1 Wrong Way to Think About the AI Boom Right Now

Nvidia, TSMC, and Broadcom reported strong AI-driven financial results, with revenue up 73% to $68.1B for Nvidia, 21% to $33.7B for TSMC, and 29% to $19.3B for Broadcom, alongside EPS growth of 82%, 35%, and 28%, respectively. The article argues the AI hardware boom remains intact, citing major new deals such as Broadcom's TPU partnership with Google through 2031 and Nvidia-linked demand at Nebius. Sentiment is supportive of AI infrastructure stocks, though the piece is more argumentative than a direct market-moving earnings release.

Analysis

The market is still pricing AI as a single-factor trade, but the more important dynamic is that the capital intensity is shifting from experimentation to industrialization. That favors the platform owners with scarce manufacturing/design bottlenecks and pricing power, while compressing returns for adjacent names that depend on the same capex wave but lack control over the stack. In practice, the winners are not just the chip designers; they are the entities that can turn wafer capacity, packaging, and custom silicon relationships into multi-year contracting leverage.

The second-order effect is that supply-chain scarcity remains the hidden bull case. When end users lock capacity years ahead, they are effectively de-risking utilization for foundries and preserving gross margin resilience even if headline AI sentiment cools. That also makes the ecosystem less cyclical than consensus assumes: order visibility extends beyond one budget cycle, so any pullback is more likely to be a sentiment air pocket over 1-2 quarters than a structural demand collapse.

The key risk is not that AI demand disappears, but that market expectations outrun near-term monetization and capex duration. If hyperscaler spending moderates or shifts from GPU-heavy builds to more efficient architectures, the most leveraged beneficiaries can de-rate even while revenue stays strong. The setup is therefore asymmetric: fundamentals can keep compounding for months, but the stocks may become vulnerable to any guidance miss, inventory digestion, or evidence that custom silicon is cannibalizing merchant accelerator growth.

Consensus is missing that this is becoming a procurement war, not a narrative trade. The clearest contrarian read is to favor the names with contractual visibility and bottleneck control over the names most exposed to cyclical multiple compression. The overdone part is the blanket 'AI bubble' call; the underdone part is distinguishing between the infrastructure toll collectors and the broader basket of AI-adjacent exposure.

Key Decisions for Investors

• Long NVDA / short a basket of weaker AI-adjacent hardware laggards over the next 1-3 months: the setup favors the highest-margin toll collector if AI spend remains intact, while limiting downside if the theme cools.
• Add to TSM on pullbacks over the next 2-6 weeks; use it as the lower-beta way to own persistent AI capex. Risk/reward is attractive as long as advanced-node demand and foundry utilization stay tight.
• Maintain AVGO as the preferred multi-month long in AI infrastructure, but size it below NVDA/TSM due to higher risk of custom-silicon share shifts. Best expressed via call spreads into earnings/guidance updates.
• Pair long GOOGL against a weaker AI-infrastructure proxy if cloud/custom silicon spend re-accelerates: Alphabet benefits from internal TPU leverage, but the better trade is owning the platform while avoiding pure hardware duration risk.
• Avoid chasing NBIS as a standalone long after headline deal news; treat it as a high-volatility financing/execution story with poor risk/reward unless there is proof of sustained contract conversion.