Is CoreWeave the Silicon Valley bank of 2026?

Remember the Bank of Silicon Valley? Its collapse in 2023 has raised fears of a wave of U.S. regional bank failures that would hamper credit growth, cause a recession and force the Federal Reserve to cut interest rates. Oops.

Instead, only a handful of small, equally poorly managed banks imploded, economic growth remained robust, and the Fed actually raised rates that month – and twice more in the months to come.

The change in expectations has occurred violently in the markets. The yield on the 10-year Treasury fell from above 4 percent to a low of 3.3 percent, before rising back to nearly 5 percent as investors realized that the economy had completely ignored the SVB mess. The KBW banking index lost up to a third of its value, but ended the year almost flat.

Barclays’ Ajay Rajadhyaksha believes something similar is happening now with CoreWeave: investors hysterically extrapolating from a single, almost single case, and wrongly questioning the entire AI investment frenzy.

Rajadhyaksha specifically argues that people like Chamath Palihapitiya and Elon Musk misunderstand CoreWeave’s convoluted balance sheet and profit from the sale of its $2 billion unsecured bonds.

This is silly, Rajadhyaksha argues, because it simply reflects the low ranking of these specific unsecured bonds in CoreWeave’s much larger debt of $35 billion – below the data center operator’s “secured deferred draw term loans,” its convertible bonds, and billions of dollars in lease obligations.

Basically, if CoreWeave runs into real trouble, these unsecured bonds will likely be wiped out, making them particularly vulnerable to even mild investor angst. Additionally, “even if bonds could be rationally priced based on their own recovery prospects, they could still be completely wrong as a signal about the viability of the entire AI CapEx cycle,” Rajadhyaksha writes.

Here’s Barclays’ more detailed breakdown of CoreWeave’s balance sheet, with Alphaville’s emphasis in bold below:

CoreWeave funds in three tiers, and they have different risk profiles. The workhorse is secure DDTLs. The majority of CoreWeave’s funding resides in special purpose vehicles, each backed by specific GPU hardware, a customer contract, and the associated data center lease. When Meta signs a multi-year take-or-pay contract and this contract is awarded to an SPV, the bank lends against Meta’s promise to pay.

The progression of recent transactions tells you that the market is still open but increasingly expensive. DDTL 4.0 in March – $8.5 billion, supported by Meta, valued SOFR+225 bps, rated A3 by Moody’s. This is investment grade. DDTL 5.0 in May – $3.1 billion, SOFR+450 bps. DDTL 5.5 in August – $2.6 billion, SOFR+550 bps. The spread is widening. But the market is open.

The stock bridge is convertibles. About $9 billion issued over the past year, at coupons between 1.75% and 2.875%. These become less attractive as the stock falls – the conversion becomes more dilutive – but the existing paper is locked in. The September deal was increased from $3 billion to $3.7 billion, suggesting there was demand. But after last week’s price movements, this channel is likely under pressure.

We view unsecured notes as the final piece of the puzzle. A liquidation involves tens of billions of dollars in receivables. A 15% yield on this paper tells you a lot about this paper. This doesn’t tell you that the company is about to go out of business.

If a company has a $35 billion mortgage and a much smaller credit card balance, that tells you a lot about the credit card. That doesn’t tell you that every mortgage borrower in this city is in trouble..

Rajadhyaksha believes that Coreweave’s order book alone is large enough to fund its investment plans, and says that markets “simply tend to find the smallest, most volatile instrument in a complex system and treat it as the barometer of the whole thing.”

Even if CoreWeave were to halt all its investment projects, the impact on the broader development of AI would be minimal, says the Barclays Global Research Chair:

CoreWeave’s 2026 CapEx forecast is $35 billion to $39 billion. That seems like a big number – and it is for just one company. But compare it to the total. The five largest hyperscalers – Microsoft, Meta, Alphabet, Amazon and Oracle – are together spending between $725 billion and $800 billion on AI infrastructure this year, with estimates north of $1 trillion for 2027. CoreWeave’s spending represents about 4% to 5% of that total. The other neo-clouds – Nebius, Crusoe, Lambda – are still much smaller, collectively 2%.

