High-Speed Chips and Slow-Moving Capital: Why Goldman’s Semi Equipment Thesis Needs a Battle-Tested Filter

The order book doesn’t lie, but interpretation often does. Goldman Sachs upgrades Lasertec, Tokyo Electron, and Disco on Intel’s 2026 capex hike. Let’s cut through the narrative. The code doesn’t care about analyst ratings. Goldman’s Monday note was clean: Intel raises 2026 capital expenditure by $3 billion, primarily for 18A and 14A nodes, plus advanced packaging EMIB-T. The three Japanese equipment makers — Lasertec (EUV photomask inspection), Tokyo Electron (coat/develop, etch/dep), and Disco (precision dicing/grinding) — are structural beneficiaries. The logic is straightforward: Intel spends more, they book more orders. The market nodded, ticking each stock up 2–4%. But I’ve been in this arena long enough to know that a rising capex tide doesn’t lift all boats equally. I used to audit ICO smart contracts in 2017, reverse-engineering bonding curves to find integer overflows before the token launch. That experience taught me that technical monopoly is the only durable edge — everything else is liquidity chasing hype. Let’s look at the order flow. Lasertec holds roughly 85% of the EUV photomask inspection market. There is no second source. If Intel’s High-NA EUV tools arrive (ASML’s first unit is already at Intel’s Hillsboro facility), Lasertec’s inspection gear is mandatory. The company’s gross margins sit around 45–55%, driven by monopoly pricing. Disco commands 50–80% in the dicing/grinding segment, especially for chiplet and HBM applications. That EMIB-T bridge Intel is pushing? Disco’s ultra-thin wafer saws are the bottleneck. Both have moats. Tokyo Electron is different. TEL is strong in coat/develop (~50% share) and competitive in etch/dep (~25–30%), but it faces constant pressure from Applied Materials and Lam Research. Intel’s internal procurement has historically favored U.S. vendors for political reasons. The CHIPS Act grants come with strings — “buy American” clauses are already creeping into subsidy guidelines. TEL’s order visibility relies on Intel’s willingness to stay diversified, which is not guaranteed. Volatility is just interest for the impatient. The market is pricing in a smooth ramp. It assumes Intel’s 18A node hits volume production by late 2025 with acceptable yield. But Intel’s track record on execution is littered with delays. The 7nm (now Intel 4) was years late. If 18A slips or yields disappoint, Intel will either stretch its capex timeline or redirect funds to fix process issues, neither of which helps equipment orders immediately. Worse, Intel’s foundry business (IFS) is still bleeding cash. The $3 billion increment is a drop in a $250–280 billion annual capex pool — split among dozens of suppliers, the actual benefit to each Japanese firm is modest. Floor sweeps happen; rug pulls are a choice. This is not a rug pull, but it’s also not a risk-free arbitrage. The contrarian angle is that Goldman’s thesis conflates “Intel spending more” with “Japanese equipment makers winning more.” The real driver is structural demand for advanced packaging and EUV-related inspection, which exists regardless of Intel’s internal execution. Disco and Lasertec benefit from AI chiplet trends (Nvidia, AMD, Google) and the shift to High-NA EUV across the entire industry. Intel is just one customer. TEL’s success is much more tied to Intel’s own fabrication volume. Liquidity is a river, not a pond. The smart money is already rotating: Lasertec’s P/E is ~45x, Disco’s ~50x. Compare to TEL at ~22x. The market is paying a premium for monopoly. That premium can compress fast if Intel disappoints or if U.S. regulators force Intel to allocate more spending to American toolmakers. Based on my 2020 DeFi arbitrage experience — where I captured 340% by exploiting spread inefficiencies, then lost 20% of profits to exchange counterparty risk — I know that the best setups are the ones where the downside is visible and manageable. Here, the downside is that Intel’s capex narrative fades, and the stocks revert to their beta to semi cycle. Takeaway: The most actionable position is to favor Disco and Lasertec over TEL. Better moats, less Intel-dependency. The catalyst is not Intel’s capex but the broad AI-driven demand for advanced packaging and High-NA EUV. The risk is overpaying for growth that may not materialize on Intel’s timeline. I’d wait for a pullback in Lasertec and Disco, then build a position with a 12–18 month horizon. TEL is a hold until Intel’s 18A yield data emerges. Code doesn’t lie — and neither should your order book.