2 Artificial Intelligence (AI) Stocks to Buy as Demand for Custom Chips Soars

Broadcom reported Q2 fiscal 2026 revenue of $22.2 billion, up 48% year over year, with AI chip sales surging 143% to $10.8 billion and adjusted EPS rising 54% to $2.44. Marvell posted record Q1 fiscal 2027 revenue of $2.4 billion, up 28%, with adjusted EPS up 29% to $0.80 and data center revenue representing 76% of sales. The article argues rising hyperscaler capex and demand for custom AI ASICs should continue to benefit both companies, though Broadcom remains the larger beneficiary today.

Analysis

The trade is not simply “AI chips up”; it is a re-rating of the custom silicon stack as hyperscalers try to lower unit compute cost and reduce dependence on merchant GPUs. That creates a second-order winner set beyond the named chip designers: advanced packaging, HBM, substrate, and EDA/tooling vendors should see stickier demand because custom ASIC deployments are more design-intensive and less easily swapped than accelerator purchases. The key nuance is that custom silicon shifts budget from one large GPU order to a longer-dated engineering funnel, which tends to smooth revenue for the suppliers that win sockets but raises execution risk for everyone else.

Broadcom’s edge is not just scale; it is customer concentration with the best-funded buyers and a longer visibility window into demand. If hyperscalers keep prioritizing cost-per-token economics, Broadcom can leverage existing designs into multi-year follow-on orders, which should support margin expansion even if growth moderates from current peak rates. Marvell is more cyclical because it is earlier in the adoption curve and more exposed to incremental wins/losses; that makes it higher beta to AI capex surprises, but also more vulnerable if hyperscalers rationalize spending or delay new tape-outs.

The consensus seems to be underestimating how much custom silicon can cannibalize adjacent GPU spend rather than merely complement it. That is constructive for the winners here, but negative for the “all AI spend is additive” narrative around Nvidia: if inference economics keep improving, hyperscalers will reallocate more of their capex mix toward ASICs and networking, not necessarily expand total spend at the same pace. The tail risk is that the market is capitalizing a sustained super-cycle while the actual driver is a few large customers pulling forward design budgets; any capex digestion phase in the next 6-12 months would hit Marvell first and could also temper Broadcom’s multiple.

Key Decisions for Investors

• Long AVGO vs. NVDA over the next 6-12 months: express the view that custom silicon share gains matter more than broad accelerator growth; target a relative outperformance trade with tighter downside if hyperscaler capex decelerates.
• Buy MRVL on pullbacks for a 6-9 month horizon, but size smaller than AVGO: higher beta to incremental AI capex upside, with the risk that it underperforms if design-win conversion slows or customer concentration bites.
• Pair long AVGO / short a basket of less differentiated AI compute exposure for 3-6 months: use the pair to isolate custom ASIC adoption while reducing market beta and multiple compression risk.
• Add exposure to AI infrastructure enablers on any weakness, especially networking, packaging, and memory supply-chain names, because custom chips increase system-level complexity even when they reduce GPU spend.
• Use options rather than outright longs if entering after a post-earnings rally: AVGO call spreads and MRVL call spreads can capture another leg of AI capex revision while limiting drawdown if the market starts pricing in a spending digestion phase.