Is The AI Trade Over? | Jack & Max on Panic in AI Credit, SpaceX Earnings, and Semis

Jack does not think the AI trade is over, and Max said an AI-trade collapse is not his base case. Max viewed SpaceX’s earnings call as bullish, while both saw upcoming large-company reports as important evidence for the outlook. Elon Musk discussed SpaceX reaching $500 billion to $1 trillion in annualized recurring revenue by the end of 2027, a target Jack stressed was not his base case. SpaceX reported 161% revenue growth, while superintelligence revenue rose 195% quarter over quarter and 830% year over year from a relatively small base. Anthropic accounts for roughly half or slightly more of that revenue, with SpaceX expected to receive about $1.2 billion to $1.3 billion per month from the customer through 2029.

Musk said SpaceX expects to bring 10 gigawatts online in 2027, in addition to 2.5 to 3 gigawatts this year, and would be surprised to have less than 13 gigawatts by the end of 2027. Jack said the plan also involves another 20 gigawatts of power and cooling to support more than 13 gigawatts of chips. If SpaceX delivers the promised scale and timing, that capacity could address concerns about whether AI chips will have enough data centers, land, and grid-connected power. Jack contrasted the plan with credit-market anxiety: CoreWeave’s unsecured 2032 bonds implied roughly a 60% cumulative default probability through 2032, while SpaceX’s compute deal implied a GPU rental price of about $5.30 per hour. He said CoreWeave stock and bonds would be buys if that price were a reasonable baseline; Max said the company may be fine at SpaceX’s stated compute costs, but the case depends on prices staying high rather than cheaper models making current spending excessive.

The semiconductor index fell for nine consecutive days, its longest such streak since 2006; Jack clarified that the record was for the length of the decline, not its percentage. He cited Anthropic’s below-expectations Q3 revenue and CXMT’s larger-than-expected planned memory-production increase as bearish factors in the October selloff. Anthropic revenue rose from $4.7 billion in Q1 to $11.6 billion in Q2 and $16.8 billion in Q3, although quarter-over-quarter growth slowed from 147% to 45%; Jack emphasized that revenue was still growing, not falling. A substantial share of semiconductor profit growth and expected growth is in memory, and Micron is the only U.S.-based company among the three leading memory companies he discussed. Jack also saw broad-market earnings optimism as concentrated in Micron, other semiconductor stocks, and oil companies.

Despite weakness in semiconductors, AI data centers, and other AI-infrastructure stocks, the S&P 500 remained within about 5% of its all-time high. Jack saw money moving into healthcare, high-quality financials, insurance, Visa, and Mastercard, but cautioned that those rallies might reflect short covering or underweight positioning rather than durable earnings growth. Max linked AI-related growth and bond issuance for compute infrastructure to higher bond yields, arguing that a slowdown could reduce both growth and bond supply. He read credit weakness alongside a bond-market rally, including during hot inflation and PPI readings, as signs of a growth scare tied to AI’s economic importance. If the AI trade collapses, Jack considers a 20% S&P decline conservative and says 30% or more is possible; Max cited Ben Inker’s estimate that equity issuance could leave the market about 18% below an average outcome, with a real economic slowdown potentially implying a 35% to 50% decline.

The financing risk reflects a mismatch between venture investors, who can accept high risk for outsized returns, and credit investors, who prioritize low defaults and recovery value. The speakers said VC and corporate investment, including Google’s and Microsoft’s investments in OpenAI, cannot fund the infrastructure buildout alone, so fixed income must contribute. A 1% chance of an AI-related catastrophe would be unacceptable in a fixed-income risk calculation, while collapsing compute prices seemed a more plausible threat to financing. A Financial Times report that CoreWeave was seeking “unlimited money” heightened anxiety; its unsecured 2032 bonds also sit behind some private-credit and special-purpose-vehicle financing. The speakers noted that private-credit firms may have different exposures: Ares reportedly had more legacy-software exposure and less data-center debt, while Apollo reportedly had little legacy-software exposure and substantial data-center lending.

The speakers did not see a near-term collapse as the base case, defining a bubble in this context as a collapse within three to six months, while allowing that one could occur in 2029 or 2031. Max viewed “is AI over?” messages from non-finance friends as a possible timing signal, but positioning data showed investors were not universally long: retail investors were among their most underweight in semiconductor stocks in three years, while hedge funds and institutions had been heavily leveraged long and were deleveraging. Max said U.S.-focused fundamental long/short funds were at the zero percentile for net equity exposure about 10 days earlier, with global positioning between the zero and ninth percentile depending on geography; neither measure isolated AI positions. He also noted that IGV fell as much as SOX and SMH during the semiconductor selloff, suggesting the long-semiconductors/short-software trade had been washed out.