Alphabet's record single-day market cap surge anchors a week of AI capex signals that cut both ways. Big Tech dropped $133B in Q1 capex (+70% YoY) while Korea DRAM exports exploded to $9.25B. The question isn't demand — it's what happens to margins when $430B/year in depreciation hits the income statement.
Big Tech deployed $133B in Q1 capex, up 70% YoY. Combined depreciation hit $41.6B in the quarter. Server equipment depreciates over 5-6 years — meaning today's record spend locks in an earnings drag through 2030. Consensus pegs annual depreciation for MSFT, AMZN, META, and GOOGL exceeding $430B within five years, against $372B of combined net income last year. Useful lives were already extended from ~4 years to 5-6, so that lever is largely spent.
↳ One observer notes the capex treadmill commodity-token-serving model may not be rewarded at 40% capital intensity. Comparisons to the Model T invention moment, not shale — structural, not cyclical.
Alphabet posted a quarter that far exceeded expectations, driven by cloud and AI services demand. Single-day market cap increase was the largest on record. Total valuation: $4.65T, now within striking distance of Nvidia's $4.85T. The read-through: hyperscaler AI investment is showing tangible returns.
↳ Goldman's Covello recommends long hyperscalers, underweight semis. His thesis: valuation compression in hyperscalers already reflects ROI skepticism, while chip stocks have been rewarded. If hyperscalers demonstrate positive ROI, they re-rate while semis actually face downside from reduced capex. Best-case scenario for the trade: hyperscalers rally on cash flow, semis get hit by demand destruction.
Preliminary April Korea semiconductor export data: DRAM $9.25B (+15% MoM, +83% QoQ). DRAM modules $6.14B (+70% QoQ). SSD $3.84B (+20% MoM, +181% QoQ). MCP $8.16B (+67% QoQ). These are extraordinary sequential ramps.
↳ Commodity memory profitability now exceeds HBM margins, giving the Memory Big 3 stronger incentive to prioritize commodity DRAM/NAND production. That tightens HBM shipment mix in the near term.
Huawei AI chip revenue projected at ~$12B in 2026, up from $7.5B in 2025 — at least 60% growth. 950PR chip entered mass production in March; upgraded 950DT planned for Q4. Beijing is pressuring Chinese tech firms to use domestic chips and restrict Nvidia to domestic-only operations. Nvidia H200 reportedly stuck in customs. Morgan Stanley forecasts China AI chip market reaches $67B by 2030, with 86% supplied domestically. The substitution trend is no longer theoretical.
David Patterson (UC Berkeley, Turing Award) flags High Bandwidth Flash (HBF) as the next compute bottleneck, with semiconductor companies actively collaborating on the problem. Separately, epitaxial wafer supply-demand is tightening faster than expected. Leading-edge logic (≤7nm) wafer demand estimated inflecting next year, reaching ~1M wpm by CY28 — roughly 10% of total 300mm equivalent demand. GlobalWafers, SUMCO, Shin-Etsu, and Siltronic are the beneficiaries.
Former TSMC packaging VP Douglas Yu has joined MediaTek, with insiders expecting his arrival to de-risk EMIB-T adoption. Industry source: "TSMC assumed customers would not leave, but they did." TSMC is reportedly rushing to fill the packaging gap between CoWoS variants. Meanwhile, low TSMC SoIC yields are causing industry to cut near-term CPO shipment forecasts. TSMC's structural moat on advanced packaging is showing cracks.
Bare metal GPU providers are sold out across AMD MI355x inventory. One operator raising prices on 8x boxes, citing zero availability among top AMD providers (Vultr, DO, TW). DigitalOcean pricing H100s at $3.39 and MI300x at $1.99 — roughly 70% discount to market. The gap between cloud pricing and bare metal scarcity suggests margins are compressing for everyone except the GPU owners.
Full analysis, implications, relative trade read-throughs, positioning reads and catalysts — delivered by email at 07:00 HKT daily.