Strong ASML, TSMC forecasts signal AI spending boom is intact

ASML and TSMC both raised annual revenue outlooks, signaling that AI chip demand remains strong and that cloud providers are still on track to spend heavily on data centers this year, with industry spending expected to exceed $600 billion. TSMC said it is increasing capital spending to meet demand, while ASML said demand is set to outstrip supply for the foreseeable future. The article is bullish for AI chip suppliers such as Nvidia, AMD, and Broadcom, but also highlights persistent capacity constraints across the semiconductor supply chain.

Analysis

The key implication is that the AI capex cycle is becoming less about experimentation and more about industrialized capacity reservation. That shifts bargaining power toward the bottleneck layer — foundry, lithography, and advanced packaging — because multi-year supply commitments typically compress optionality for buyers and improve visibility for suppliers. In the near term, that should support the entire semi value chain, but the cleaner second-order winner is TSM/ASML, not the headline AI compute names, because their revenue is tied to capacity expansion rather than end-demand monetization.

The market is likely underestimating how quickly inference demand can re-rate the mix. Training gets the headlines, but inference is where utilization compounds, which tends to favor systems integrators and custom accelerators over pure GPU unit growth over a 6-18 month horizon. That creates a subtle risk for NVDA/AMD/AVGO: demand can remain strong while pricing power migrates toward buyers that can lock in supply or design around generic accelerators, capping upside multiples even as top-line stays robust.

The main reversal catalyst is not a collapse in AI interest; it is a delay in monetization evidence from hyperscalers, likely around the next 1-2 earnings cycles. If capex continues rising faster than revenue disclosures, investors may start demanding proof of ROI, which could compress sentiment in MSFT/META/AMZN/GOOGL before it ever hits chip orders. On the other hand, if these platforms reaffirm multi-quarter spend plans on April 29, the trade becomes a duration extension rather than a one-quarter pop.

Contrarianly, the setup argues for owning the bottleneck and fading the most crowded beneficiaries of AI optimism. The asymmetry is best where supply is constrained and contractual, not where expectations for product monetization are already elevated. In that sense, the risk/reward is better in ASML/TSM than in the more consensus long AI basket.

Key Decisions for Investors

• Long TSM / short a basket of NVDA-AMD-AVGO into the next 2-6 weeks: thesis is that capacity owners capture the next leg of the cycle while chip designers face valuation compression if demand shifts toward inference and custom silicon.
• Add to ASML on any post-guidance pullback, 3-12 month horizon: supply scarcity and multi-year tool demand should support estimates even if end-market enthusiasm wobbles; use weakness as a better entry than chasing strength.
• Buy July-August call spreads on TSM or ASML rather than outright stock: the market may rerate the supply bottleneck after hyperscaler earnings, but call spreads better balance upside with risk if capex rhetoric fades.
• Hedge long AI exposure with a short in the hyperscaler group into April 29 earnings: if capex remains elevated without matching ROI language, MSFT/META/AMZN/GOOGL are the names most likely to face margin and sentiment pressure.
• If already long NVDA, reduce outright exposure and replace part of it with a pair against a semiconductor equipment laggard only if the latter has weaker exposure to AI fab spend; otherwise, the cleaner trade is simply trimming beta and keeping exposure via suppliers with contracted demand.