Market Predictions, Rates & Inflation, DOGE, CES, AI Compute | BG2 w/ Bill Gurley & Brad Gerstner

Episode Summary

Executive Summary: The conversation centered on 2025 investing themes: AI as a historic capital cycle, the likelihood of compute constraints persisting, and how CapEx, power, regulation, and rates could shape market returns. The speakers argued that mega-cap tech and NVIDIA remain beneficiaries, but high valuations, higher rates, and fiscal uncertainty create major risks. They also emphasized federal AI policy over fragmented state regulation and predicted broader reasoning-model and agent adoption.

Main Topics: AI as a once-in-a-generation capital cycle (Priority: 5/5): The speakers framed AI as a phase shift comparable to earlier platform waves, arguing that the scale of investment in compute, inference, and data centers could reshape labor markets and corporate earnings. Big Tech CapEx and compute constraints (Priority: 5/5): They discussed hyperscaler spending surging in 2024 and likely rising again in 2025, with demand for training and especially inference pushing firms to keep building despite investor concerns about capital intensity. Market valuation, earnings expectations, and 2025 positioning (Priority: 5/5): The discussion focused on high valuations across the S&P 500 and mega-cap tech, expected earnings acceleration outside tech, and why the investor is carrying less net exposure despite optimism. Interest rates, inflation, and fiscal policy as the main macro risk (Priority: 5/5): The speakers argued that rising long rates, potential stimulus from tax cuts, and uncertainty over whether spending cuts will offset them are the key macro variables that could derail equities. AI regulation and federal preemption (Priority: 4/5): They warned against state-by-state AI regulation, especially a Texas proposal, and argued for a federal framework that preserves U.S. competitiveness while addressing safety. Winners across the AI supply chain (Priority: 4/5): Beyond NVIDIA, the conversation highlighted memory suppliers, power infrastructure, and energy assets as likely beneficiaries of the AI buildout, while noting the market still prices many of them as cyclical. Google, xAI, and the next phase of model competition (Priority: 4/5): They evaluated Google’s broad asset base, xAI’s massive cluster strategy, and the likelihood that reasoning models, coding agents, and consumer assistants will expand AI’s utility in 2025.

Key Arguments: AI could automate nearly all human cognitive labor within three to four years, implying trillions of dollars of economic value could shift toward compute owners and model builders. Hyperscaler CapEx is not just hype; it reflects real compute shortages and expected demand from inference, agents, and reasoning models. The market is assuming strong non-tech earnings acceleration in 2025; if that does not materialize, equity returns may disappoint. High rates are the main macro 'boogeyman'; if the 10-year Treasury stays elevated or rises further, it will pressure valuations and growth stocks. The right AI policy is federal, not state-by-state, because fragmented regulation would slow innovation and hand an advantage to China. NVIDIA remains the clearest beneficiary because much of the incremental AI infrastructure spend still flows into its data center ecosystem. Memory, power, and grid infrastructure are underappreciated AI beneficiaries because inference-time compute and data centers require massive supporting capacity. Reasoning models and agents will broaden AI use cases, but current demand already exceeds available compute, so shortage and pricing power may persist into 2025.

Data Points: AI cognitive-task timeline: 3 to 4 years - Elon’s estimate for when AI can do every cognitive task a human can do S&P 500 valuation peak: 22x earnings - Recent peak valuation cited for the S&P 500 in 2021 S&P 500 valuation trough: 16x earnings - Trough valuation cited in 2022 S&P 500 current valuation: 23x earnings - Current level cited at the start of 2025 Meta valuation: 23x earnings - Example of mega-cap valuation cited Google valuation: ~21x earnings - Example of mega-cap valuation cited NVIDIA valuation peak: 66x earnings - Peak multiple cited historically NVIDIA current valuation: 36x consensus; ~28-30x speaker estimates - Current valuation discussion for NVIDIA MAG 5 earnings growth in 2024: 44% - Microsoft, Meta, NVIDIA, Amazon, and Google earnings growth S&P 500 earnings growth in 2024: 9% - Overall index earnings growth aided by mega-cap tech Other 495 companies earnings growth in 2024: 2% - Non-MAG 5 earnings growth in the S&P 500 MAG 5 expected earnings growth in 2025: 21% - Consensus forecast discussed for 2025 Non-tech expected earnings growth in 2025: 11% - Consensus forecast for the rest of the S&P 500 Healthcare earnings forecast: +4% to +20% - Expected turnaround cited for 2025 Industrials earnings forecast: -4% to +16% - Expected turnaround cited for 2025 Materials earnings forecast: -10% to +17% - Expected turnaround cited for 2025 Big Tech CapEx 2024: ~$160B to ~$260B - Combined CapEx of Google, Meta, Amazon, Microsoft, Apple, and Oracle Meta CapEx intensity: Above 25% of revenue - CapEx as a share of revenue for the largest spenders Microsoft CapEx intensity: Above 25% of revenue - CapEx as a share of revenue for the largest spenders Apple CapEx intensity: Below 5% of revenue - Apple cited as not investing in frontier models OpenAI inference revenue: $10B - Satya’s cited inference revenue figure and expected growth OpenAI revenue run rate: $4B-$5B - Rumored end-of-year run rate cited ChatGPT weekly users: 300M+ per week - Usage level cited as evidence of demand exceeding compute NVIDIA data center revenue 2023: $61B - Consensus data-center revenue cited NVIDIA data center revenue 2025: Nearly $200B - Consensus estimate cited Increment in NVIDIA data center revenue: ~$60B - Increase from 2024 to 2025 estimates 10-year Treasury yield: ~4.7% - Recent level cited as a market risk Federal spending 2019: $4.4T - Historical baseline spending level Federal spending 2024: $6.7T - Estimated actual spending level Baseline federal spending estimate: $5.0T - 2019 spending grown by 2.5% annually Baseline vs actual spending gap: $1.7T - Difference used to argue there is room for deficit reduction Potential annual spending cuts needed: $300B-$400B - Estimated amount needed to offset tax-cut stimulus Trump tax-cut stimulus: ~$400B annually - Estimated annual fiscal stimulus if tax cuts pass Diablo Canyon power contribution: 10% of California’s green, clean power - Used to argue for extending the plant beyond 2029

Pivotal Quotes: "If you believe that to be true, the value of all that human labor that you're replacing is measured in trillions." — Brad: Argument about the economic magnitude of AI automating cognitive work "I really think that it's this inflation and higher rates combined with higher valuations that could put a damper on at least the market in the first part of this year." — Brad: Core macro risk framing for 2025 markets "If we care about the race with China, then the first thing we need to do is to reduce the impediments to us running the fastest race we can run." — Brad: Explanation for opposing state-level AI regulation

Implications: 2025 may be defined by AI infrastructure spending, elevated rates, and policy choices. Winners are likely to be compute, memory, and power suppliers, but markets could lag if earnings or fiscal discipline disappoint.