The AI trade's doubters are naming names: short CoreWeave, peak earnings in 2027

The short side names its trade

Independent investor George Noble did not hedge his language on the October 22 episode of Bloomberg Surveillance. He called AI "the biggest misallocation of capital in history" and disclosed that CoreWeave is the biggest short position in his fund. Oracle, he added, is another short target, on concerns over fixed-income pricing. Utility stocks, bid up as AI plays, are no longer a safe hedge in his view; gold, he said, is the only viable one. m:…7592

Noble's economics run through the user base. He claims roughly 1% of users account for 80% of revenue in current AI business models, and that subscription pricing is unsustainable: he puts the required monthly fee at $200 to cover compute and energy costs. His warning is that the market is in the early stages of realizing it has been had. m:…7592

The peak-earnings calendar

Edge Wealth chief investment officer Dawson, on Bloomberg Surveillance TV the same day, gave the timing version of the bear case. He argued that tech stocks are likely to peak before the AI adoption cycle peaks, a pattern he described as typical of every technology adoption cycle, and he put the earnings peak in 2027. His valuation logic is cyclical: semiconductor stocks trade on counter-cyclical multiples, meaning the lowest multiple gets applied to the highest earnings. m:…7314

The numbers behind the caution: the semiconductor index is expected to grow earnings by over 100% in 2026, while AI infrastructure EBIT margins expanded from 16% to 42% this year. Dawson flagged Micron's capacity coming online in mid-2027 and 2028 as a test of 85% gross margins, noted that credit spreads for AI assets and data-center bonds have widened, and cited estimates that power constraints could leave only 25% to 50% of planned data centers built. m:…7314

The other side of the trade

Blackstone president and COO Jon Gray, on Bloomberg Talks the same day, gave the bull case its clearest formulation. The AI build-out is not a bubble, he argued, because demand is parabolic while supply is constrained by power, chips and entitlements. In a bubble you build capacity and hope demand arrives; here, in his telling, you build capacity and hope to keep up with demand. m:…7655

Gray's evidence is spending, not sentiment. AI model spending at Blackstone's portfolio companies is up 20-fold year on year, and NeoCloud providers like CoreWeave are seeing pricing power for compute rise 50% year over year. Blackstone itself raised $63 billion in the third quarter and $270 billion over the last twelve months, and Gray positioned the firm as a capital provider, from investment-grade credit to equity, for the data-center and energy build-out. An Anthropic IPO, he added, would be a positive market catalyst by letting the company tap cheaper capital as an investment-grade issuer. m:…7655

What a VC should take from the shouting

Strip out the volume and both camps are pricing the same question: who absorbs the margin compression, and when. The bears point at credit spreads, arguing bond markets are already charging AI assets more for the risk. The bulls point at revealed demand, with portfolio companies spending twenty times more on models than a year ago. For a venture investor, the useful signal is the split itself: when the people financing the build-out and the people shorting it both agree the binding constraint is power, the argument has moved from software to the physical layer. m:…7314 m:…7655

One marker to watch: Gray flagged the Anthropic IPO as a catalyst. Whether it happens, and on what terms, will say more about AI valuations than another quarter of earnings commentary. m:…7655