Powerpei's Profile | Binance Square

Last week, I wrote about CoreWeave. My closing line was: Stress-testing a business model with fear gets you closer to the truth than hyping its valuation with enthusiasm. This week, the test spread. CoreWeave fell nearly 12% in a single day, bringing its losses this week to around 25%. Bond yields kept climbing. Last week, it was one company’s debt pressure. This week, the entire AI infrastructure chain is being reassessed. The Philadelphia Semiconductor Index plunged, and memory stocks got hammered too. Micron, AMD, Broadcom, Intel, SK Hynix, SanDisk—the chip stocks fell across the board. Most big tech stocks closed lower: Meta -3.7%, Google -2.5%, Tesla -2.1%, Nvidia -1.4%. Apple was the only one to edge up, +0.43%. // I wrote in my last post that neo-clouds sell “certainty”—and the higher the leverage, the more fragile they are. This week, rising rate expectations and higher financing costs were added to the mix. That fragility spread from one company’s balance sheet to share prices across the entire chip chain. // Jensen Huang responded publicly, saying these fluctuations “will pass.” Nvidia sells the shovels and has backstop agreements, so he can afford to say that. But the market is testing whether downstream customers can repay their debts. That Gartner figure is still hanging over the sector: 80% of neo-clouds could shut down by 2030. For downstream companies bearing financing costs above 15%, that statement is, for now, just a hope. ———— Apple’s slight gain was, by contrast, the clearest signal. Everything getting hammered was tied to AI computing power and infrastructure leverage. Money moved into big tech furthest removed from AI infrastructure as a safe haven. The keyword in my last post was “if.” This week’s market action is effectively telling you that more and more people are betting that the “if” won’t hold. // Utilization, contract quality, and cost of capital. In my last post, I said these three things were being tested at CoreWeave. This week, the whole chain is being tested. Those who can withstand this round of interest-rate pressure will stay at the table. Those who can’t will be forced out. This filtering process may take a while. Data is based on public disclosures from various parties and does not constitute investment advice. #CoreWeave #AI基建

Last week, I wrote about CoreWeave. My closing line was: Stress-testing a business model with fear gets you closer to the truth than hyping its valuation with enthusiasm.

This week, the test spread.

CoreWeave fell nearly 12% in a single day, bringing its losses this week to around 25%.
Bond yields kept climbing.
Last week, it was one company’s debt pressure. This week, the entire AI infrastructure chain is being reassessed.

The Philadelphia Semiconductor Index plunged, and memory stocks got hammered too. Micron, AMD, Broadcom, Intel, SK Hynix, SanDisk—the chip stocks fell across the board. Most big tech stocks closed lower: Meta -3.7%, Google -2.5%, Tesla -2.1%, Nvidia -1.4%.

Apple was the only one to edge up, +0.43%.

//

I wrote in my last post that neo-clouds sell “certainty”—and the higher the leverage, the more fragile they are.
This week, rising rate expectations and higher financing costs were added to the mix. That fragility spread from one company’s balance sheet to share prices across the entire chip chain.

//

Jensen Huang responded publicly, saying these fluctuations “will pass.”

Nvidia sells the shovels and has backstop agreements, so he can afford to say that. But the market is testing whether downstream customers can repay their debts. That Gartner figure is still hanging over the sector: 80% of neo-clouds could shut down by 2030. For downstream companies bearing financing costs above 15%, that statement is, for now, just a hope.

————

Apple’s slight gain was, by contrast, the clearest signal. Everything getting hammered was tied to AI computing power and infrastructure leverage. Money moved into big tech furthest removed from AI infrastructure as a safe haven.

The keyword in my last post was “if.”
This week’s market action is effectively telling you that more and more people are betting that the “if” won’t hold.

//

Utilization, contract quality, and cost of capital. In my last post, I said these three things were being tested at CoreWeave. This week, the whole chain is being tested.

Those who can withstand this round of interest-rate pressure will stay at the table. Those who can’t will be forced out.
This filtering process may take a while.

Data is based on public disclosures from various parties and does not constitute investment advice.

#CoreWeave #AI基建

Powerpei

CoreWeave fell 15.5% in a single day on Wednesday, closing at $65.78. The yield on its debt due in 2032 surged past 15%, and its CDS spread jumped above 1,000 basis points. Based on market pricing, the probability of default within five years is around 63%. Chamath posted a chart, Musk replied “Yikes,” and timelines were flooded with claims that leveraged AI infrastructure was about to collapse.

But the same company has around $129 billion in contracted revenue backlog, with Meta accounting for 27% and OpenAI for 17%. BofA has a Buy rating and a $140 price target, implying 105% upside. The CEO himself says the market is “irrational.” The bond market thinks it’s about to die; equity analysts think it could double—and they’re looking at the same financial statements...

What CoreWeave does is different from traditional hardware sales. It borrows money up front to stock up on GPUs, packages them as computing services, and rents them to customers through long-term contracts. Its balance sheet is full of debt, but the revenue side is supposed to turn into stable, recurring cash flow. Why are customers willing to sign long-term contracts? Because building their own clusters is too slow, too expensive, and too uncertain. In effect, CoreWeave takes on that uncertainty first by leveraging its own balance sheet, then sells the locked-in computing capacity. The more leverage it takes on, the more certainty it can offer—but the more fragile it becomes.

A few details are worth noting. Gartner says around 80% of neocloud providers could shut down by 2030. Nvidia and CoreWeave have a backstop agreement: Nvidia will take over any computing capacity that remains unsold by April 2032. This has barely been discussed, but it may be the most important safety net in the whole story. There have also been reports that OpenAI’s actual revenue is $20 billion lower than previously estimated. If expectations for demand from the largest customer are weakening, then the $129 billion contracted backlog deserves a discount.

The fear-driven narrative is certainly alarming, but it’s also overly simplistic. If the contracted backlog converts on schedule, then interest coverage is just a matter of timing. The key word is “if.” Whoever can survive this surge in the cost of capital will have the strongest moat in those long-term contracts. For now, let’s keep watching.

Data is based on public disclosures from the respective parties and does not constitute investment advice.

NVDAB-0,55%

METAB-0,36%

AAPLB-0,16%

Powerpei

Trending Topics