Big Tech Q1 capex hit $133B (+70% y/y) while annual depreciation for the mega-cap cohort is on track to exceed their combined net income within five years. Huawei AI chip revenue surges past $12B as China accelerates decoupling. Korea April DRAM exports signal memory cycle inflection.
Big Tech Q1 capex: $133B, up 70% y/y. Combined depreciation hit $41.6B in the quarter alone. Server equipment depreciates over 5-6 years, meaning 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. Useful lives already extended from ~4 to 5-6 years. That lever is spent.
↳ Alphabet's Q1 blowout ($4.65T market cap, closing on NVIDIA's $4.85T) validated the spend for now. Goldman's Covello now recommends long hyperscalers / short semis, arguing compressed valuation multiples give hyperscalers more room to re-rate while chip names are priced for perfection.
Huawei AI chip revenue projected at ~$12B in 2026, up from $7.5B in 2025. 950PR in mass production since March; 950DT upgrade planned Q4. Morgan Stanley sizes the China AI chip TAM at $67B by 2030, with 86% from domestic suppliers. This year's domestic supply alone: ~$21B. NVIDIA H200 chips reportedly unable to clear Chinese customs — regulatory friction from both sides.
↳ Beijing directing tech firms to use domestic chips and restrict NVIDIA to domestic-only operations. U.S. regulators demanding the same from the other direction. The double-squeeze is real and accelerates Huawei's share capture.
Korea April DRAM exports: $9.25B (+15% MoM, +83% QoQ). SSD exports: $3.84B (+20% MoM, +181% QoQ). Epitaxial wafer supply-demand tightening faster than expected — leading-edge logic wafer demand inflecting toward ~1M wpm by CY28, roughly 10% of total 300mm equivalent. Major wafer makers (GlobalWafers, SUMCO, Shin-Etsu, Siltronics) in line to benefit.
↳ Critical nuance: commodity memory profitability now exceeds HBM, giving the Memory Big 3 stronger incentive to prioritize standard DRAM/NAND production in the near term. Apple confirms memory is a cost headwind, not a supply bottleneck — TSMC leading-edge SoC capacity is the actual constraint limiting shipments.
Monolithic Power Systems Q1 read-through: AI demand broadening into CPU servers, optical modules, switches, storage, DDR5 infrastructure, and higher-voltage rack-level architectures. Bare metal GPU availability is zero — top AMD MI355x providers (Vultr, DO, TW) all sold out. H100s pricing at $3.39, MI300x at $1.99.
↳ Applied Digital signed a 15-year lease for 300MW critical IT load with a second IG hyperscaler, adding ~$7.5B in contracted value. Total contracted revenue now $23B+, over 50% backed by investment-grade tenants. Initial operations mid-2027.
Goldman puts 90% upside on MediaTek (TP5000) with next-gen AI ASIC ASPs multiple times higher than current generation, plus GM-accretion from compute die design involvement. Former TSMC packaging VP Douglas Yu joins MediaTek — expected to de-risk EMIB-T adoption. Industry source: "TSMC assumed customers would not leave, but they did."
↳ Qualcomm also seeing a bump on hyperscaler ASIC read-through. Cohu pivoting from automotive/industrial test toward HBM thermal management — Eclipse handler targeting extreme power density AI accelerator applications.
Low TSMC SoIC yields driving down 2026 CPO shipment forecasts. AXT confirms indium phosphide demand accelerating faster than qualified supply — procurement shifting to multi-year allocation security. China AI optical ecosystem scaling rapidly. Aixtron sizes optoelectronics at 80-100 G10 tools per year (~€300-400M annual TAM).
↳ UC Berkeley's David Patterson flags High Bandwidth Flash (HBF) as the next bottleneck after HBM — "highly likely to stand at the center of the next surge in demand." Active semiconductor collaboration underway.
Full analysis, implications, relative trade read-throughs, positioning reads and catalysts — delivered by email at 07:00 HKT daily.