3 AI Chip Stocks to Buy as the Sell-Off Continues

AI chip stocks have cooled amid concerns that AI infrastructure spending could slow, but hyperscalers are signaling capex will rise next year. Bank of America projects worldwide cloud/AI data center capex will jump 40%–50% YoY to about $1.5T in 2027, which the piece frames as a buying opportunity. It highlights Nvidia at ~15x fiscal 2028 estimated earnings with 85% revenue growth last quarter, AMD levered to inference/agentic AI (data center CPU TAM seen growing to $120B by 2030), and Broadcom valued around ~19x fiscal 2027 estimates despite custom-chip revenue expected to exceed $100B next year.

Analysis

The selloff looks more like a positioning reset than a thesis break: if hyperscaler capex continues to inflect, the next leg is not just GPU upside but a broader mix shift toward networking, custom silicon, and CPU-heavy agent workloads. That creates a relative winner stack: AVGO benefits from the custom-chip and interconnect content per dollar of AI spend, while AMD has more operating leverage if inference demand broadens beyond one or two flagship customers. NVDA remains the quality anchor, but its upside is increasingly tied to sustaining share as buyers diversify architectures; that can cap multiple expansion even if absolute revenue stays strong.

The key second-order effect is margin dispersion inside the AI supply chain. As spend moves from training to inference/agentic deployments, the revenue mix should tilt away from the most monopoly-like SKU and toward lower-visibility platforms where pricing power is weaker and design wins matter more. Over 1-3 months, the market will trade on backlog commentary and hyperscaler budgets; over 6-18 months, the real question is whether custom ASICs and CPU content dilute NVDA’s share of wallet faster than total AI capex grows. A reversal would likely come from any capex pause, export restriction, or evidence that inference monetization is not scaling fast enough to justify the buildout.

Consensus may be missing that 'AI spend up' is not automatically bullish the same names in the same proportions. If the mix keeps moving to custom silicon, AVGO can compound faster with less narrative risk, while NVDA may deserve a lower peak multiple than in the training-only phase. The contrarian risk to being too bearish on the pullback is that large-cap semis rarely stay dislocated if bookings and guide still point higher; the better tell is whether 2027/2028 estimates keep rising, not just stock prices.

Key Decisions for Investors

• Long AVGO / short NVDA as a 3-6 month relative-value pair: thesis is mix shift toward custom silicon and networking. Target 8-12% spread capture if AVGO re-rates on design-win visibility; stop if NVDA reaccelerates share gains or AVGO custom-chip commentary disappoints.
• Buy AMD on weakness for a 6-12 month trade: higher beta to inference and agentic-AI CPU content gives more torque if hyperscaler orders convert. Use a staged entry after next earnings cycle; thesis breaks if shipment ramp slips or gross margin guidance is cut.
• Use SMH/XLK as a hedge overlay rather than an outright macro bet: if AI capex headlines remain positive but single-name dispersion widens, the ETF captures beta while pair trades express stock-specific differentiation.
• Set a catalyst watch on hyperscaler capex prints and supplier backlog over the next 1-2 quarters. If GOOGL/META commentary or order data imply any moderation, reduce long exposure quickly; if capex guides up again, the current pullback is likely buyable.
• Avoid chasing NVDA into the bounce until the market confirms estimate revisions are still moving up. The stock can work, but the risk/reward is worse than AVGO unless the next print shows renewed share gains and no sign of customer diversification.