Is The AI Selloff A Buying Opportunity? | Warren Pies

Warren Pies has been overweight equities since late 2023, but has downgraded his stance and neutralized risk after a roughly 63-day consolidation of 6%–7% near market highs weakened momentum. He sees a potential breakout or breakdown, not a clear buying signal: he would consider buying a breakout or a period of extreme fear, but says neither condition is present. The current setup calls for caution rather than assuming the pullback is already an opportunity.

Market leadership and credit are adding to his concern. Financials have rolled over relative to consumer staples, a pattern Pies says has preceded past consolidation breakdowns, while staples gained 14% in October as investors rotated away from AI-related areas. Hyperscaler credit spreads also moved above their July wides before equities did, a warning because the AI buildout requires substantial financing. Estimated all-target positioning remains about one standard deviation above its rolling mean, suggesting investors could still cut exposure; October weakness also included mutual-fund year-end tax-loss selling. Pies cautions that technical analysis is imperfect and that a market led by energy and staples is unlikely to offer a strong tailwind for equities.

Pies remains bullish on AI’s long-term potential and does not think the trade has already become a burst bubble. He tracks GPU availability and rental rates: demand for agentic AI tightened availability, B200s were unavailable, and rental prices kept rising, contrary to his data-center model’s assumption that GPU rents would fall 25% annually. Higher forward rental rates can indicate a capacity premium rather than weakening demand. He would reconsider his fundamental view if availability rose sharply, rents plunged, or evidence of overbuilding emerged. Semiconductors had grown to about 20% of S&P 500 market capitalization, and their equity-line spread against the S&P 500 across roughly 50 industry groups exceeded 105%; after worrying about a top in June and hoping July marked a bottom, he says semiconductor charts have broken down again.

Other analysis discussed points to improving AI economics, though not without qualification. Anthropic’s revenue per dollar of compute spend rose from $0.57 in Q2 a year earlier to $2.08, then dipped somewhat with its Q3 release; revenue per dollar of inference compute spend rose each quarter, including Q3. Inference compute spend was declining as a share of Anthropic’s total compute spend, while hyperscaler cloud revenues were beginning to ramp up. These indicators support the case for returns emerging, but do not eliminate the financing and execution risks of scaling infrastructure.

Those risks include interest rates, power, materials, labor, supply bottlenecks, and political decisions. Pies says AI companies’ reliance on external funding makes the buildout more exposed to rate hikes than when it was funded from excess cash flow; VC-backed startups account for an estimated 25% of lab ARR, which could be vulnerable if higher rates sharply curtail venture funding. A separate construction analysis found no current data-center supply shortfall or signs of overbuilding, and its satellite tracking aligned so far with semiconductor earnings and revenue estimates; its models showed more than 27 gigawatts of capacity potentially coming online over 2027. It interpreted rising GPU rental prices alongside falling hyperscaler and semiconductor shares as broad liquidation, or “degrossing,” rather than evidence of a supply crunch.

Pies identifies macro conditions, especially rates, as the intermediate-term threat: inflation that fails to cool could constrain funding and slow AI investment, though he sees this as a difficult path rather than necessarily a death blow to the trade. He expects financial instability to limit further Fed hikes and a credit-market wobble to end the hiking cycle, potentially followed by rate cuts. With the two-year yield at 4.6%, he sees room for one more hike and a possible hawkish bias; a signal that hikes are definitively over could lower the two-year yield and support rate-sensitive areas. He expects an AI-related wobble to bring choppiness rather than recession, and plans to retain fundamental conviction while using technical signals to manage risk. For balanced portfolios, he favors overweighting energy as a hedge against the conflict worsening, while warning that oil markets remain disrupted and that refining stocks are cyclical.