Lam Research Founder 林杰屏: AI Supercycle and 40 Years of Chip Cycles

Summary

• 林杰屏 sees AI becoming a more important semiconductor driver than Moore’s law, with AI and advanced packaging marking 2 major inflection points happening simultaneously. Moore’s law is hitting a physical limit as dimensions shrink so far that quantum effects may make transistor switching ambiguous; economically, each node-to-node transition now costs hundreds of millions of dollars and is approaching its ceiling. Advanced packaging offers another way to manufacture chips. His broader view: AI is “not merely an industrial revolution, but a cognitive industrial revolution.”
• He acknowledges that more than $700B in capex from major tech companies this year—“almost equivalent to the US defense budget”—reflects some overspending and overconfidence, but says it is fundamentally different from the internet bubble. In 1999–2000, 80% of listed startups had zero revenue, P/E ratios reached 100x and peaked at 200x, and “trillions of dollars in assets vanished” after the bubble burst; today’s protagonists are well-capitalized incumbents such as Amazon and Google, while Nvidia can generate 71% gross margins and tens of billions of dollars in net income each quarter. The real constraint is electricity: “Power shortages will be far more serious than our concerns about whether spending is excessive.”
• He is explicitly betting that 陈立武 can turn Intel around: “If I wanted to bet, I would bet that he can win”—but “it will be difficult, and it will not happen quickly.” He once wondered whether 陈立武 was simply better at investing, but Cadence’s turnaround changed his mind; the counterexample is 保罗·欧德宁’s decision around 2007 to reject Apple’s iPhone chip business because gross margins were too low. In 林杰屏’s view, Intel made a “double mistake” by missing low-margin mobile products and putting a nontechnical executive in charge of a technology company, “completely missing the entire mobile market.”
• The central lesson from 40 years of cycles is that cycles cannot be predicted; only inflection points can be observed. Equipment buyers are squeezed either way: expand too aggressively and get fired when the fab is finally built after the boom has passed, or stay conservative and get fired when competitors take the share. “That is why nobody can predict” the cycle. Inflection points leave incumbents unable to adjust in time, creating openings for new entrants: “Opportunity is always hidden in turning points.”
• The equipment business is “lumpy”: you either spend a few million dollars on one tool or $180M on one ASML lithography system; you either get an order or get nothing. Customers can shut off capex almost instantly, and fabs may stop paying “almost overnight.” In the roughly $140B equipment market cited by the host, lithography accounts for about 25%, etch and deposition roughly 20% each, and metrology and inspection about 12%; 林杰屏 expects that mix to change very little over the next 5 years.
• Multibeam’s strategy is maskless, multicolumn electron-beam direct writing aimed at applications optical lithography cannot handle. A mask is like a cookie cutter and takes weeks to make; after changing the layout, an e-beam system can rerun it in about 2 hours. Its depth of focus reaches 5–10 microns versus roughly 100–200 nanometers for lithography, fitting advanced-packaging 3D stacking and quantum chips much larger than photomasks. There were about 4 companies in the field 5–10 years ago; the others have gone bankrupt, disappeared or abandoned their projects, leaving Multibeam focused on multicolumn parallel writing.
• He does not want to use parallel e-beam writing to solve DRAM’s EUV bottleneck—strategic discipline comes before market temptation. Giants may move slowly, but “they have thousands of people, while we have 75 … they have billions of dollars in the bank, while we have millions.” The better strategy is to find inflection points where the giants cannot compete. He read The Innovator’s Dilemma twice, taking the small-company perspective the second time; the CEO of a 5¼-inch floppy-drive company said, “We can make it, but nobody wants it,” and ultimately missed the laptop wave.

Deep dive

1. AI + advanced packaging: 2 inflection points at once

• 林杰屏’s opening judgment is that AI is now driving the semiconductor industry, “even more importantly than the previous driver, Moore’s law.” Moore’s law has hit a physical ceiling: dimensions have shrunk until one can “almost only see atoms,” creating the possibility of quantum effects that make it difficult to distinguish whether a transistor is on or off. It has also hit an economic ceiling: moving from one node to the next costs hundreds of millions of dollars, and “the money earned is not even enough to pay for all the R&D.”
• Advanced packaging is the successor, “another way to manufacture chips.” The fact that AI and advanced packaging are arriving as 2 simultaneous inflection points makes the shift “even more significant for the industry.” 林杰屏’s upgraded characterization: AI is “not just a turning point, but a true industrial revolution … a cognitive industrial revolution.”

