Bernstein's senior US semiconductor & semicap analyst (NVDA, AMD, INTC, QCOM, AVGO, AMAT, LRCX, KLAC) — running synthesis of his video/podcast appearances, with per-transcript breakdowns and a stock index.
Upgraded (late, admittedly): genuinely good server CPUs taking huge share from Intel (up ~70%+ this year), OpenAI/Meta multi-gigawatt GPU deals — ~$20 EPS possible by 2028, two years ahead of target.
His monetization proof-point: run-rate "gone vertical" (~$1B → $44B annualized in a year); agentic coding is the first AI use case at "takeoff velocity."
Monopoly on lithography — the critical patterning step (~90% share, ~100% in advanced/EUV); also the choke point that keeps China's chips uncompetitive.
AI (custom chips incl. Google's TPUs + AI networking) guided to $100B next year — "probably a lot better"; lagged like Nvidia, and its VMware software is infrastructure AI runs on, a baby thrown out with the SaaS bathwater.
Accelerating (85% growth, ~75% GM, $91B guide) yet lagging the bottleneck names whose businesses hinge on it — the divergence "has to normalize" with Nvidia coming up; plus a ~$20B CPU story off agentic AI.
Bought half Intel's Ireland fab for $11B when Intel was desperate ('PE doesn't work for free') — and just got paid a handsome return when Intel bought its way out.
"Lucky over good": uncompetitive product sells anyway in the CPU shortage, balance sheet fixed, 18A/14A statements carefully worded (yields still poor) — and unshortable because "Trump wants the stock to go up."
"Poor little Meta" — $135B capex with no rentable cloud, but real internal AI returns and an AMD GPU customer; a Meta capex cut is the derailment signal that arrives too late.
Memory poster child — pricing gone crazy, order-of-magnitude earnings revisions, yet single-digit PE; the bottleneck trade that must eventually normalize against Nvidia (colleague's coverage).
Eisman's "shell game / weak sister" bear theory; Rasgon doesn't take the bait — points to monetization evidence instead. Also one of AMD's two big GPU-deal customers.
~75% smartphones + the licensing annuity, chasing data center: Humain CPU/200MW rack deals (nothing seen yet) and a detail-free hyperscale ASIC win that still moved the stock 70% — "tells you how nuts things are."
China's chips "are not competitive" — semicap sanctions (no ASML) are working; Huawei pursues known-but-early techniques "because they have no choice."
In one line: the AI semi boom is real and accelerating — the rally is all earnings (SOX multiples actually compressed), the depreciation bear case fails against rising old-GPU rental prices, and monetization is visible (Anthropic ~$1B→$44B run-rate, agentic coding at "takeoff velocity") — so the trades are the lagging compute names (NVDA, AVGO) normalizing up toward the bottleneck names, and the agentic-AI CPU wave (AMD upgraded); the real long-run risks are AI returns disappointing and, physically, power.
AI drags everything; investors play bottlenecks. One sub-sector at a time becomes the constraint (memory → semicap → optical → power → CPUs) and rips; "you could have owned anything in the space." The earnings, not multiples, did the work — a bear must explain why earnings are unsustainable.
The normalization trade. Nvidia (85% growth, accelerating, ~$20B CPU story) and Broadcom (AI guided to $100B next year) lag the constraint names whose businesses hinge on them — "one of them has to be wrong… either the constraints come down or Nvidia comes up."
Agentic AI is a CPU demand wave. Agents do real-world tasks on CPU cores (attach 1:8 → 1:2 → toward 1:1), lifting NVDA/AMD/INTC alike; AMD upgraded — server share gains off Intel plus OpenAI/Meta GPU deals point to ~$20 EPS by 2028.
Bear cases triaged. Burry's depreciation argument fails (old-GPU rental prices rising; fully-depreciated GPUs very profitable). The real doomsday — no/poor return on AI — can't be front-run via capex (cuts show up too late; meanwhile capex only goes up and is existential for hyperscalers); monitor proxies like token usage and Anthropic's run-rate instead.
Power is the binding constraint. "The US has chips but no power; China has power but no chips." US grid can't add enough → on-site generation (3-yr turbine lead times, SMRs, Three Mile Island restart); China's chips stay uncompetitive while semicap sanctions (no ASML) hold.
Transcripts
One dated page per appearance — each has its full stock table, talking points, and the saved transcript. Newest first.
Stacy Rasgon appearances discovered via YouTube search (Stacy Rasgon), not yet processed — verify publish dates & channels before processing. (None queued yet — his prior Eisman Playbook appearance, Ep 28 "Inside the AI Chip War", ~Nov 2025, is a candidate.)