Big Tech's $133B Q1 capex locks in a depreciation drag that could exceed combined net income within five years. Huawei's AI chip revenue surges 60% to $12B as Nvidia's H200 gets blocked at Chinese customs. Epitaxial wafer ASPs are turning as leading-edge demand inflects toward 1M wpm by CY28.
Big Tech deployed $133B in Q1 capex, up 70% y/y. Combined depreciation hit $41.6B in the quarter. Server equipment depreciates over 5-6 years—today's record spend locks in 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. The useful-life extension lever from ~4 to 5-6 years is already spent.
Goldman's semiconductor team now recommends long hyperscalers, short semis—arguing compressed valuation multiples already discount capex skepticism, while chip stocks have priced in the upside.
↳ The capex-to-depreciation treadmill is the defining tension for mega-cap tech through decade-end. If hyperscalers demonstrate ROI, semis face revenue compression as cloud spend normalizes. If they don't, both legs sell off together.
Huawei is set to capture the largest share of China's AI chip market this year, with revenue jumping at least 60% to ~$12B from $7.5B in 2025. The 950PR chip entered mass production in March; an upgraded 950DT is planned for Q4. Beijing has directed Chinese tech firms to support domestic suppliers and restrict Nvidia chips to overseas operations only. U.S. regulators are simultaneously blocking Nvidia H200 customs clearance in China.
Morgan Stanley forecasts China's AI chip market at $67B by 2030, with 86% served by domestic suppliers—roughly $21B from domestic sources this year alone. Separately, SMIC has reportedly produced early 3nm test wafers as of February 2026.
↳ The bifurcation of global AI compute supply chains is accelerating. Nvidia's China revenue floor is being actively reshaped by regulatory action from both Washington and Beijing.
Epitaxial wafer supply-demand balances are tightening faster than expected. Leading-edge logic (≤7nm) wafer demand is modeled to inflect next year, reaching ~1M wpm by CY28—roughly 10% of total 300mm equivalent demand. GlobalWafers, SUMCO, Shin-Etsu, and Siltronic are primary beneficiaries. Wafer ASPs are turning.
AXT reports indium phosphide substrate demand is accelerating faster than qualified supply, with customers shifting to multi-year allocation agreements. Geographic demand is broadening from U.S. hyperscaler supply chains to include a rapidly scaling China AI optical ecosystem.
TSMC SoIC yields are reportedly low, leading the industry to cut near-term 2026 CPO shipment forecasts. Advantest appears to be gaining share in CPO testing relative to Teradyne, based on Teradyne's own Q1 call commentary.
↳ Former TSMC packaging VP Douglas Yu has joined MediaTek, expected to de-risk MediaTek's first EMIB-T adoption. Industry sources note: "TSMC assumed customers would not leave, but they did." TSMC is now rushing to fill the packaging gap between CoWoS and SoIC variants.
Major bare-metal GPU providers—including Vultr, DigitalOcean, and others—are fully sold out of AMD MI355x inventory. One provider is raising prices across 8x bare metal configurations, citing zero market availability. DigitalOcean is pricing H100s at $3.39/hr and MI300x at $1.99/hr—a ~70% discount to prevailing rates, suggesting either loss-leading or legacy inventory clearing.
↳ The sold-out signal on MI355x is a positive read-through for AMD data center revenue timing. Vultr reportedly moved massive MI355x volume ahead of the squeeze.
Turing Award laureate David Patterson identifies High Bandwidth Flash (HBF) as the likely next compute bottleneck after HBM, citing surging demand and active collaboration with semiconductor companies.
Commodity memory profitability has risen above HBM margins, giving the Memory Big 3 stronger incentive to prioritize commodity DRAM/NAND over HBM shipments. Korea April DRAM exports hit $9.25B (+15% MoM, +83% QoQ); SSD exports reached $3.84B (+20% MoM, +181% QoQ).
↳ If HBM shipment mix tilts lower while commodity margins stay elevated, the HBM revenue thesis for SK Hynix and Samsung faces near-term dilution even as AI capex accelerates.
Applied Digital signed a 15-year lease at its Delta Forge 1 campus with a second U.S. investment-grade hyperscaler. The deal covers 300MW of critical IT load, adds ~$7.5B in contracted value, and brings total contracted revenue to $23B+. Over 50% of revenue now backed by IG tenants. Initial operations expected mid-2027.
Alphabet surged to $4.65T market cap after a blowout quarter—cloud and AI demand framed as proof AI investment is generating returns. Within striking distance of Nvidia ($4.85T). Apple's Q2 FY2026 call confirms memory is not the bottleneck; SoC and leading-edge node capacity at TSMC is the actual constraint limiting unit shipments. Mac mini and Mac Studio are the most supply-constrained SKUs.
Full analysis, implications, trade read-throughs, positioning reads and catalysts — delivered by email at 07:00 HKT daily.
Full bull/bear breakdown, broker targets, and catalyst calendar — delivered by email at 07:00 HKT daily.