TSMC just raised its 2026 capital spending forecast to between $52 billion and $56 billion. That’s a sharp jump. Demand for its most advanced chips keeps outpacing what the company can produce. Executives point to one driver above all others. Artificial intelligence.
The world’s largest contract chipmaker reported second-quarter results that topped forecasts. Revenue climbed. Profits beat estimates. Yet shares dipped more than 2% in the session after the numbers came out. Investors focused less on the beat and more on the price tag attached to future growth. Higher spending. Thinner margins in some areas. And a stock that had already run hard.
AI Demand Forces Record Spending Plans
But the numbers tell a story of sustained strength. TSMC lifted its full-year revenue growth outlook. Management now sees more than 40% expansion for the year. The phrase “AI megatrend” came up repeatedly on the earnings call. It isn’t marketing. It’s shorthand for orders from Nvidia, Broadcom, Apple and others racing to build data centers that train and run ever-larger models.
One week after the report, the company unveiled plans for another $100 billion in U.S. manufacturing. That brings its total American commitments since 2025 to $200 billion, according to CNBC. The latest tranche targets advanced logic and packaging facilities in Arizona. Four new fabs in total. The announcement arrived amid fresh pressure from Washington to move production stateside. President Trump highlighted the deal on social media, calling it a win for American jobs and semiconductor independence.
Demand signals stretch years into the future. TSMC now guides with visibility out to 2029 in some segments. Hyperscalers aren’t slowing down. Meta plans to put its own custom AI accelerator into production in September. The chip, designed with help from Broadcom and built at TSMC, aims to double the social network’s computing capacity. Reuters first reported the timeline and scope.
Capacity remains the bottleneck. Even with aggressive fab builds, TSMC executives said AI-related demand exceeds what they can supply through at least 2027. That mismatch explains the capex surge. It also explains why the stock, despite strong results, trades at 25 times forward earnings. Reasonable for a company growing north of 40%. Still, the market wants proof that all this spending will translate into sustained high margins.
Overseas expansion carries costs. U.S. fabs come with higher labor and regulatory expenses. Water usage in Arizona has already sparked local debate. Recent X posts from accounts monitoring the Phoenix area highlight concerns that TSMC could become the metro region’s largest water customer once all facilities come online. The company counters with heavy investment in recycling systems. Net consumption matters more than gross withdrawals, officials argue.
Geopolitics adds another layer. Taiwan remains the heart of TSMC’s manufacturing. Yet the firm accelerates diversification. Arizona. Japan. Germany. Each new site insures against potential conflict in the Taiwan Strait. Each also dilutes near-term profitability. Second-quarter earnings reflected some of that pressure. Margins compressed slightly on the higher mix of mature nodes and ramp costs for new U.S. capacity.
Analysts remain largely bullish. A Barron’s analysis after the results showed how AI has reshaped the entire semiconductor value chain. TSMC sits at the center. Its CoWoS advanced packaging technology, critical for high-bandwidth memory integration in AI accelerators, remains oversubscribed. Customers pay premiums for priority access. That dynamic supports pricing power even as capex climbs.
Competitors watch closely. Intel ramps its own foundry ambitions. Samsung pushes into advanced nodes. Neither has matched TSMC’s yields or scale in 3-nanometer and 2-nanometer processes. The gap shows in the financials. TSMC’s market capitalization has more than doubled in the past year. Its dominance in leading-edge production now exceeds 90% for chips below 7 nanometers.
Investors who focus only on near-term margin pressure miss the bigger picture. AI training clusters keep scaling. Inference workloads are just beginning to move to specialized silicon. Every major cloud provider builds its own chips. All of them rely on TSMC. A Motley Fool piece from June noted the widening moat. The article argued TSMC stands in a league of its own within the AI supply chain.
Supply chain breadth matters too. The spending doesn’t stop at TSMC. Memory makers SK Hynix and Micron ramp high-bandwidth memory output. Equipment suppliers like ASML see multi-year tailwinds. Power and cooling specialists benefit from the enormous electricity draw of new data centers. One recent analysis on X mapped how a single $100 billion hyperscaler AI investment ripples across dozens of vendors. TSMC captures a sizable slice.
Still. Risks exist. A slowdown in AI enthusiasm could leave expensive fabs underutilized. Trade tensions between the U.S. and China add uncertainty. Beijing restricts certain advanced chip imports. Washington tightens export controls on manufacturing tools. TSMC walks a narrow path between its largest customer base and its home government’s security concerns.
Executive commentary struck a measured tone. They acknowledged higher spending but emphasized long-term visibility. Revenue growth above 40% this year. Mid-teens percentage growth expected in 2027 and beyond. The “AI megatrend” comment wasn’t throwaway. It reflects internal forecasts that show no letup in orders for 3nm, 2nm and the upcoming A16 process.
Recent social media chatter on X reinforces the momentum. Multiple accounts noted that beating earnings no longer suffices for AI-related stocks. Markets now penalize any sign of margin dilution or slower-than-expected capacity ramps. TSMC felt that dynamic firsthand. Shares recovered some ground in subsequent sessions as analysts raised price targets.
The company also faces questions about concentration. Nvidia represents a growing share of revenue. Estimates suggest AI-related sales could approach 60% of TSMC’s wafer revenue by 2027. Diversification across customers helps. But the fortunes of the largest AI accelerator maker still matter enormously.
Production timelines for the new Arizona facilities stretch into 2027 and 2028. Initial output will focus on less advanced nodes before shifting to leading-edge. That gradual ramp gives the company time to train local workers and optimize processes. It also means near-term costs hit financials before the revenue fully materializes.
Industry watchers point to 2026 as a pivotal year. TSMC’s raised guidance surprised to the upside. Few analysts had penciled in capex that high. The move signals confidence. It also signals that even the most optimistic forecasts may have understated how quickly AI infrastructure scales.
Meta’s custom chip move adds another data point. By bringing design in-house and manufacturing at TSMC, the company reduces dependence on Nvidia and AMD. Other hyperscalers pursue similar strategies. Google, Amazon and Microsoft all run custom silicon programs. Each design wins requires TSMC capacity. The cumulative effect keeps the order book full.
Water and power constraints could limit further expansion in certain regions. Arizona’s desert climate makes cooling especially expensive. Local officials and TSMC continue discussions on sustainable usage. Recycling rates above 80% form part of the solution. Long-term planning, not short-term fixes, will determine success.
Investors weighing the stock today confront a mix of signals. Strong demand. Record capex. Geopolitical tension. Attractive valuation relative to growth. The pullback from June highs created an entry point for some. Others wait for clearer signs that margins have bottomed.
TSMC doesn’t design the chips that power AI. It makes the chips possible. That distinction matters. Without manufacturing leadership at the frontier, the entire industry stalls. The company’s latest moves show it intends to stay years ahead. The price tag is high. The payoff, if demand holds, could be higher still.
Recent coverage from CNBC on the Arizona expansion underscores the dual nature of the bet. More American production satisfies policymakers. It also raises costs in the short run. How those costs balance against pricing power in advanced packaging will shape profitability for years.
The AI buildout continues. TSMC remains the indispensable partner. Its willingness to spend at this scale reflects conviction that the megatrend has staying power. Markets will keep testing that conviction. Results over the next several quarters will provide the answers.
WebProNews is an iEntry Publication