Sigurd raises prices, signs long AI capacity contracts, eyes higher 2026 capex

The back-end of semiconductor manufacturing is undergoing a structural repricing. Sigurd's move signals a paradigm shift from spot-market dynamics to an allocation-based model—long-term contracts locking capacity, pricing anchored annually rather than negotiated quarterly. The economics of AI chip test have fundamentally changed. On the technical front, AI GPU and ASIC test complexity runs three to five times that of conventional CPUs. Multi-die interconnect validation, high-power verification, and silicon photonics optical coupling checks multiply per-unit test time. HBM's KGD screening demands 100% full test, consuming multiples of standard DRAM capacity. These three vectors compound, making test capacity far less elastic than front-end wafer production. Competitively, ASE and Amkor have expanded AI test lines, but ATE equipment lead times from Advantest and Teradyne impose a rigid 12-to-18-month delivery lag. Sigurd's 2026 capex commitment is essentially a bet that ASIC-based inference chips—not just training GPUs—will become the dominant test demand driver. Risk: US export controls on ATE equipment haven't directly hit Taiwan, China-based providers, but customer concentration among US AI giants creates order volatility exposure. Long-term contracts lock revenue but also lock technology roadmaps—if architectures pivot toward photonic computing or 3D stacking by 2027, existing test assets face accelerated depreciation. Outlook: Over the next 12-24 months, test capacity will replicate the 2020-21 fab-shortage pattern, but persist longer due to higher capital intensity. A 15-25% price upside is likely sustained.