in this week's video, I explain why the AI infrastructure build-out has shifted from a compute bottleneck to an unbounded compute requirement as we enter the agentic era. Markets remain resilient, S&P flat for the week, rates volatility at pre-2022 levels, junk spreads at all-time tights while labor indicators confirm what major employers already know: AI is redesigning every job. The critical insight: 2025 marks the transition from cloud-based model training to AI agents, digital employees, and edge device upgrades, creating the revenue inflection that skeptics have been waiting for. The thesis centers on why compute demand will never be satisfied. With multimodal architectures (VLMs combining vision, language, audio, video) and recursive self-improvement now underway, frontier model leaders confirm compute requirements are unbounded. Sam Altman, Demis Hassabis, and Mustafa Suleyman all emphasize the same point: OpenAI has never had excess compute capacity, scaling is the mandatory path to AGI, and Microsoft alone plans hundreds of billions in spending over the next decade. As enterprises shift from scattered AI pilots to unified AI operating systems, infrastructure spending intensifies across cloud compute, vector databases, API gateways, and on-premise edge deployments. I also go through an interview with Jim Chanos for his views. For traditional investors, this creates a forced rotation into previously avoided sectors. The AI-power-capex cycle favors small caps as hardware diffuses through supply chains, while physical infrastructure constraints shift alpha toward hardware with the synchronized upgrade cycle, enterprise needs for AI agents and the rise of robotaxis. Micron exemplifies the opportunity, sold out through 2026 for HBM despite analyst calls for oversupply while Cisco benefits from the enterprise networking build-out as companies bring AI in-house for security. Tesla maintains its position as the top MAG7 theme for 2026, with 6.8 billion autonomous miles versus Waymo's 100 million and a vision-only approach that scales without LIDAR dependency. Next year, revenues arrive through AI agents, agentic commerce (Shopify/Etsy checkout in ChatGPT), digital employees replacing SaaS subscriptions, and tokenization/stablecoin infrastructure. Bitcoin ties the AI-energy nexus together as miners act as virtual batteries, stabilizing grids while benefiting from the same physical constraints that restrain Big Tech revenue timing. Timestamps • (0:00–5:40) Market setup and AI infrastructure thesis: Weekly recap shows S&P flat, rates volatility back to pre-2022 levels, junk spreads at all-time tights. J • (12:18–18:25) Small-cap rotation risk and portfolio construction: Mike Wilson positive on small caps, but biggest risk is rotation from most crowded trade (Nvidia $5T vs Russell 2000 $3T). Used AI (Grok) to build portfolio of 20 S&P stocks turning year-over-year positive, identified 10 macro regimes, all reflationary/expansion themes including value rotation, industrial metals, weaker dollar. • (19:18–24:16) Jim Chanos interview and dot-com comparison: Oracle concerns (Michigan data center in limbo), but Chanos rational on bear case, free cash flow negative, revenues not yet in. Positive on hyperscalers and Nvidia. • (24:16–29:15) AI agents are the revenue inflection: Sam Altman interview confirms enterprise inflection has arrived, companies request single AI platform, GPT 5.2 achieves 70-74% of knowledge work tasks at expert level. Next year benchmarking expected to show 95% of knowledge worker jobs can be replaced. New AI-first devices coming, every car getting AI. • (29:15–33:45) Enterprise AI platform spending and PMI implications: Morgan Stanley example shifting to single AI operating system requires heavy investment in cloud compute, vector databases, API gateways, new network hardware. GPU compute highest spend intensity historically, next phase is cloud infrastructure/networking/security. Expect PMIs well into the 60s as we go into hardware build-out with AI inside every company. • (33:45–38:18) Bill Gurley and Demis Hassabis on compute: Asking if it's a bubble is wrong question—every technological revolution has real innovation and speculative excess simultaneously. Demis emphasizes multimodality (VLMs) as biggest frontier, video understanding next major leap, multimodal architectures expand compute needs exponentially. • (40:15–44:12) Mustafa Suleyman and labor deflation: Compute requirements unbounded for recursive self-improvement, Microsoft plans hundreds of billions over next decade. • (44:12–47:34) Tesla as top MAG7 theme for 2026: 6.8 billion autonomous miles versus Waymo's 100 million, fleet of 5.1 million cars versus 2,500. Vision-only approach wins on cost per mile. • (48:13–50:30) Bitcoin and tokenization year ahead: Brutal price action since October, won't turn positive until close above $92K three days in a row.. Next year is tokenization and stablecoin year.