Nasdaq’s 5% Drop Recasts the Risk in AI Infrastructure Spending

The Nasdaq fell 5% on November 5, pushing the tech-heavy index into correction territory, which John Coogan described as a decline of at least 10% from a recent high. The Dow was down about 1,600 points and the S&P 500 about 4%, on pace for its worst day since April 2025. Coogan cited Jay Hatfield of Infrastructure Capital Advisors, who called it “a true panic” and said the move was about chip stocks and the possibility of a slowdown in AI infrastructure spending after the midterm elections.

The timing was part of the puzzle. Prediction markets had pointed to a Democratic wave before Election Day, and Jordi Hays noted that the result in Texas was known by election night or the next morning. But the market did not immediately make a decisive move. Hays found that delay notable in a market capable of moving hundreds of billions of dollars—or more—on algorithmic trading. Investors, he suggested, still needed time to work out what the result meant for their positions.

The day’s losses erased the previous day’s post-election gains. The Nasdaq 100 was said to be roughly 6% from a bear market. Yet Hays also pointed out that the Nasdaq remained up 5% for the year. Coogan said the contrast showed how much optimism had already been priced in, even as technology adoption continued. The hosts discussed a sharp reversal in market expectations, not evidence that adoption had stopped.

Reports of fund liquidations added to the sense of pressure, though the hosts emphasized that some claims circulating online were unconfirmed. Posts said Walleye and Engineers Gate were shutting down equity pods, while other posts claimed that Lone Pine had liquidated. Hays said Walleye had issued a letter about closing one long-short strategy while continuing to focus on other products, but said the details of the other reports might not be known for at least 24 hours.

The hosts were particularly cautious about online claims that Leopold Aschenbrenner’s fund, Situational Awareness, had “blown up.” Hays said people did not know what Aschenbrenner was doing and that it was fair to challenge posts stating the fund had collapsed. Coogan noted that some of the stocks named in the rumors were down 50% or 60%, and Hays said such a portfolio would be under real pressure if the thesis remained unchanged and leverage was involved. But they also allowed that Aschenbrenner might be out of the market. The selloff could produce liquidations; the hosts did not treat every liquidation claim as confirmed.

The fear was concentrated in the infrastructure trade. Coogan said the largest technology companies were holding up better than many chipmakers and data-center suppliers. One explanation discussed was that hyperscalers already have facilities, power, and sites in place; they may be able to put newer chips into existing data centers rather than depend as heavily on new projects and permits. Hays added that companies with capacity could potentially sell into spot demand. Coogan cautioned that reconfiguring older facilities could be slower than the approach of “neocloud” providers, but said he expected announcements in the coming weeks about where those companies planned to build next.

Those plans face a political constraint as well as a financing one. Coogan argued that projects could win local support if they bring community benefits, use clean power, operate quietly, or are built in remote locations. But the election results, he said, suggested that voters in several states object to the existing terms of the data-center buildout. Some concerns may be based on misunderstandings, he allowed; others are legitimate.

The hosts also cited a post from investment firm Citrini saying it had revised its thematic model book and judged the risk around AI infrastructure spending to have shifted to the downside. Citrini expected higher U.S. data-center costs after Democratic midterm wins and warned of a late-cycle stagflationary macro shift. That view sat alongside a bullish argument from Lambda chief Stephen Balaban: local resistance could constrain supply of compute just as demand continues to grow. The hosts presented competing views about whether demand and local constraints would support the economics of new infrastructure.

For John Coogan and Jordi Hays, the market’s deeper concern was not just the cost of data centers. It was what comes after AI coding as a major source of revenue for OpenAI and Anthropic. Coogan cited candidates including AI for science and chip design, but said none had yet given investors a clear answer about when those applications would become meaningful revenue. Rumors of breakthroughs, he said, were not the same as knowing when they would produce money.

Hays framed the commercial constraint in terms of spending per employee. Many businesses, he said, want to use more AI but cannot yet identify a way to spend much more than a few hundred dollars a month per person. AI spending remains small beside other business costs; the willingness to spend more may exist, but the applications are not there yet across enough kinds of work.

That gap is especially visible outside software engineering. Coogan said an engineer earning a high compensation package could plausibly justify a large annual token budget if the tools make the engineer more productive. He invoked Jensen Huang’s suggested one-to-one relationship between compensation and AI spending, saying it did not sound implausible for some roles. It is harder to make the case for a lawyer, marketer, or finance executive. Companies might want to give those employees $20,000 a month to spend on AI, Coogan said, but today the budget may be closer to $200 or $2,000.

Applications in science and medicine could eventually support larger spending, but they have their own adoption timelines. Coogan noted that a drug designed with AI would still face demands for human evidence and lengthy trials before broad use. At the same time, he pointed to GLP-1 drugs as an example of a market assigning substantial value before widespread access. A breakthrough in science or chip design could still support financing ahead of mature adoption, he argued, even if the revenue path is slower.