2. $700B in capex: a bubble, but not another 1999

• 曹卿云 put the number on the table: major tech companies’ combined capex has exceeded $700B this year. Is that rational investment or strategic overspending? 林杰屏’s answer is that customers are spending because their own customers need chips, and these companies are “very good at adjusting” their capex budgets. But is there a bubble? “Yes, I think so too.”
• The sharpest contrast is with the internet bubble. At the time, “80% of listed startups had zero revenue”; P/E ratios were pushed to 100x and peaked at 200x, before the March 2000 collapse vaporized trillions of dollars. Today’s participants are mature companies such as Amazon and Google, while Nvidia can achieve 71% gross margins and generate tens of billions of dollars in net income every quarter.
• His real concern is not excessive spending but electricity: “Once the data centers are built, where will they get the power needed to run?” Lam Research is working with at least 8 partners to address the problem. Bubbles typically damage companies that are unprepared; the survivors may have a chance to become the winners of the next cycle.

3. History lesson: Intel exits memory, TI Japan’s makeup room and the Plaza Accord

• Intel’s 1985 decision to abandon memory and focus on logic was “one of the best strategic decisions Intel ever made.” 林杰屏 added a story he witnessed firsthand: TI Japan’s yields were consistently higher than those in Dallas. The explanation was makeup powder from female fab technicians landing on wafers. In Japan, employees could be required to arrive at work without makeup and use a makeup room before leaving; “in the US, you cannot do that.”
• He attributes Japan’s semiconductor decline to the Plaza Accord, signed by Japan, the US, the UK, France and Germany. The US wanted Japan to change its exchange rate to slow exports to the US, triggering the “lost decade”—which, in retrospect, was 30 years of slowing growth. “The 2 decisions Japan made were the real reasons its semiconductor industry declined.” The downturn did not spread to the US: Lam Research went public in 1984 as planned and continued to grow.

4. Cycles cannot be predicted; opportunity sits in the inflection point

• The procurement dilemma captures the difficulty of navigating cycles. Discover new demand and expand capacity, and by the time the fab is up and running, the boom may be over and you may be fired for spending too much. Stay cautious, and when the cycle arrives there may be nothing to sell while competitors take share; you may also be fired for negligence. “That is why nobody can predict” the cycle.
• 林杰屏’s method is not to guess when a cycle starts or ends, but to watch for inflection points—the moments when the slope of the growth curve changes sharply. Incumbent leaders “cannot adjust quickly enough” at those points, opening the door to new entrants. But an inflection point is only an opportunity; the next question is, “Do we have the technology to solve the problem?” “Opportunity is always hidden in turning points.”

5. Intel’s double mistake and the bet on 陈立武

• 保罗·欧德宁 spent 31 years at Intel and was highly respected. Around 2007, he rejected 乔布斯’s iPhone chip order because the gross margin was too low—a standard shaped by more than 20 years of PC dominance. 林杰屏’s principle is the opposite: “When you see a turning point, the first product you launch is unlikely to have a high gross margin. If you make a little money, consider yourself lucky.” Intel consequently missed the entire mobile market and “there was no second chance.”
• The second mistake was “letting a nontechnical person run a technology company.” 林杰屏 believes a board should not choose a CEO without an engineering or science background, while emphasizing, “This is not a criticism, just an observation.”
• On 陈立武, whom he has known for more than 20 years, 林杰屏 admitted his earlier doubt: “He might be very good at investing. Is he good at running a large company?” Cadence’s turnaround left him “amazed”—quick-minded and strategic in his decisions. His conclusion came with every qualification intact: “It will be difficult, and it will not happen quickly … but if I wanted to bet, I would bet that he will win.”

6. The equipment industry’s “lumpiness”: an overnight shutdown

• Equipment sits at the top of the value chain and is sold to foundries. AMD and Broadcom are called “chipmakers,” but in reality they only design chips. Equipment companies do not deal directly with consumers. “That is why our name recognition is lower than that of chipmakers.”
• Why can equipment become a constraint more easily than chips? “I call it ‘lumpy.’” You either spend a few million dollars on one tool or $180M on one ASML lithography system; you either get an order or get nothing. Once customers see a problem coming, they “shut off the capex valve,” and a fab can stop paying “almost overnight.” Lithography is the most complex and precision-intensive segment, with longer and more complicated lead times and supply chains, and higher prices overall.
• In the roughly $140B equipment market cited by the host, lithography represents about 25%, etch and deposition about 20% each, and metrology and inspection about 12%. 林杰屏 expects the mix to change very little in 5 years. Rising complexity is a slow-moving process—“I would not worry about it at all.” Geopolitics and supply chains are realities smaller companies cannot influence; they can only be handled one issue at a time.
• He cannot identify a fixed cycle pattern in the purchasing behavior of foundry, memory and IDM customers. A shortage of memory or logic chips can both drive equipment orders; there is no standing rule. He watches the market and order changes every day to protect the company, not to predict the future.

