Signals from the leadership of Taiwan Semiconductor Manufacturing (TSM.US) regarding the AI computing power supply chain indicate the world's largest foundry will be unable to catch up with AI-driven demand for "years," a long-term positive signal for the stock price and fundamentals of TSMC and global AI supply chain participants, according to a Friday research report from Wall Street investment firm Wedbush Securities.
While recent investment fervor in the AI computing power chain has shown signs of cooling, with TSMC's U.S. ADRs falling nearly 3% in Friday's pre-market trading following Broadcom's AI semiconductor growth outlook falling short of Wall Street expectations, the messages from TSMC management and Wedbush strongly reinforce the long-term thesis of robust AI infrastructure demand.
At the annual shareholder meeting, CEO C.C. Wei stated that even with new U.S. capacity coming online, TSMC will struggle to fully meet AI-driven demand for several years. This suggests the bottleneck extends beyond short-to-medium-term CoWoS or HBM capacity, pointing instead to a systemic shortage across the global AI computing wafer supply chain involving advanced logic chips, advanced packaging, AI accelerators, and high-performance computing wafer capacity.
The series of positive signals conveyed by TSMC's CEO this week highlight that AI computing infrastructure supply constraints are evolving from a singular chip/wafer shortage into a full-stack, long-term capital expenditure cycle encompassing data center CPUs, high-performance networking infrastructure, optical interconnect chips, and 2.5D/3D advanced packaging capacity.
In the view of analysts at Bank of America, AI computing infrastructure is entering a more enduring and broader capital expenditure supercycle. Nearly simultaneously, a report from Morgan Stanley indicated the AI computing arms race is entering a system-level expansion phase, with AI infrastructure demand showing a rare "inelastic" trend—where tech giants continue to build AI data centers regardless of cost curves—a trend expected to bolster U.S. economic resilience and S&P 500 earnings growth.
Wedbush's Analysis of TSMC's Long-Term Demand Outlook
Wedbush noted that CEO C.C. Wei's latest comments align with Broadcom CEO Hock Tan's remarks about major customer orders extending into 2028, implying leading-edge logic and advanced packaging could remain in "severe undersupply" well beyond 2026. This positions TSMC to potentially exceed Wall Street's optimistic revenue and margin expectations for 2026-2027 through 2nm capacity ramps, product mix improvements, and measured price increases.
Senior analyst Matt Bryson wrote in a client note that Wei's statement about demand exceeding supply for years matches the information from Hock Tan, reinforcing the long-held investment thesis that the most advanced CPU/GPU logic chips and advanced packaging will remain structurally undersupplied with firm pricing well beyond 2026.
For TSMC, Wedbush believes that measured price increases, positive product mix effects from the 2nm ramp and additional leading-edge capacity, coupled with stable mature node output, should allow the company to surpass Wall Street sales and margin expectations in 2026 and 2027, assuming continued execution on capacity and technology roadmaps, particularly for 2nm.
At the shareholder meeting, Wei repeatedly emphasized that TSMC will be unable to meet the seemingly insatiable AI infrastructure demand for years, even with new U.S. advanced process or packaging capacity. He reaffirmed TSMC's overall revenue growth will exceed 30% this year and expressed willingness to raise prices, albeit in a "measured, sustainable manner," contrasting this with the sharp hikes seen in the memory chip sector.
Wei also identified AI humanoid robots and autonomous driving as TSMC's next long-term growth drivers.
Market Reactions and the Underlying Demand Thesis
The recent sell-off in U.S. AI supply chain stocks and the South Korean market appears more reflective of a valuation digestion phase for a crowded trade theme rather than a discrediting of underlying AI demand. Broadcom's post-earnings drop primarily reflects the market's extreme scrutiny of AI semiconductor expectations, where any guidance that isn't explosively above consensus triggers profit-taking.
Similarly, declines in South Korea's KOSPI index and stocks like Samsung and SK Hynix highlight the sensitivity of AI-related assets to any marginal disappointment after significant year-to-date gains. In essence, secondary markets are rotating from high-crowding AI themes toward defensive, lower-valuation leaders, while signals from AI supply chain giants like TSMC and Hon Hai continue to validate that "AI capital expenditure and computing orders are still being realized."
On the specific topic of AI capital expenditure, Wei's shareholder meeting comments that he "doesn't know where the peak is" and "doesn't see any indicators of demand stopping" represent perhaps the most bullish supply chain commentary from an industry titan.
From an engineering perspective, TSMC's latest stance carries the highest industrial weight, as it sits at the core "physical bottleneck layer" of AI infrastructure. Whether it's Nvidia/AMD AI GPUs, Broadcom/Marvell ASICs and networking chips, AMD AI accelerators, or cloud provider custom ASICs, all depend on advanced processes, advanced packaging, and high-yield mass production. If TSMC remains supply-constrained for years, it signals AI data center construction is not a single-quarter inventory pull but a multi-year capex cycle expanding from training GPUs to inference, Agentic AI, robotics, autonomous driving, and sovereign AI.
TSMC itself emphasized in shareholder materials that it will continue investing in Taiwan's leading-edge processes and advanced packaging, advancing multi-phase 2nm-class capacity construction to meet strong computing demand driven by AI.
The Shift to Profitability in AI Applications
For AI monetization prospects, the recent strong revenue trajectory and better-than-expected profit path announced by Anthropic—which previously rattled global software stocks with its AI agent launches—are likely to further accelerate tech giants' efforts to build "AI computing super-empires."
Anthropic's expectation for Q2 revenue to double and achieve its first operating profit highlights the AI industry's shift from a "cash-burn narrative" to a "cash-flow cycle narrative." Most significantly, Anthropic's monthly ARR surge of $110 billion is equivalent to the combined scale of SaaS giants Palantir, Snowflake, and Databricks over a decade—an unprecedented phenomenon.
Anthropic expects Q2 revenue to jump from $4.8 billion in Q1 to $10.9 billion, with operating profit around $559 million, demonstrating that frontier AI applications are not just consuming computing power but are beginning to transform enterprise programming, agent workflows, cybersecurity, and data analysis into high-value token revenue.
The explosion in demand for Claude and AI tools is pushing Anthropic toward its first profitable quarter, with computing costs per dollar of revenue dropping from about 71 cents in Q1 to roughly 56 cents in Q2, showing scale effects and inference efficiency are improving the AI application economic model.
This underpins why Morgan Stanley proclaims the AI computing arms race is entering a system-level expansion phase. The firm has significantly raised its 2026 capital expenditure forecast for U.S. mega-cap tech giants from $433 billion a year ago to $805 billion, with 2027 capex potentially reaching $1.1 trillion, up from a prior forecast of $950 billion. It further predicts that by 2028, nearly $3 trillion in AI-related infrastructure investment will flow through the global economy, with over 80% of that spending still ahead.
Comments