So in the most extreme downside scenario – CoreWeave stops spending altogether – we lose about 4-5% of total AI investments. If its expenses are cut in half, which might be a more realistic downside, the impact is 2%. On an estimate of over a trillion dollars for 2027, that’s $20 billion. Not nothing. But not the cycle.

The risk is of course contagion. The idea that if Coreweave is in trouble, that means Google will no longer be able to fund its data center. We think that’s a really big way to go – extrapolating from a neocloud to one of the most profitable companies in the world. Additionally, a large portion of hyperscalers’ spending is still funded by free cash flow. Microsoft, Meta, Alphabet and Amazon are not extremely dependent on the unsecured bond market, although they would like to – and need to – issue more.

The logical chain that the market is implicitly heading towards – CoreWeave bonds at 15%, so AI infrastructure funding is broken, therefore the investment cycle is over – does not hold. CoreWeave’s unsecured obligations constitute a tiny portion of CoreWeave’s capital structure. CoreWeave’s investments represent only a tiny portion of total AI investments. And the hyperscalers who do the vast majority of spending don’t finance it like CoreWeave does.

Alphaville very much agrees with this argument. But we think Rajadhyaksha is overdoing it, for three reasons.

First, given the dominant importance of AI trading in virtually everything happening right now – even Goldman Sachs says the entire global stock market has virtually become “one big market” – it’s only natural that investors would react warily to even indirect hints that it might be weakening.

Second, CoreWeave is not a small and obscure security, like the $KRE bank stock ETF that Rajadhyaksha uses as an example of the overblown US financial rout of 2023. CoreWeave is an example of AI development that is heavily backed by Nvidia, Jane Street and Magnetar Capital, and its biggest clients are said to be Microsoft and OpenAI. Sure, it’s not Nvidia or Meta, or even Oracle, but it’s deeply embedded in the AI ​​ecosystem.

Third, the reaction was hardly as violent or as broad. The Nasdaq 100 fell just 5.5 percent last week and started this week in good shape. Oracle’s bond yields also jumped, but they fell for other hyperscalers thanks to the recovery in the Treasury market. This is not a massive panic price overhaul that we saw in March 2023.

So what does Rajadhyaksha think? would be Could this be a more ominous harbinger of the AI ​​investment holiday? All signs that the cost of computing is plummeting. But this is not yet the case, he believes:

CoreWeave’s business model – and the business model of every AI infrastructure company – is a widespread business. Buy GPUs, rent them, pocket the difference between rental yield and financing cost. This gap persists as long as demand for computing exceeds supply and rental rates remain high enough to cover debt service, depreciation, and operating costs.

If compute prices collapse – if the market for GPU hours falls enough that the economics of building no longer work – then the bears are right, and CoreWeave bonds are the canary in the coal mine.

But so far, that just hasn’t happened. TSMC’s results tell you that demand for chips is increasing. ASML’s delay tells you manufacturing capacity is limited for years. The hyperscalers’ capital spending numbers — $725 billion to $800 billion in 2026, with estimates of $1 trillion or more in 2027 — tell us that compute buyers aren’t backing down. CoreWeave’s own backlog stands at $104 billion in take-or-pay contract revenue.

Unless the spot and futures price of compute declines drastically – and nothing in the supply chain data supports this conclusion – we think last week’s pullback in AI infrastructure stocks should abate. Quality should reassert itself, even if investors are more wary of margins and wary of neoclouds. Just as the 2023 recession never happened, we don’t expect the AI ​​investment cycle to break. What broke was a small, subordinate part of a company’s capital structure. Investors, as they did in March 2023, should not confuse the part with the whole.

The market response so far this week suggests that investors broadly agree with Rajadhyaksha, that it was a flurry and nothing more. But the complex economics of the calculation may be worth watching.