AI's "Safe Bet" Stocks Are Cracking — and CoreWeave Is Ground Zero

Okay, so this one actually surprised me — and I don't say that lightly after watching this market for years.

CoreWeave, the AI cloud infrastructure darling that seemed untouchable just a few months ago, had its bonds trading at 81 cents on the dollar on Wednesday. That's not a minor wobble. That's a company's debt pricing in serious distress — yielding over 15% — on bonds that were issued only a few months ago. A prominent chip ETF is now on track for its eighth consecutive down day. Anthropic, in a private investor meeting, quietly confirmed its Q3 revenue growth is slowing. And all of this is happening against a backdrop of hot PPI data and a bond market that's been flashing warnings for weeks.

The AI trade — the one that felt like a one-way bet for the better part of two years — is unwinding. And this week, that unwind moved from "uncomfortable" to something that deserves real attention.

The CoreWeave Number That Should Make You Stop Scrolling

Let me sit with that CoreWeave bond price for a second, because 81 cents doesn't sound catastrophic on its face. But here's what it means: when you buy a bond at issuance for $1,000 and it's now trading at $810, the market is telling you it wants a dramatically higher return to hold that risk. At 15%+ yield, the credit market is essentially screaming that CoreWeave's ability to service its debt is genuinely in question.

This matters beyond CoreWeave itself. The company is a "neocloud" — it rents out Nvidia GPU clusters to AI companies that don't want to build their own data centers. Their entire business model is predicated on AI spending staying elevated, ideally forever. Traders, not exactly known for subtlety, have been loading up on what they're calling "bubble puts" on CoreWeave — options designed to profit if the stock falls further. That's not hedging. That's conviction that things get worse.

And here's what I actually think about this: CoreWeave's bond stress is a canary. If AI infrastructure spending is solid and growing, these things don't fall to 81 cents. They just don't. The fact that professional credit investors — people who do this for a living and are generally less prone to narrative-driven optimism than equity folks — are bailing tells you something the stock price alone might not.

Anthropic's Revenue Update Changed the Mood

Now here's where it gets interesting.

Anthropic — maker of Claude, one of the main GPT competitors — gave a quarterly revenue update in a private investor meeting. The numbers weren't secret for long: Q3 revenue came in at $16.8 billion for the September quarter. That sounds like a big number. But the detail that leaked out and hit the tape? Growth is slowing.

For context, the AI investment thesis has always leaned heavily on exponential adoption curves. If you're paying $40B+ valuations on AI companies, you need the revenue chart to be going straight up. A slowdown in growth — even from an already large base — breaks the math that justified those valuations.

Wall Street has a funny way of forgetting that "still growing fast" and "growing fast enough to justify the price" are very different things. And investors, apparently, are remembering that distinction this week.

The ripple effect ran straight into chip stocks. Micron, Nvidia, and the broader semiconductor complex faced renewed selling pressure Thursday. The chip ETF drawdown — eight days in a row — is the longest consecutive losing streak in over a year for that sector. When Anthropic's spending slows, the demand for the Nvidia H100s and Micron's HBM memory that power those AI systems gets questioned. It's all connected.

Who's Actually Buying Right Now

If money is leaving AI and chip stocks, it has to go somewhere. Investors this week have been piling into government bonds and defensive stocks — utilities, consumer staples, healthcare names. Which is a reasonable move, except for one complication.

Treasury yields are sitting above 5% on the 10-year. The U.S. is running a $2 trillion deficit. The IMF literally issued a warning about U.S. fiscal sustainability this week (which the Treasury dismissed somewhat sharply, but still — when the IMF is flagging you, you've got people's attention). The federal debt is above $40 trillion.

Buying Treasuries as a "safe haven" trade while 10-year yields are at 5% and the deficit trajectory is what it is — that's not exactly a no-brainer. You're getting paid more than you have in years to hold that debt, sure. But the pile of that debt is enormous and growing. There's a real scenario where the "rotation into safety" trade turns out to be jumping from one tricky situation into another.

Defensive stocks like Walmart at least have a concrete story: Deutsche Bank upgraded them to buy Thursday, arguing that persistent inflation is actually driving more cost-conscious consumers through Walmart's doors. That's a real, earnings-driven thesis right there.

What's Underneath the AI Selloff

Look, I could be wrong here, but I don't think this is just profit-taking or a normal pullback. The structure of this breakdown looks different.

When you see:

• Equity investors dumping AI and chip names for 8+ straight sessions
• Credit markets pricing CoreWeave debt at 15% yields
• Options traders aggressively buying downside protection (the "bubble puts")
• Private investor meetings leaking data showing slowing growth at a top AI startup

...that's not a random bad week. That's multiple markets, using multiple instruments, coming to a similar conclusion at the same time. Each of those signals in isolation is noise. All of them together, in the same week, pointing in the same direction — that's information.