7. Multibeam: maskless, deep focus, and outside lithography’s reach

• The second startup was almost accidental. After retirement, 林杰屏 “got tired of being retired,” and a VC asked whether he would acquire the assets of a startup that had operated for 10 years before failing but held strong patents. When he once said he was working on e-beam lithography, “everyone laughed me out of the room”—a fine pen is obviously slower than a “cookie cutter,” or mask.
• The reversal lies in the full workflow. Masks take weeks to produce, and complex chips take even longer. Maskless direct writing uses onboard software to convert layout data layer by layer into e-beam exposure patterns: “2 hours later, you change the layout and run it again.” The system can use up to 3 modules, with multiple beams writing in parallel and joining seamlessly. “You can manufacture a chip as large as an entire wafer.” Quantum chips “absolutely love us” because they do not require stitching.
• The second advantage is depth of focus. Lithography offers only 100–200 nanometers, making connections in advanced-packaging 3D stacks difficult. Multibeam reaches the micron scale: “5 microns, 10 microns? No problem”—2 orders of magnitude greater. There were about 4 companies in the field 5–10 years ago: the European one went bankrupt, the Japanese one “quietly disappeared,” and the US company abandoned its project. “We are the only company in the world that can perform multicolumn parallel writing on a wafer.”

8. Strategic discipline: do not volunteer to solve the EUV bottleneck

• Asked whether parallel e-beam writing could ease DRAM’s bottleneck from limited EUV supply, 林杰屏’s answer was blunt: “In theory, perhaps. Strategically, I do not want to do that.” The giants may move slowly, “but the moment I try to take a share of their business, they will wake up.” “They have thousands of people. We have 75 … they have billions of dollars in the bank. We have millions.”
• He is looking for asymmetry: “Our technology is outstanding. They cannot compete with it.” He says “our system costs only 5% of other systems.” Customers initially complained that it was expensive, but accepted it after cost analysis and modeling showed that the total cost was far lower. The focus areas are AI, quantum, compound semiconductors and advanced packaging. If ASML slows production for any reason and creates an equipment bottleneck, he will not enter the market: “These people are too powerful. I simply cannot compete with them.”

9. “Good workers, not good racehorses”: from discrimination to prepared entrepreneurship

• In 1979, 林杰屏 decided to leave after being denied a promotion that was not even to manager. The role went to a white employee who had been on the job for only 11 months after a postdoctoral program. The prevailing line at the time was that “Asians are good workers, but not good racehorses.” Today, role models include 黄仁勋, 苏姿丰, 陈福阳 and 陈立武. Bias has eased, but “geopolitics and other factors will replace it—and those factors may be even more dangerous.”
• His advice to engineers today is direct: “You can be angry, but never leave in a rage … starting a company is a very serious matter.” The commitment involves years of time, investors’ money and employees’ expectations. “You need to have something that makes the company more likely to succeed.”
• His preparation was unusually practical. He made the decision in January but did not leave until June, using the intervening months to take night classes in finance, accounting, marketing and management at Foothill and De Anza community colleges. After leaving, he spent 5 months learning sales. Drawing on 2 papers in plasma chemistry from his master’s and PhD work, he identified plasma etching’s Achilles’ heel: “You can get a result in the morning, but you cannot reproduce it in the afternoon.” At 2 a.m. on Sundays, he rented the cheapest machine at a computing center to build financial models. “When I walked in the door in January 1980, I already had everything I needed.” The company launched a product within 1 year and went public in 4 years. Of the namesake company that later moved away from his original path, he says: “That is their work, not mine.”
• The book that shaped him was 克莱顿·克里斯坦森’s The Innovator’s Dilemma, published by Harvard Business School Press in 1997. He read it twice: the first time from the large-company perspective, to understand how not to be ambushed; the second time from the small-company perspective, to learn how to carve out a piece of the market. At a trade show, the CEO of a 5¼-inch floppy-drive company picked up a 3½-inch drive and said, “We can make it, but nobody wants it.” A departing executive built the smaller drive and caught the laptop wave. “The timing of a turning point is not always obvious, but you need to pay very close attention to it.”