Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon

Episode Summary

Executive Summary: Stacey Raskin argues semiconductors are not near a cycle peak despite stock pullbacks and multiple compression. Earnings, order visibility, AI infrastructure spending, and memory/storage tightness all point to stronger fundamentals. He sees demand as real, constrained by power and clean-room capacity, and favors AI leaders, semicap equipment, and AMD.

Main Topics: Semiconductor valuation vs. fundamentals (Priority: 5/5): The sector’s earnings have surged while multiples compressed, which Raskin says can happen near cycle peaks—but he believes current fundamentals are still improving rather than deteriorating. AI demand and hyperscaler spending (Priority: 5/5): He argues AI infrastructure demand is authentic, with hyperscalers, model companies, and enterprise customers continuing to spend aggressively and seeing returns. Bottlenecks, capacity constraints, and cycle risk (Priority: 5/5): The conversation centers on whether current shortages are real demand or double ordering; Raskin says tight supply, power limits, and buildout constraints are slowing supply, not demand. Memory market squeeze: DRAM, HBM, NAND, HDD (Priority: 4/5): He explains why memory is especially tight: HBM consumes much more wafer capacity, NAND supply is constrained, and storage demand extends even into hard drives and possibly tape. Semicap outlook and WFE growth (Priority: 5/5): Raskin is bullish on wafer fab equipment because more AI data centers require more fabs, clean rooms, and tools; he sees equipment spending rising strongly through 2028 and beyond. NVIDIA, Broadcom, AMD, and market-share fears (Priority: 4/5): He downplays competitive threats as overblown, arguing the real question is whether AI remains a huge market; if it does, multiple players can win. Intel turnaround and relative improvement (Priority: 3/5): He says Intel remains a long shot, but he is more positive than in years because of demand strength, better execution, and a more realistic management approach.

Key Arguments: Semis may look like a peak because earnings multiples compress when investors fear the cycle is maturing, but Raskin says earnings and forward commentary still point up. Management teams do not have perfect foresight; they only see orders in front of them, so guidance is imperfect but still currently strong. AI infrastructure spending appears real because customers are deploying capital only when they expect a return, and early evidence suggests payback is plausible. Buildout constraints like land, power, and clean-room availability are limiting near-term supply, which may actually strengthen the cycle by preventing overexpansion. Memory is exceptionally tight because HBM uses several times more wafers per gigabyte than conventional DRAM, while NAND and HDD demand are also rising. The market-share debate in AI semis is less important than the size of the market; multiple winners can coexist if AI demand keeps expanding. Semicap is attractive because higher AI demand translates into more fabs, more clean rooms, and ultimately more WFE spending. Intel is improving mainly due to strong server demand and a better CEO/operating story, but foundry success remains years away.

Data Points: Semiconductor sector earnings growth: About 100%+ - Earnings in the semiconductor sector are described as roughly doubling year over year. Semiconductor sector stock performance: Up roughly 80% year-to-date - The sector index is still far above the start of the year despite recent pullbacks. NVIDIA and Broadcom expected revenue growth: 70% to 100%+ next year - Raskin says both companies guided to extraordinary growth, with Broadcom implying possible 100% growth again into 2028. Top five hyperscaler spending: About $1.5 trillion next year - Used to illustrate the magnitude of AI capex already being deployed. AI infrastructure spending outlook: $3 trillion to $4 trillion annually by end of decade - A Jensen Huang forecast Raskin says once sounded crazy but now looks plausible. NVIDIA stock price: About $240 - Used to illustrate valuation despite very high growth expectations. Broadcom stock price: About $363 - Used to discuss implied valuation versus projected earnings growth. WFE spending this year: About $150B+ - Raskin’s semiconductor equipment forecast for the current year. WFE spending next year: About $204B - His published forecast for the following year. WFE spending in 2028: About $250B+ - Forecast for continued strong equipment demand. Potential WFE by end of decade: $300B+ - He thinks this level is possible if clean-room constraints ease. Memory cycle timing: Worst memory cycle since the tech bubble 18–24 months ago - Explains why current memory profits and margins are recovering from a very weak base. HBM wafer intensity: 3–4x as many wafers per gigabyte as standard DRAM - Key reason HBM is consuming capacity and tightening the market. NVIDIA NVL72 content mix: ~80% of wafer area is memory - Illustrates how AI racks are heavily memory-intensive. Semicap stocks covered: AMAT, LAM, KLA - Raskin says these are among the names he follows, with AMAT slightly preferred. NVIDIA revenue guide: 70% growth - He says NVIDIA guided conservatively because of land/power/shell constraints. Broadcom AI revenue: $115B next year - A figure Raskin cites as a rough expectation in the discussion.

Pivotal Quotes: "At this point, fundamentals seem better or not worse." — Stacey Raskin: Summarizing why he is not yet worried about a semiconductor cycle peak. "Nobody has enough compute." — Stacey Raskin: Describing the dominant theme from hyperscalers and AI customers across the supply chain. "I do not think that there are like huge warehouses all over the place filled with GPUs that are just lying on the floor." — Stacey Raskin: Rejecting the idea that AI chips are being stockpiled without being used.

Implications: For investors, the message is that AI-linked semis may still have room to run, but supply constraints and sentiment swings can cause volatility. Semicap and memory remain key beneficiaries if AI demand stays real.