TSMC’s Capacity Crunch Squeezes Nvidia and Broadcom

TSMC, the world’s largest contract chipmaker, has told major customers Nvidia and Broadcom that it cannot fully meet their growing demand for production capacity, highlighting mounting stress across the semiconductor supply chain as AI-driven demand accelerates.

The situation exposes a core paradox in today’s chip industry. While demand for advanced processors powering artificial intelligence, data centres, and high-performance networking is surging, the ability to manufacture these chips remains severely constrained. TSMC’s most advanced process nodes—particularly 3nm and 5nm—are already running close to full utilisation, leaving little flexibility even for its biggest clients.

For Nvidia, the constraint is especially critical. As the dominant supplier of AI accelerators, Nvidia’s growth depends not just on design leadership but on access to advanced manufacturing at scale. Any limitation on wafer supply risks slowing GPU shipments to hyperscalers and enterprises, turning manufacturing capacity into a strategic bottleneck rather than a routine supply issue.

Broadcom faces similar challenges. Its custom silicon and networking chips are becoming essential to AI data centre architectures. Restricted access to production capacity could delay customer deployments, forcing staggered rollouts and potentially pushing revenue recognition further out.

From TSMC’s perspective, the crunch reflects the realities of capital-intensive chipmaking. Advanced fabs take years to build and require investments running into tens of billions of dollars. Despite aggressive expansion plans across Taiwan, the US, and Japan, near-term capacity remains limited. As a result, TSMC must carefully prioritise customers while protecting long-term margins and relationships.

The broader takeaway is stark: AI demand is now outpacing the semiconductor industry’s ability to scale quickly. Capacity constraints at TSMC underline the growing importance of foundry diversification, advanced packaging technologies, and long-term supply agreements. Until new fabs come online, even the world’s most powerful chip designers may find their ambitions capped—not by market demand, but by manufacturing reality.