HSINCHU — On the day technology stocks around the world were being sold off, Taiwan Semiconductor Manufacturing Co. told investors the opposite story. The world’s largest contract chip maker said it expects the global semiconductor market to reach $1.5 trillion by 2030, roughly double its current size, with nearly all of the growth driven by chips for artificial intelligence.
Executives delivered the forecast at an investor conference on Tuesday, hours after the KOSPI in Seoul had halted trading twice and Nvidia had fallen as much as 15% in New York. The timing appeared deliberate: a statement from the industry’s most important supplier that the buildout of AI computing capacity would continue regardless of a day of panic in the stock market.
“We are not slowing down,” one executive said, according to a person who attended the meeting. “The orders we see are not speculative.”
TSMC manufactures the most advanced processors sold today, including Nvidia’s AI accelerators and Apple’s iPhone chips, and its capacity decisions effectively set the pace of the entire semiconductor industry. A doubling of the market to $1.5 trillion would require the company and its rivals to keep adding fabrication plants at a record clip for the rest of the decade, analysts said.
The company has been building toward that scale for years. It has broken ground on new facilities in Arizona, Japan and Germany, on top of its sprawling campuses in Taiwan, and has repeatedly raised its capital-expenditure budget to keep pace with demand from hyperscale customers such as Microsoft, Amazon and Google. Each of those customers has said it will spend more on AI infrastructure this year than last, and TSMC’s order book reflects that.
The forecast is not the company’s first bullish call on AI. But the timing distinguishes it, analysts said. By publishing the projection on the worst trading day of the year for chip stocks, TSMC was signaling to investors that near-term volatility in share prices would not alter its multi-year plans for capacity, pricing or hiring.
There are risks embedded in the forecast. A $1.5 trillion market by 2030 assumes that AI demand grows at a compound rate that has no precedent in the industry’s history, and that memory, logic and packaging capacity all expand in lockstep. It also assumes that the geopolitics of chip manufacturing, a factor that has reshaped the industry’s supply chains since 2020, remain manageable.
The company’s customers are making the same bet. Nvidia, AMD and Broadcom have all booked wafer capacity through 2027, according to people familiar with the arrangements, and TSMC’s advanced packaging lines, a bottleneck for AI accelerators, are sold out through next year. Those bookings convert the market forecast from aspiration into committed revenue, analysts said.
For investors, the projection cuts both ways. Bulls read it as evidence that the AI capital-expenditure cycle has years to run, and that Tuesday’s selloff was a valuation event rather than an end to demand. Bears note that chip companies have a history of extrapolating current shortages into permanent growth, and that the last time the industry promised a supercycle, in 2021, it was followed by one of the deepest downcycles on record.
TSMC’s executives acknowledged the tension but did not concede the point. The company said its backlog is longer than it has ever been, and that customers are signing take-or-pay agreements that commit them to capacity even if demand softens. Those agreements, more than the market forecast, are the concrete evidence that the buildout will proceed, analysts said.
The company’s own share price did not escape the rout: TSMC’s Taiwan-listed shares fell along with the broader market on Tuesday. But executives’ message was that the gap between stock price and order book is temporary. If the forecast proves right, the paper losses of June 23 will be a footnote in a decade of growth; if it proves wrong, the industry will have built far more capacity than it can use.
The forecast rests on a specific assumption about technology: that AI models will keep getting bigger and that running them will require more chips, not fewer. TSMC’s executives pointed to the industry’s transition to the 2-nanometer manufacturing node and beyond, where the company’s advanced packaging technology, which stacks chips and memory together, has become as important as the fabrication process itself. The company’s advanced packaging lines are sold out through next year, according to people familiar with its capacity plans, and the bottleneck has become a talking point for the entire industry. Nvidia’s chief executive has described the packaging constraint as the difference between what AI can do and what it can deliver, and TSMC’s expansion of that capacity is the single largest capital program in the company’s history.
The geography of the expansion is as important as its scale. TSMC has spread its new capacity across Taiwan, Arizona and Japan, a diversification driven as much by politics as by economics, and each new site has brought its own complications, from water supply in the desert to earthquakes on the island. The company’s executives said the buildout is being managed with an eye on both demand and risk, and that the multi-site strategy gives customers a hedge against the concentration that has defined the industry for decades. The result is a manufacturing map that looks nothing like the one that produced today’s chips, and the transition is being paid for with the profits of the current boom.
For now, the order book is the fact that matters, and it points in one direction. TSMC’s fabs are full, its customers are committed, and its executives are telling investors to plan on a market twice today’s size. The market’s job, one analyst said, is to decide whether to trust the forecast or the panic.
