The Thesis — Gavin's Argument
"AI makes global GDP inexorably more silicon-intensive. Old SaaS: 85% gross margin, 5% to compute. AI companies: 40% gross margin, 50% to compute. Every dollar of AI revenue is structurally more silicon-intensive than any prior technology." — Gavin Baker
🔩 Picks & Shovels
Nvidia, TSMC, and Broadcom are the irreplaceable infrastructure layer. Every dollar of AI revenue translates directly into silicon demand. NVDA could sell $2–3 trillion of GPUs in 2026–27 if Jensen had his way. TSMC is the single most important company in the world — their capacity discipline is the difference between a generational build-out and a catastrophic bubble.
"The most extraordinary moment in the history of capitalism." — Gavin Baker
💾 DRAM
Gavin's highest conviction long. Micron trades at mid-single-digit forward P/E while semicap equipment sits at 40x. Both can't be right. HBM (High Bandwidth Memory) demand is growing exponentially — every AI training cluster needs it. The cross-sectional valuation incoherence is the core arbitrage of this entire thesis.
"I don't know anyone not bullish on DRAM. No one. That worries me." — Gavin Baker
⚙️ Semicap Equipment
ASML, KLAC, LRCX, and AMAT win regardless of who wins the AI model war. TSMC, Intel, and Samsung are all expanding capacity — all of them need semicap equipment. The catch: at 40x earnings, semicap may already be priced for perfection. Own it, but watch the valuation carefully.
"Semicap at 40x vs DRAM at single digits. Those can't both be true." — Gavin Baker
⚡ Power / Watts
The timing trade. A near-term power shortage is real — zoning and approvals are now the binding constraint, not engineering. But capitalism will solve this by 2027–28. GE Vernova at 40x "builds in an unfathomable amount of Nvidia share loss" into its valuation — a red flag for quality-chasing at the wrong moment.
"Capitalism will solve this — 2027-28." — Gavin Baker
Gavin's Four Watches
The Valuation Incoherence
"Semicap equipment companies trading at 40x next quarter's annualised earnings and DRAM companies trading at mid-single digits. At the peak of the last cycle it was 5x vs 12x. Those can't both be true." — Gavin Baker, May 2026
Forward P/E by Group — the signal Gavin watches most closely
If the AI build-out continues, DRAM must re-rate. The gap between DRAM (green) and semicap (blue) is the core arbitrage in this thesis. Watch it close.
Picks & Shovels — Core Holdings
DRAM — Highest Conviction Long
Semicap Equipment — Essential Regardless of Who Wins
Power / Watts — 2027-28 Timing Inflection
Catalyst Calendar — Events That Matter
WATCH · ~2 months
Fed FOMC
Jul 29–30 2026
Rate path affects tech multiples. Cuts = multiple expansion for long-duration growth assets.
KEY · ~3 months
NVDA Q2 FY27 Earnings
~Aug 2026
Data center revenue and HBM demand guide. Most important earnings in the thesis.
HIGH · ~4 months
SpaceX 90-Day Lockup Expiry
~Sep 2026
First window for ~1,700 employees and early investors to sell. Watch for liquidity drain from tech. Note: staggered lockup structure under discussion — supply may start earlier and spread across months, not a single cliff.
WATCH · monthly
TSMC Monthly Sales
10th of each month
Monthly revenue report. Rising = AI capex intact. Flat/down = thesis pressure.
KEY · ~6 months
NVDA Q3 FY27 Earnings
~Nov 2026
Earnings near SpaceX lockup expiry. Double event risk window — elevated volatility likely.
HIGH · ~7 months
SpaceX 180-Day Lockup Expiry
Dec 15–27 2026
Full unlock. Could slip to Jan 2027 if structured that way — employees save a full year's marginal tax on the sale, so there is incentive to push for January. Dec 28 vs Jan 2 is a material date to watch. Largest scheduled supply shock in US equity market history.
Adjacent Macro Risks — Not the Thesis, But Will Move the Market
These don't change whether Gavin is right. They change whether the price will reflect it when he's right.
HIGH
SpaceX IPO Lockup Unlock
At $2T, the largest scheduled supply shock in US equity history. Roadshow targeting week of June 8, 2026 — S-1 not yet filed, terms not confirmed. Lockup likely 90–180 days, potentially staggered (some selling day one, then tranches — worse than a single cliff). 180-day expiry may be structured to land Jan 2027 to save employees a full year's tax — watch for this. Bull case: capital rotates into tech. Bear case: employees permanently diversify out of growth equities. Coincides with Nov NVDA earnings — double risk window.
