Newsletter · 2026-09-27 · Ashutosh Agarwal

Bill Gurley Called CoreWeave the Canary and It Stopped Singing - Track Record - October 2025 to September 2026

A historical scorecard of Benchmark's Bill Gurley, who on the BG2 podcast on October 14, 2025 flagged AI's circular financing deals and named CoreWeave as the pure play to watch, measured through the September 25, 2026 cutoff. CoreWeave fell about 35% while the S&P 500 rose about 16% and Nvidia, whose deals he criticized, rose about 25%. Verdict: Right on CoreWeave, too early on the broader overbuild thesis.

Track Record

Bill Gurley, October 2025 to September 2026

This is a historical review of a call made on October 14, 2025, with performance measured through the September 25, 2026 cutoff, about eleven months. It was published on September 27, 2026; the prices and grades below are as of the September 25, 2026 cutoff.

This issue: Benchmark's Bill Gurley, and his October 2025 warning that AI's circular financing deals raised the odds of an overbuild, with CoreWeave named as the pure play to watch. Verdict: Right on the specific, checkable CoreWeave call, too early on the broader overbuild thesis.

On October 14, 2025, Bill Gurley sat down for what he said would be his last regular episode of BG2, the podcast he co-hosted with Altimeter's Brad Gerstner. He was leaving to finish a book and work on policy.

He didn't sign off quietly. He spent his last big segment raising red flags about how the AI boom was being paid for, and he named one stock as the place to look first when things started to go wrong.

That stock was CoreWeave ($CRWV). It closed at $134.06 that day. On Friday it closed at $87.59.

The Call

Who: Bill Gurley, general partner at Benchmark. He was an early investor in Uber, Zillow, Grubhub and OpenTable. Where: BG2Pod with Brad Gerstner and Bill Gurley, "AI Bubble, Stablecoin Boom, and Runnin' Down a Dream" (October 14, 2025). Ticker: $CRWV (CoreWeave). The wider target was the AI buildout as a whole.

Gerstner opened the segment by pointing to that morning's OpenAI and Broadcom chip deal, which added to what he called "well over $1 trillion of incremental CapEx." (CapEx means capital spending: money spent on data centers, chips and power.) Then he asked Gurley to lay out his worries.

Gurley's main point was about circular revenue. That's when a company invests money in a customer and the customer uses that money to buy the company's products. The sale shows up as revenue, but some of it is really the seller's own cash coming back. Gurley said he described these deal structures to ChatGPT without naming any companies, and the AI brought up Enron and WorldCom on its own:

"There's like 6 different transactions that have happened now that I would say are non-normal... And the AI itself, you know, would find its way toward company names like Enron and WorldCom and those kind of things merely by describing the type of transaction."

He traced the habit back to the original Microsoft and OpenAI deal, where Microsoft invested in OpenAI partly in cloud credits that OpenAI then spent on Microsoft's Azure cloud:

"That's a cashless transaction. Like there's no cash, but it becomes an income statement revenue item for Microsoft. And I don't think that's ideal from an economic standpoint."

He also called out Meta for agreeing to cover the debt on a data center it doesn't technically own: "to me, that's classic off-balance sheet financing."

Then he made the call. He didn't predict a crash on a date. He said these deals make an overbuild more likely, and more likely to stay hidden for a while:

"I think it increases the chance that we go over the top, that, that we end up overprovisioning. And I, I kind of felt like that was unavoidable anyway, but now I think it's higher. And, but I think it maybe pushes out when we find out that happens, because you've, you've just created more virtual leverage on the whole system, and you might be hiding some of the signs that would tell you things are slowing down."

Then he named the stock:

"One of the more peculiar of all the deals is, and this was disclosed in a CoreWeave filing, was NVIDIA has promised to buy any of CoreWeave's service availability that they can't sell to anyone else. That is very unusual... It also means as an investor, we don't know what's going on with real demand for CoreWeave... And if you'd have said to me, what would you look for to see if we've, you know, kind of reached a point where things are slowing a little bit, you'd say, well, let's look at one of the pure plays."

What he predicted, in plain English:

• Direction: Cautious to bearish on the most indebted, most financially engineered parts of the AI buildout. CoreWeave was his example of that, and of the stock that would crack first.
• Thesis: Circular deals and off-balance-sheet financing make overbuilding more likely, and they hide the warning signs. CoreWeave is a "pure play" (a company that does only one thing, here renting out AI computing power), and Nvidia's promise to buy its unsold capacity makes its real demand hard to see.
• Timeframe: None given. He said the problem might take longer to show up ("pushes out when we find out").
• Price target: None. Gurley is a venture capitalist, not a short-seller, and he never said "sell."

Gerstner pushed back on the air: "I'm not concerned as a shareholder in NVIDIA with what I'm seeing NVIDIA do today." He did agree that yellow flags could show up "further and further to these startup neo clouds."

What Happened

All prices are daily closes. The S&P 500 is the benchmark.

CoreWeave broke down almost right away. Over the two and a half months after the podcast, the stock lost 46.6%, falling from $134.06 to $71.61 by New Year's Eve. The S&P 500 gained 3.0% over the same stretch. The monthly closes show the drop: $133.71 at the end of October, then $73.12 at the end of November.

It then went sideways at the lower level. CoreWeave bounced to $111.60 by the end of April 2026, then slid to its lowest close in the whole period, $60.82 on July 29, 2026. That was 54.6% below where it stood on the day of the call. It has since climbed back to $87.59.