The AI infrastructure buildout has consumed staggering amounts of capital. Nvidia's data center revenue, Micron's HBM sales, CoreWeave's GPU leases — all of it was predicated on AI application revenue growing fast enough to justify the spend. If Anthropic's Q3 is showing cracks, and if the broader enterprise adoption of AI is slower than the infrastructure spending implied, then there's a genuine mismatch between the supply of AI horsepower and the actual demand for the output.

And I'll be honest — this is the part that genuinely worries me. Not because AI is going away. It isn't. But because markets priced in a lot of perfection, and perfection rarely shows up on schedule.

Going a step further: the hot PPI report that also hit this week isn't helping. If inflation stays elevated, the Fed's path toward rate cuts closes off, which keeps borrowing costs high across the board — including for the capital-intensive AI infrastructure companies that burned cash to build out. Higher rates and slower AI revenue growth is a rough combination for a sector that assumed both would get easier.

Has This Kind of Repricing Happened Before?

Yeah. A few times.

The dot-com bust of 2000–2002 is the obvious reference point, but that's a bit extreme — that was a near-total wipeout of speculative tech. A closer analog might be the cloud computing selloff of 2022, when the "everything cloud" trade peaked out and names like Snowflake, Cloudflare, and Datadog fell 50–75% from their highs before eventually recovering. The underlying technology didn't stop being real or valuable. But the valuations had overshot what the actual revenue trajectories could support.

The neocloud stocks — CoreWeave and its peers — occupy a similar space today. They're real businesses doing real things. CoreWeave's revenue isn't zero. But when your bonds are yielding 15%, the market is saying your business model has execution risk that the equity price maybe hasn't fully priced in yet.

And this is where I think most people get it wrong: they treat a high-yield bond price as a company-specific problem, when really it's a signal about the entire ecosystem's health. If CoreWeave's customers were flush with cash and confidently expanding, CoreWeave's debt wouldn't be trading at a discount. The bond market sees something the bullish equity narrative wants to ignore.

That said — and I want to be fair about this — some of the selling could absolutely be overdone. Eight down days in a chip ETF is genuinely brutal, and markets often overcorrect in both directions. Nvidia isn't Pets.com. But "the technology is real" has never been sufficient justification for any particular price.

What the S&P 500's Worst Day Since July Is Actually Telling You

Wednesday was the S&P 500's worst single day since July. That's not a catastrophe — it's not 2022, it's not March 2020 — but it's a signal the AI trade had become large enough that its repricing moves the whole index.

One veteran market watcher with 30 years of experience said this week he's "alarmed" by the S&P's record-chasing behavior and is actively looking elsewhere. There's also a hedge fund manager making the rounds saying there's a "monstrous bull market" developing in smaller names underneath the surface, even as the megacaps drag. Both takes have some merit. The index is so concentrated in a handful of AI-adjacent names that their pain becomes the market's pain.

If you've been in an S&P 500 index fund this year feeling pretty good, it's worth understanding that your exposure to Nvidia, Microsoft, and the rest of the AI infrastructure stack is larger than you might think. When they sneeze, your portfolio catches a cold. That's not a reason to panic-sell — it's just reality to sit with. We've talked before about how concentrated the index actually is, and this week is a pretty vivid illustration of that.

What to Watch From Here If You Own Any of This

If you hold chip stocks or AI infrastructure names directly:

• Watch CoreWeave's bond price as a real-time sentiment gauge — if it stabilizes above 85 cents, that's a sign the credit fear is abating; if it breaks below 75, buckle up
• Keep an eye on Anthropic and OpenAI's next reported revenue figures — those are now market-moving data points even though they're private companies
• The next Fed meeting is in two weeks, and at least one Fedwatcher is now calling for a rate hike — a hike would apply additional pressure to growth-stage AI names

If you've rotated into Treasuries as a flight to safety:

• The 5%+ yield is genuinely attractive versus recent years, but the $2T deficit and $40T debt pile are real tailwinds for yields staying high, which means bond prices could keep falling — you're catching yield but potentially still losing on price if you're in long-duration bonds
• Short-duration Treasuries (2-year, 1-year) give you the yield with far less price risk right now; that's where I'd be sitting personally

If you're in a standard S&P 500 index fund and just watching:

• A sector rotation toward defensives (utilities, staples, healthcare) doesn't require you to do anything, but it's worth knowing what's driving your index — a hedged allocation to value-oriented names might smooth the ride

For anyone considering buying the AI dip:

• Don't conflate "the technology is real" with "this price is justified" — those are two separate questions, and Tuesday's market answered the second one pretty clearly
• If you do buy, the chip equipment makers (the companies that make the machines that make the chips) tend to be more insulated from hyperscaler spending swings than the hyperscalers themselves