny over SRT repo leverage grows. The ECB is asking questions, at least one bank (Deutsche) has pulled in its horns, and rumours are swirling about others. Banks quitting the business might be banks that didn’t have terribly large SRT repo businesses in the first place, or they might just find that the returns don’t justify the hassle.

We don’t think there’s any need to panic about SRT repo leverage, but regardless of the rights and wrongs, the direction of travel seems to be towards a smaller range of counterparties.

Of course, pending any regulatory changes, there remains the art of financial structuring. The funded / unfunded challenge is well understood, and there are routes around it.

Celeste has a great piece surveying the insurance scene and looking at different solutions, such as funded repacks (where a bank contributes the funding while the insurer takes the portfolio risk) or even solutions from both sides of an insurer’s balance sheet, fusing the cash investment potential with the risk-bearing liability side and fund involvement to provide an efficient package.

Sell them better

Much lobbying and advocacy effort has been expended in trying to convince European regulators to change the rules for securitisation, in the hope of reviving the market and implicitly, catching up to the much more dynamic US market.

But some challenges in the European market are basically self-inflicted, and could probably, with the right willpower, be sorted out among market participants. One example is portfolio sale processes, which are long and expensive.

If there are 10 bidders for a deal, nine lose, and each one is in the hole for a couple of months of Magic Circle-level legal fees… that’s a drag on overall buyside profitability in the industry (though good for the lawyers!).

Funds doing asset-based finance are generally either bidding big portfolios from the established banking industry or financing smaller new entrant fintechs, and the big portfolios, being big, are a more important driver of overall economics.

The cost of lost deals particularly smart when a fund isn’t winning much. That is presumably because they didn’t pay the most money. But for a fund which wants to be appropriately conservative, and maximise returns by avoiding overpaying, will find itself racking up legal fees without much to show for it.

Is there a better way? Some US portfolio processes work differently, and allow detailed due diligence after the sale. A portfolio can trade, and if, on going line-by-line, property-by-property, it is found that certain loans were not in fact eligible or didn’t fit in the credit box which was the basis for the sale, these can be plucked out of the portfolio.

Portfolio bidding practices sit downstream from corporate M&A convention and legal requirements. In the US, a potential buyer of a company doesn’t have a legal certain funds requirement — hence the notorious Drexel “highly confident” letter.

Committed acquisition financing is nice to have, is common practice, and certainly strengthens a bid, but it’s not a requirement as it is in much of Europe. The UK Takeover Code for listed companies even requires bidders to set out the details, terms and counterparties for any acquisition financing prior to the shareholder vote.

Part of the reasoning behind European practice may also come down to the fundamental motivation for portfolio sales or asset disposal.

Generally banks have sold assets in size because they’re part of non-core divisions or major balance sheet cleanup operations. Selling institutions want nothing more to do with the portfolios; reps will be grudging and minimal, and arrangements to kick out loans non-existent. Bidding will be all or nothing, so if a problem didn’t come up in diligence, you’re stuck with it.

But if you can detach detailed financing arrangements and detailed due diligence from the basic bidding process for portfolios, the whole thing becomes much easier and cheaper to run. The same funds are lining up bids, writing investment memos and going through credit committee, but with far less external cost associated and on a much more compressed timeline.

Take a view, find a price, and get through to exclusivity (or at least final two), at which point legal costs and full due diligence becomes easier to swallow.

But practice and culture has a way of getting embedded, even when the rationale for certain habits falls away. The same bidding institutions and the same sellside advisors are in a repeated game with a well-understood ruleset, and may not share the goal of making portfolio sales easier or faster to transact.

Our customers receive this content ahead of the crowd — find out more about 9fin’s news and analysis.