Taiwan Semiconductor Manufacturing Company raised its 2026 capital expenditure guidance on January 1, extending a broader semiconductor buildout triggered by AI chip demand. The move follows Intel's December 2024 Malaysia investment and Infineon Technologies' Dresden facility expansion, where construction is underway with production targets set for late 2026.
The capex increases come as chip designers pursue specialized AI connectivity infrastructure. Marvell Technology entered acquisition talks with Celestial AI in December 2024, targeting the startup's optical interconnect technology. These photonics-based chips move data between GPUs in training clusters, addressing bandwidth constraints that slow large language model development.
Infineon confirmed final funding approval for its Dresden site, which will manufacture power semiconductors for AI data center cooling systems. Microchip Technology reported "broad-based recovery" across end markets in its revised guidance, citing AI infrastructure as a primary growth driver alongside automotive and industrial segments.
The coordinated expansion suggests chip manufacturers expect AI hardware demand to persist through 2027. TSMC's guidance increase specifically targets advanced node capacity for 3nm and 2nm chips used in Nvidia's Blackwell and future Rubin GPU architectures. Intel's Malaysia facility will produce packaging substrates for AI accelerators, a component facing supply shortages since mid-2024.
Optical interconnect acquisitions like Marvell-Celestial AI reflect a strategic shift toward solving data movement problems rather than pure compute power. Current copper-based chip-to-chip links hit physics limits at 800 Gbps, while training GPT-5-scale models requires 1.6 Tbps interconnects. Celestial AI's optical technology promises 10 Tbps bandwidth using silicon photonics.
The semiconductor M&A and capex pattern creates a testable market hypothesis: AI chip revenue growth should correlate with optical computing acquisitions through 2026. TSMC, Intel, and Infineon spending patterns will indicate whether manufacturers see AI demand as cyclical or structural. Early signals point to structural bets—Infineon's Dresden timeline extends to 2028, suggesting confidence beyond near-term AI hype cycles.