The same uncertainty shaped the discussion of an Anthropic IPO. Jordi Hays cited Jim Cramer’s view that Anthropic might need to accept a discount, while pushing back on the idea that a lower valuation would necessarily be a problem for a well-capitalized company. If the company’s last financing was around $960 billion, Hays reasoned, an IPO at $1.2 trillion would still be an increase for those investors—even after expectations had risen toward $2 trillion.

John Coogan saw a different risk in the timing: whether the company could go public while businesses in its supply chain were sharply falling. He could also imagine investors looking past short-term losses and treating a lower IPO price as an opportunity. Neither host claimed to know which response would dominate.

The possible offering has levers beyond valuation. Hays suggested Anthropic could reduce the number of shares offered as well as lower the price. But he noted that a smaller offering or discount would not be ideal if the company needed to raise capital or provide liquidity. The hosts also discussed the possibility that retail investors could account for a larger share of demand than in other recent IPOs, particularly if the price were attractive and the company’s social-media marketing had made it widely known.

Coogan cited a prediction-market estimate that the chance of an IPO before December 1 had fallen from 88% on November 3 to 61%. Hays said the move might reflect market sentiment, or the judgment of people with experience taking technology companies public. The hosts did not know which explanation was behind the shift.

A further concern was the location of data centers. Hays described Cramer’s focus on facilities in Texas for IREN and CoreWeave, and noted that Nebius appeared to have less Texas exposure. The question for investors, he said, was whether projects still under construction could get approvals, face a pause, or have their terms renegotiated. He also pointed to messaging around protecting power costs and making water use more explicit—evidence, in his view, that companies were already trying to address some of the objections.

A Wall Street Journal interviewer challenged Take-Two CEO Strauss Zelnick on whether the company depended too heavily on Grand Theft Auto. Zelnick answered with a list: 15 franchises had each sold at least five million units in an individual release, he said, including Red Dead Redemption, NBA 2K, WWE 2K, BioShock, Borderlands, Civilization, and mobile games such as Words With Friends. He cited more than 65 million units sold for the most recent Red Dead Redemption release.

Jordi Hays acknowledged the force of the question even while approving of Zelnick’s answer. GTA is an unusually large franchise, he said; its scale makes the company’s catalogue look different to an audience that knows the name GTA but may not follow the wider games business. John Coogan added that GTA 6 must succeed on two related but distinct fronts: the single-player game, with its story and mechanics, and the ongoing business of GTA Online. A strong game can keep players engaged and support spending, but each is a different creative and commercial challenge.

The exchange also opened onto the argument about AI in games. Coogan said Zelnick had described GTA as carefully handcrafted while also observing that games have long been made with computers. The hosts pushed back on calls for “no AI” in video games by distinguishing generative AI from the systems that already govern computer-controlled characters. Enemies and other characters make decisions in the game world; remove those systems, Coogan argued, and that world becomes empty. The debate over new generative tools is different, but the phrase “AI in games” can obscure how much computer-driven behavior is already part of the medium.

Take-Two was up 10% over the previous five days, Coogan noted, while much of the market was falling. Hays pointed to the company’s ability to produce hits and its range of established intellectual property as reasons investors might see it as a comparatively durable business.

A Wall Street Journal report on comments by OpenAI president Greg Brockman described a difficult first half of 2026, when Anthropic had gained ground with business customers. Jordi Hays said OpenAI’s coding tools found traction in the third quarter, with enterprise revenue growth accelerating sharply. He emphasized the change in growth rates, not just the companies’ relative revenue: OpenAI, he said, reported more than 100% quarter-on-quarter enterprise growth, while Anthropic’s growth had slowed to around 49%.

Brockman’s account, as relayed by Hays, linked the change to new coding tools and a new frontier model. Hays described it as a “revenue inflection,” following behind-the-scenes work that had not been visible in the first half.

The hosts then turned to Dot, an iPhone-based assistant Hays had been using. He described new integrations and ways to call it from the iOS Action button, along with faster pickup. The delay when calling had felt awkward, he said: unlike a chat interface that immediately acknowledges a request, the phone rang while the system spun up the necessary models and computing resources.

Hays’s more substantive use case was calendar synthesis. He wanted Dot to combine personal and work calendars into a readable daily plan—something that could tell him when a show ended, when he needed to leave, and what else was happening at home or while traveling. Dot could already produce and text him that summary, but the iOS widget showed only the character he had created. He wanted useful, changing information directly on the home screen, without building an app himself. The example illustrated the product question he was testing: whether an assistant could turn a general capability into practical, repeated use.

John Coogan cited a Bloomberg report that an anonymous seller was seeking buyers for a portfolio of about $500 million in data-center loans, in a rare auction days after the election. The loans included a $25 million slice tied to a New Mexico data-center site associated with the OpenAI, SoftBank, and Oracle Stargate project. Coogan noted that Oracle had sent the developer a notice citing force majeure. The portfolio also included debt tied to operators such as Vantage Data Centers and CyrusOne.

Bids were due November 19, with a reported minimum bid of $10 million. The hosts did not know who was selling or what the sale signaled about the broader market. The listing put a specific financing question alongside the day’s stock moves: what confidence investors still have in the debt behind the data-center expansion, and at what price they are willing to take that risk.