MMarket PulseScope요청 / 문의
All market news
AI HARDWARE

AI Server Buildout Widens as Wiwynn Opens Texas Hub and Google Flags One-Year Silicon Payback

Wiwynn opened a US$1.6 billion AI server plant in Socorro, Texas, while Google Cloud's CEO said in-house silicon pays for itself in about a year. PwC and Oxford Economics project US$31.6 trillion of cumulative global data centre capex by 2050, with power emerging as the binding constraint.

AI server manufacturing capacity is moving closer to its end customers. Wiwynn has officially opened an advanced manufacturing facility in Socorro, Texas, with US$1.6 billion of investment, deepening the Taiwan-based server maker's US footprint at a time when hyperscalers want shorter, more resilient supply lines for rack-scale systems. The investment follows a broader pattern of Taiwanese ODMs shifting final assembly for GPU and ASIC racks into North America.

The economics of that capacity are holding up. Google Cloud CEO Thomas Kurian revealed that Google's overall payback period on AI server investments is less than two years, with in-house silicon recovering its cost in half that time, roughly one year. DIGITIMES reads the disclosure as upside for MediaTek and Broadcom ASIC revenues, since custom silicon is where the fastest payback sits, and it reinforces the shift from general-purpose GPU procurement toward custom accelerators for inference workloads.

Long-horizon demand assumptions keep being revised up. PwC and Oxford Economics released their Global Data Centre Outlook 2026-50 in early September, projecting cumulative global data centre capex of US$31.6 trillion by 2050 under the baseline scenario. That figure is not a near-term order book, but it is what underwrites the multi-year commitments now being signed for land, substations, cooling capacity and long-lead electrical equipment.

Power, not silicon, is increasingly the limiting factor. DIGITIMES Intelligence reports that the semiconductor sustainability and energy forum at SEMICON Taiwan 2026 centred on whether Taiwan will have enough electricity to support continued AI-driven semiconductor expansion, and on the practical limits of meeting that additional demand with renewable energy. For AI hardware suppliers, the implication is that thermal and power-delivery content per rack keeps rising even when unit shipment growth is throttled by grid interconnect and substation lead times.

Sources