Hyperscalers Are Trying To Replace Nvidia's GPUs - TSMC Gets The Upside Either Way

TSMC is benefiting from the AI chip buildout as Google’s custom AI chips add another major foundry order, reinforcing its central role across competing AI architectures. Management also raised its long-term gross margin floor to 56% in Q1, a constructive sign for profitability, though the article notes a potentially large dilution schedule.

Analysis

The key equity implication is that AI capex is becoming less of a winner-takes-all race and more of a toll-road for TSMC. If hyperscalers keep diversifying inference/training silicon away from Nvidia, the lost unit share at one customer is offset by incremental wafer demand from multiple custom-chip programs, which should make TSMC’s revenue mix more durable than either GOOGL- or NVDA-centric narratives imply. The second-order effect is that every major AI CPU/GPU/ASIC road map now competes for advanced-node capacity, so the bottleneck migrates from design wins to allocation power.

The market is probably still underestimating the asymmetry in margins versus volume. A higher long-term floor suggests pricing power is becoming embedded, but governance and dilution matter because the equity story can be capped even if operating performance stays strong. For holders, the right lens is not peak gross margin; it is whether incremental cash generation can outrun share count growth over a 12-24 month horizon.

For NVDA, this is not an immediate demand collapse story; it is a multi-year mix-risk story. Custom silicon from GOOGL and others erodes the “default” platform tax, but Nvidia still benefits from the ecosystem and software moat in the near term. AMD and INTC are more likely to gain optionality than near-term earnings, because any diversification away from Nvidia first shows up as more design engagements and better bargaining leverage before it shows up in revenue.

Consensus is likely overpricing the idea that all AI semis move together. The better contrarian trade is that the scarce asset is not AI demand but advanced packaging and leading-edge wafer capacity, which should keep TSMC structurally ahead even if the end-equipment narrative rotates. The main reversal risk is a capex pause or export-control shock that forces customers to delay tape-outs; that would hit second-half estimates first and sentiment fastest, but the core thesis remains intact on a 6-18 month view.

Key Decisions for Investors

• Long TSM on any 2-3% pullback; target a 3-6 month hold. Risk/reward favors continued multiple support as advanced-node allocation remains the binding constraint, but trim if dilution concerns start pressuring forward EPS revisions.
• Pair trade: long TSM / short NVDA for 3-6 months. Thesis: the market still over-credits Nvidia’s share of AI capex while underpricing the toll-collector economics at TSMC. Stop if NVDA reaccelerates on enterprise software monetization rather than hardware demand.
• Buy GOOGL vs. NVDA on a 6-12 month horizon. GOOGL’s custom silicon reduces inference cost and improves bargaining power, while NVDA absorbs the first-order share loss but not the full replacement volume. Best executed after any post-earnings NVDA bounce.
• Small tactical long AMD vs. INTC for 1-3 months. AMD benefits from any supply-chain diversification narrative; INTC remains a turnaround story with less near-term AI monetization. Keep sizing modest because this is more sentiment-driven than fundamentals-driven.
• Use TSMC upside calls rather than common if available. The catalyst path is orderly but durable, and options can capture rerating while limiting exposure to governance/dilution overhangs.