Views on
Companies

The most important company in the world that nobody watches closely enough. Every AI dollar flows through TSMC fabs. Conservative capacity expansion is rational — 99.9% fixed costs, no competitive pressure to overbuild. Chip shortage ~2029 is near-certain. Nvidia has surpassed Apple as TSMC's largest customer.

"Hotel California" lock-in. Three simultaneous architectures — GPUs, Groq LPUs, Vera CPUs — leave no competitive openings. CPU:GPU ratio shifting from 1:8 toward parity as agents drive orchestration demand. Customers cannot easily exit the ecosystem. Dominance is clearest in training and answer inference; agentic inference may favor different architectures over time.

Ad flywheel supercharged by LLMs three ways: better query understanding increases ad relevance; long/complex and non-English queries increase inventory; Gemini data improves targeting across all surfaces. $403B revenue, $132B profit in 2025. Cloud grew 48% with 30% margins.

GEM transformer model drove +24% ad revenue growth. Zuckerberg: "The largest and most successful agent in the world today is Facebook advertising." Meta's ad model is underrated and its execution has been the strongest among Big Tech in the AI era.

Trainium strategy is a commodity compute play — lower cost structure wins in commodity markets. Ad business at $21B/quarter, +22% YoY. More cautious here: CapEx exceeds projected operating cash flow. At risk if AI agents demote Amazon from aggregator to aggregated.

Hardware execution world-class. TSMC supply constraints are the binding limit — Cook cited this three times on one earnings call. John Ternus as CEO signals hardware-defined future. China sales driven by industrial design, not iOS stickiness.

Trillions of data points from billions of transactions. Orders from AI search up 15x since Jan 2026. Physical-world interaction and pre-existing payment rails are competitive advantages. Holds over 14% of US e-commerce.

Disciplined capital allocation — walked away from Paramount/WBD at the right price. Netflix stock jumped on news they lost the deal. Netflix will likely eventually acquire both Paramount and WBD, possibly simultaneously, within the decade.

Already has the right business model (subscription + ads). AI improves recommendations. Building a unique language-to-music dataset. 751M MAUs, 290M premium subscribers. 90M have used Interactive DJ.

AI is sustaining for its dominant enterprise position. "Joe managing the tire shop doesn't care about best of breed." But Azure missed numbers, Xbox likely exits first-party hardware, and growth may plateau well below prior expectations.

Growth threatened even if survival is not. Seat-based pricing faces headwinds. Shift from "growth company" to "stable company" valuation is a massive haircut. Potential PE opportunity: buy at low multiples, fix cost structure, run as cash generators.

800M users who are weekly-active, not daily-active, with "very, very shallow usage." Benedict Evans' framing: a "cargo cult" — copying Meta's playbook without understanding why it worked. Initial ad product (banner ads in conversation) is wrong. Needs product-market fit in 2026 or faces commoditization.

Shut down CPU lines at the wrong time — just as agentic workloads are driving CPU demand back up alongside GPU demand. This is a major strategic error with lasting consequences.

Charging for multiplayer and raising prices are stunting their own network effects. Roblox is eating their lunch by being free and open. Memory crisis is delaying PS6 (possibly to 2028–2029) and raising hardware costs across the board.