Against the market, the gap is 51 percentage points. CoreWeave fell 34.7% while the S&P 500 rose 16.5% and set a record close of 7,798.99 on August 13, 2026.

Oracle fell even further. Gurley didn't name Oracle on October 14. It is, though, the clearest example of the debt-funded buildout he was describing. Oracle closed at $299.00 on the day of the call, peaked at $313.00 two days later, and now sits at $137.10, down 54.1%.

The company he criticized most did fine. Nvidia, whose CoreWeave backstop he called "very unusual," is up 25.0% since the call and ahead of the S&P 500. The circular deals did not sink the chipmaker.

The Verdict: Right on CoreWeave. The bigger thesis is Too Early.

What he got right:

• He named the right stock at nearly the right moment. CoreWeave's closing high for the whole period was $143.08 on October 9, 2025, three trading days before the podcast. Gurley pointed to the most fragile name in the AI trade within days of its top, and it lost nearly half its value in 11 weeks.
• He had the reason right. The market started treating debt-funded, customer-concentrated neoclouds (cloud companies built only to rent out AI chips) very differently from the cash-rich giants. That's exactly the split he described: worry about the companies with thin balance sheets, not the ones generating hundreds of billions in cash.
• "Pushes out when we find out" fit what happened. CoreWeave bounced back in April and May 2026 before making a new low in July. That's messy, drawn-out price action, and it matches his warning that the damage would be hard to time.

The uncomfortable parts:

• This wasn't a short call, and we shouldn't grade it like one. Gurley never said "sell CoreWeave" or gave a target. He said it was the stock to watch. Treating a warning as a trade gives him extra credit he didn't ask for.
• The broad overbuild hasn't happened yet. His headline worry was that the industry would "go over the top" and overbuild. Eleven months later the S&P 500 is at record highs and Nvidia is up 25%. There's no sign of a sector-wide glut. On the big picture, he's early at best.
• His target was Nvidia's deals, and Nvidia's shareholders came out ahead. If you had sold Nvidia because of Gurley's circular-revenue argument, you'd have missed a 25% gain.
• He's an interested party. Benchmark is a venture firm that competes for AI deals. That doesn't make him wrong, but readers should know it.

Grade: RIGHT on the specific, checkable call (CoreWeave). TOO EARLY on the broader overbuild thesis.

Have They Changed Their Tune?

No. He has repeated the warning, more strongly each time.

January 2026: he says the correction has already started. On Power Players with Brian Sozzi, "Why Bill Gurley says AI investing is becoming crowded" (January 26, 2026), Gurley pointed to the damage as proof:

"We've already seen some types of corrections. I think the Oracle boom-bust thing happened very fast. CoreWeave stock is way off of what it used to be. We had the Rick Perry SPAC thing, data center thing that went up and way down... You know, I do think that there will eventually be a correction."

He also offered a new reason. AI startups are "losing massive amounts of money, more than Uber ever lost... more than Amazon ever lost." When investors stop funding those losses, he said, the companies "have to do what Uber did, which is transition from losing money to being cash flow positive. And once, as they do that, the pricing for these products is going to shift." He called out OpenAI as a risk to the whole system: "It's almost like everyone's dependent on them surviving." On data centers: "just the pure mental exercise of assuming that they would make it on budget, on time, is probably incorrect."

What happened next: From that day's close of $98.31, CoreWeave is down 10.9%, while the S&P 500 is up 11.4%. Oracle is down 24.9% (from $182.44). The second warning also lagged the market.

March 2026: his harshest version yet. On Prof G Markets, "The AI Divide: Who Wins and Who Gets Replaced, ft. Bill Gurley" (March 27, 2026), talking with Ed Elson and Scott Galloway, he dropped the careful wording:

"I think the circular deals are horrific. I don't think the auditor should have approved them. And I think whenever you eventually have an unwinding, they're going to make it worse because they won't be sustainable. You shouldn't be able to move cash from your balance and create revenue on your income statement."

He again singled out the Nvidia and CoreWeave arrangement as "probably most questionable." He also gave a reason for Nvidia's stock that is easy to test: "People want to understand why the NVIDIA multiple won't go higher. I think it's the circular deals." (A multiple is how many times its earnings investors will pay for a stock.) He summed up where things stand: "it's become normalized... and it's gonna stay normalized until it's not."

What happened next: This time the timing was poor. He doubled down close to a bottom. The S&P 500 made its low for the period on March 30, three days later. CoreWeave has gained 17.1% since that podcast (from $74.81), and Nvidia has gained 34.4% (from $167.52). Both are close to or ahead of the S&P 500's 21.6% rebound. Anyone who read his March comments as a reason to sell AI stocks lost money.

Where he is now. His latest big podcast was All-In, "Bill Gurley: Searching for Feynman" (September 19, 2026). He spent it on regulatory capture, meaning industries that end up controlling the agencies meant to oversee them. His main example was Boeing and the FAA, where he said "Boeing and the FAA were working as one." In a separate talk on Money Rehab with Nicole Lapin, "Legendary Venture Capitalist Bill Gurley on the AI Bubble, Why IPOs Feel Rigged and How to Find Your Dream Job" (March 4, 2026), he gave everyday investors a simple warning: "It is a highly speculative time right now... I would be cautious in general as a retail investor, even towards AI." He added that if you own an index fund, "you have pretty good AI exposure already."

Running Scorecard

Bill Gurley

Track Record so far, across all investors

Next issue: Cathie Wood (ARK Invest).