Advanced packaging, cooling, and critical materials in the physical stack beneath every AI chip.
NVIDIA has reserved approximately 60% of TSMC's total CoWoS output through 2027, leaving rival AI chip makers scrambling for allocation. CoWoS capacity is growing at 80% CAGR and still falling short of demand. Advanced packaging — not silicon fab — is now the binding constraint on AI chip output.
NVIDIA GB300 NVL72 racks draw 120–150kW per rack — beyond the thermal threshold where air cooling is physically viable. Direct-to-chip liquid cooling is now a system requirement, not an option. Vertiv guided liquid cooling revenue growing >60% Y/Y in 2026 as hyperscaler deployments scale.
NVIDIA committed $2B each to Lumentum and Coherent in March 2026 — a $4B lockup of EML laser supply — alongside additional investments in Scintil Photonics and Ayar Labs. The move secured priority capacity and future access rights for NVIDIA through 2027+, leaving rival AI cluster builders unable to source EML supply at any price.
Taiwan's ABF substrate capacity is fully booked for 2026. Spot prices have jumped 30–40% since Q1 2026 as AI GPU demand overwhelms supply. The supply-demand gap is now expected to extend through 2027. Unimicron, Kinsus, and Nan Ya PCB — Taiwan's three major ABF substrate makers — all guided sold-out capacity for the year.