The AI Trade, the Fed and the Next Phase of the Bull Market | Warren Pies

Episode Summary

Executive Summary: The episode argues that AI is still driving a durable bull market, with the main risks being regulation, Fed tightening, and sentiment rather than open-source competition or token-cost collapse. The guests see hyperscaler capex crowding out housing, supporting inflation and semis, while labor remains mixed and not yet recessionary. They stay constructive on equities, semis, and AI-related earnings.

Main Topics: AI bull market: what bearish arguments matter (Priority: 5/5): The discussion separates credible AI risks from noise. Open-source panic and token-maxing concerns are viewed as overblown, while regulatory pressure, poor lab messaging, and future financing/sentiment risks are taken seriously. Crowding out of housing by tech capex (Priority: 5/5): A central macro thesis is that data-center and IT spending are displacing residential fixed investment and labor, creating a K-shaped economy where tech investment rises as housing weakens. Fed dilemma and inflation from tech spending (Priority: 5/5): The guests argue that hyperscaler capex is feeding into core PCE and forcing the Fed to weigh price stability against employment, with rate hikes unlikely to stop AI spend unless policy becomes much tighter. GPU demand and real-time compute indicators (Priority: 4/5): They explain why GPU availability across neo-clouds is a cleaner supply-demand signal than announced capex or revenue, and say their data still shows intense compute scarcity despite recent worries. Token demand, open source, and frontier model competition (Priority: 4/5): Using OpenRouter and internal pricing data, they contend that frontier token demand is still growing even as open-source usage rises, so open-source gains are complementing rather than replacing frontier demand. Semiconductors as the meta-group under AI (Priority: 4/5): Semis are framed as the key beneficiary of the AI buildout. The recent drawdown is seen as a buyable dip rather than a cycle top, though valuations now reflect a less cyclical business than history would suggest. Earnings, valuations, and the equity outlook (Priority: 4/5): Strong AI-led earnings growth and fiscal support keep the market constructive, but multiple expansion should be limited if the Fed remains on hold or turns hawkish. The team still maintains a positive S&P 500 outlook.

Key Arguments: Open-source AI is not yet an existential threat to frontier-model demand; if it were, frontier token prices and usage would be falling materially, but the data show continued growth. The biggest AI-related risk is regulation and messaging, not model collapse; public fear about data centers and job loss could shape policy by 2028. AI and hyperscaler capex are crowding out residential construction and labor-intensive sectors, creating a bifurcated economy and supporting inflation. The Fed faces a difficult dual-mandate tradeoff because inflation is being driven more by capex and fiscal deficits than by a tight labor market. GPU availability across neo-clouds provides a more timely read on compute supply-demand than capex announcements or revenue figures. Token demand is expanding: open-source tokens are growing fast, but frontier tokens are still rising too, implying total demand is enlarging rather than shrinking. Semis remain in a powerful multi-year cycle; the current drawdown resembles prior cyclical pullbacks inside a larger secular AI uptrend. Equities can still work if earnings remain strong, but returns depend on the Fed not shifting into a true tightening cycle. The market is not in a broad bubble, but some pockets of exuberance exist, especially in semis where valuation behavior no longer matches historical cyclicality.

Data Points: Tech vs. housing fixed investment: ~$1.5 trillion vs. ~$1.1 trillion - IT/information-processing and equipment spending now exceeds residential fixed investment, illustrating crowding out. Core PCE contribution: ~7 bps month-over-month - Tech spending categories are estimated to be adding to core PCE inflation. Residential construction payroll drawdown: 2% decline over two months - Seen as a yellow-light warning for the labor market; historical recession signal is closer to an 8% decline. Nonfarm payrolls ex healthcare, education, government: Flat over the last year - Used to argue the labor market is not broadly strong despite headline payroll gains. Workers receiving no raise: Almost 14% - From wage tracker data; suggests wage growth is stagnating rather than overheating. Hyperscaler capex share of GDP: ~2.5% of GDP - Shows how large AI infrastructure spending has become in the macro economy. AI-related earnings growth: ~25% this year - Earnings strength is front-loaded by AI and semis, supporting the equity bull case. Five-year earnings growth estimate: ~15-16% including this year; ~14-15% excluding this year - Contrasts with overstated long-term estimates that imply unrealistic 25% annual growth. S&P 500 target: 7,850 - The guests say they are holding their year-end target despite higher rates and stronger earnings. Semiconductor cycle length in the late 1990s: 63 months - Historical comparison used to argue the current AI/semis cycle could have further room to run. Semiconductor bull-market multiple: ~11x in the 1990s vs. ~4x now - Used to suggest the current cycle may be shorter on time so far but still has room on valuation and duration. Oil market short positioning: ~40% managed money short - Indicates extreme pessimism and the possibility of a short-covering rally. Oil clearing-price range: $75-$80 Brent - The long-term view is that oil is unlikely to settle materially above this range absent renewed geopolitical disruption. OpenRouter inference growth: ~40% month-over-month - Shows overall token demand is still rising rapidly. Open-source token growth on OpenRouter: ~60-65% month-over-month (implied) - Open source is gaining share quickly, but total usage is still expanding. Frontier token growth on OpenRouter: Double-digit month-over-month - Evidence that frontier demand has not rolled over despite open-source growth. GPU availability: Collapsed to 0% for Blackwell and at one point for all GPUs - A real-time signal of intense compute scarcity and strong AI demand. S&P 500 yearly trailing 12-month earnings growth: ~23% this year - Part of the argument that earnings are stronger than many bearish models assume.

Pivotal Quotes: "Bull markets don't die of old age, they get murdered by the central bank." — Warren Pies: Used to frame the main macro risk as Fed tightening rather than valuation alone. "If our fate, if the AI fate is left in the hands of Dario and Sam, we're all screwed, in my opinion." — Warren Pies: A blunt comment on poor AI messaging from lab leaders and its regulatory implications. "The most recent thing that's been cropping up is two things... the open source panic." — Warren Pies: Introduces the key bearish AI narrative the guest argues is overstated.

Implications: Listeners should view AI infrastructure, semis, and select equities as still in a structural uptrend, but with rising regulatory and Fed-policy risk. Housing and other labor-heavy sectors may remain under pressure as capital shifts toward tech.