“AI-related demand continues to be extremely robust. The shift from generative AI and the query mode to agentic AI and the command and action mode is leading to another step up in the amount of tokens being consumed. This is driving the need for more and more computation, which supports the robust demand for leading-edge silicon.”
TSMC CEO C.C. Wei’s assessment highlights a fundamental re-rating of the global semiconductor supply chain. As the industry pivots from passive query-based models to autonomous, action-oriented agents, the compute bottleneck is migrating. The industry is shifting from a GPU-exclusive shortage to a broader, systemic scarcity where demand for CPUs and other components is surging alongside accelerators.
TSMC’s Q2 2026 results provide the hard data to back this shift. The foundry reported revenue of $40.2 billion — a 36% year-over-year increase — and a staggering 77% jump in net income to $22.36 billion. With gross margins hitting 67.7% and High-Performance Computing (HPC) now accounting for 66% of total revenue, TSMC is effectively the toll booth for the AI revolution. The company’s decision to raise its 2026 capex to $60–64 billion, up from the previous $52–56 billion, and lift revenue guidance to above 40%, signals that the demand for leading-edge silicon is not just high — it is accelerating.
However, the financial success masks a deepening structural crisis. TSMC has explicitly identified a “really big demand-supply gap” for AI silicon, exacerbated by critical bottlenecks in tester capacity and CoWoS advanced packaging. While CoWoS capacity is scaling to 115k–140k wafers per month, the supply-demand gap is expected to narrow only to 10% by the end of 2026. This is a supply chain operating at its absolute limit.
The most significant consequence of this shift is the sudden, intense pressure on CPUs. According to BofA, the rise of agentic AI is driving a surge in CPU demand, leaving Intel and AMD inventories nearly sold out and setting the stage for price hikes of 10% to 50%. AMD CEO Lisa Su has underscored this transformation, projecting the server CPU market to reach $120 billion by 2030 with a CAGR exceeding 35%. Crucially, the industry is seeing a shift in the CPU-to-GPU ratio from 1:8 to 1:1, as agents require more sophisticated orchestration and logic processing than simple generative tasks.
TSMC is betting its future on this sustained demand. The company is committing an additional $100 billion to build four more 2nm fabs in Arizona, bringing its total investment in the region to $265 billion. With 70% to 80% of its total capex now funneled into advanced nodes, TSMC is effectively locking the global supply chain into a high-cost, high-stakes trajectory.
Despite these record-breaking results, TSM stock saw a sell-off, reflecting investor anxiety over the sheer scale of capital expenditure and the fragility of the supply chain. Yet, with analysts raising price targets to $500, the long-term thesis remains clear: the compute squeeze is no longer a GPU-specific phenomenon. It has expanded into a comprehensive, structural shortage across CPUs, advanced packaging, and memory, ensuring that the cost of compute will continue to rise.