MEDIUM
AI Capex Cycle Risk
Hyperscalers guiding $60–80B+ for 2026. If any single one guides down — the AI trade de-rates violently. Watch Microsoft, Google, Meta, and Amazon earnings commentary closely. Demand destruction would hit NVDA, TSMC, and semicap simultaneously.
HIGH
China / Taiwan / Export Controls
TSMC is in Taiwan. ASML is Dutch. KLAC, LRCX, and AMAT are US. Any Taiwan escalation or export control tightening directly impairs the thesis supply chain. China revenue is 15–25% for some semicap names — a material exposure.
LOW–MED
Fed Path / Multiple Compression
These are high-multiple, long-duration assets. Higher-for-longer rates mechanically compress multiples. The thesis can be right on fundamentals and still lose money. Market is currently pricing 2 cuts in 2026.
12-Month Relative Performance vs QQQ
"Tech essentially got as cheap as it's been versus the rest of the market at any point over the last 10 years. We had the most extraordinary moment in the history of capitalism — wildly bullish for AI — and you got a chance to buy AI at really attractive valuations." — Gavin Baker, April 2026
Thesis Stocks vs QQQ Baseline · 12 months · normalised to 0%
Cumulative outperformance vs Nasdaq 100. The lens Gavin uses to judge when his thesis is cheap or expensive relative to the market.
Natural Gas NG1 · 12 months
The watts proxy. NG down = US manufacturing cost advantage over Asia/Europe. "Natural gas was down 20%. Our relative manufacturing competitiveness improved overnight." — Gavin
Thesis Layer Status
"AI makes global GDP inexorably more silicon-intensive. Old SaaS: 85% gross margin, 5% to compute. AI companies: 40% gross margin, 50% to compute. Every dollar of AI revenue is structurally more silicon-intensive than any prior technology." — Gavin Baker
| Layer | Thesis | Key Positions | Avg Fwd P/E | Avg 1Y Return | Gavin's Signal |
|---|
Thesis Health — Is It Playing Out?
DRAM re-rating in progress
MU at ~7x fwd P/E. Still deep value vs semicap. Intact.
DRAM/Semicap gap still extreme
26x gap. Consensus widely aware. Watch for narrowing without re-rating.
AI capex accelerating
All 4 hyperscalers guiding $60B+ for 2026. Demand side confirmed.
HBM demand tightening
SK Hynix and Samsung HBM sold out through H2 2026.
Power timing uncertain
Watts shortage real but 2027-28 easing timeline unclear. Power names lagging.
TSMC capacity disciplined
Not overbuilt. Capex disciplined. Bubble prevention intact.
NVDA consensus crowded
"I don't know anyone not bullish on Nvidia." Low short interest.
SpaceX lockup risk building
$2T IPO, roadshow June 2026. Lockup 90–180 days, possibly staggered. Expiry Dec 2026 or Jan 2027. Potential liquidity drain from growth equities.
Latest Signals — @GavinSBaker & Podcast Alerts
Loading signals…
What Would Break This Thesis
1
TSMC guides down capex materially
"If TSMC starts pulling back on capacity expansion, the whole thesis is wrong." TSMC's discipline is what separates a generational build-out from a bubble.
2
Hyperscaler AI capex goes negative YoY
"If Microsoft/Google/Meta/Amazon reduce AI infrastructure spend in any quarter, the demand side breaks." This is the single most watched data point.
3
DRAM price collapse (>30% in a quarter)
"Either supply outran AI demand, or AI demand disappointed." A DRAM price collapse invalidates the highest-conviction long in the thesis.
4
Nvidia loses >20% accelerator market share
"If AMD or custom silicon takes real share, the NVDA position needs rethinking." Custom ASIC progress from hyperscalers is the key watch signal.
5
China/Taiwan escalation impairs TSMC
"TSMC IS the thesis. Any scenario that impairs their 3nm/2nm operations invalidates the supply chain." Geopolitical risk is unhedgeable.
6
AI revenue model fails
"If AI products fail to monetize at scale, demand for AI silicon collapses." The thesis depends on AI companies spending 50% of gross margin on compute — that spend must be real and durable.
Full Metrics Table
| Stock | Price | Exchange Time | 1D % | YTD % | 1Y % | Fwd P/E | Trail P/E | P/S | Gross Margin | Rev Growth | Mkt Cap |
|---|