1 Wrong Way to Think About the AI Boom Right Now

Nvidia, Broadcom, and Taiwan Semiconductor posted strong results, with sales up 73%, 29%, and 21% respectively and earnings growth of 82%, 28%, and 35%. The article argues AI hardware demand remains intact, citing major new deals including Broadcom’s long-term Google TPU agreement, Broadcom’s Anthropic-related capacity deal, and TSMC’s reported expansion of U.S. manufacturing. It also highlights Nvidia management’s forecast for 77% revenue growth in fiscal Q1 2027 to $78 billion, reinforcing the view that AI spending is still accelerating.

Analysis

The key market error is treating AI hardware as a monolith when the economics are actually bifurcated between compute demand and capacity bottlenecks. The near-term winners are the supply-constrained infrastructure layer — especially foundry, packaging, and custom silicon — because hyperscalers are now optimizing for supply assurance as much as performance. That favors TSMC and AVGO more durably than pure-play GPU sentiment, since both benefit from long-duration design-ins and switching costs that rise as AI deployments get embedded in production workflows.

What the market may be underestimating is that the second leg of the AI buildout is less about model hype and more about capex normalization across cloud, enterprise, and sovereign compute. If capex keeps compounding, the revenue durability for NVDA is strong, but the marginal upside increasingly depends on server/networking attach rates and cadence of next-gen product transitions rather than simple unit growth. Conversely, any slowdown in top-tier hyperscaler spending would hit NBIS and other data-center intermediaries first, because their valuation is more sensitive to financing conditions and customer concentration.

The contrarian issue is that consensus is anchoring to the right direction but the wrong duration: the trade is not about whether AI spending exists, but whether growth decelerates from explosive to merely strong over the next 2-4 quarters. That transition could compress multiples even if earnings keep rising, especially for names with the most crowded positioning. The market is also likely underpricing supply-chain capex as a constraint on free cash flow at TSMC and, to a lesser extent, the margin trade-off for AVGO if custom silicon demand broadens beyond a few anchor customers.

Key Decisions for Investors

• Stay long TSM and AVGO versus NVDA into the next 1-2 quarters; both have more durable pricing power from supply bottlenecks and customization, while NVDA is more exposed to any slowing in growth-rate optics.
• Buy NVDA on 5-10% pullbacks only, with a 3-6 month horizon; upside remains intact, but the risk/reward is worse after strong runs because multiple expansion is now more sensitive to guidance inflections than fundamentals.
• Short NBIS against a long basket of NVDA/TSM for a 1-3 month relative-value trade; if AI capex stays strong, the upstream semiconductor names should outperform the financing-sensitive data-center proxy.
• Use call spreads on AVGO or TSM rather than outright shares for a 3-6 month expression; the trade is to monetize continued AI capex without overpaying for terminal multiple risk.
• If hyperscaler capex commentary weakens, rotate quickly out of NBIS and META-adjacent infrastructure names first; those are the earliest canaries for a broader AI-spend deceleration.