Evercore finds strong AI networking demand in latest channel checks By Investing.com

Evercore’s AI networking checks indicate scale-out demand remains strong but constrained by supply, supporting Outperform ratings on Nvidia, Marvell, Broadcom and Macom. The note highlights Nvidia’s NVLink and upcoming CPO adoption, Broadcom’s leadership in 1.6T DSP and scale-up CPO, and Marvell’s improving 1.6T share outlook, while Astera Labs and Macom see mixed but generally constructive demand trends. The article is analyst-driven rather than event-driven, so the likely market impact is modest and mostly stock-specific.

Analysis

The key takeaway is not simply that AI networking demand is healthy, but that the bottleneck is migrating from compute silicon to interconnect architecture. That shifts bargaining power toward the suppliers that control the “plumbing” of AI clusters: the companies with credible paths to scale-up fabrics, co-packaged optics, and high-speed DSPs should capture a larger share of the next leg of capex even if GPU unit growth normalizes. In practice, that favors names with multiple technology touchpoints and penalizes point-solution vendors whose roadmap is more exposed to a single standard winning.

The second-order effect is that supply-constrained demand often looks bullish for everyone until customers begin dual-sourcing or reprioritizing architectures to preserve deployment schedules. That creates a subtle winner’s curse: the most “proven” vendor can actually accelerate ecosystem commoditization as hyperscalers push custom ASICs and alternative protocols to reduce dependence on a single scale-up stack. For Marvell and Broadcom, the market may be underestimating how much share can still be won through design-win persistence in 1.6T and CPO, even if one product line loses mindshare.

The more interesting contrarian read is that near-term enthusiasm may be overweighting standards adoption speed. The 2027-28 CPO and next-gen networking timelines leave a long gap where execution risk, reliability qualification, and customer migration friction can delay monetization by quarters, not years. That argues for expressing the bullish view through leaders with near-term DSP and switch exposure, while being more selective on newer entrants whose upside depends on a clean protocol or packaging inflection that may arrive later than consensus expects.

Key Decisions for Investors

• Overweight NVDA vs the AI networking basket for the next 6-12 months: best path to monetizing interconnect scarcity, with lower execution risk than the younger suppliers; add on dips rather than chasing strength.
• Pair long AVGO / short a broad semiconductor ETF for 3-6 months: Broadcom has the cleanest exposure to 1.6T and scale-up CPO content, while the short hedges any sector-wide de-rating if AI capex expectations get marked down.
• Maintain a tactical long MTSI position for 12-18 months only if buying weakness: the upside is share gain from 1.6T migration, but the risk/reward is narrower because it is more exposed to standards and customer timing slippage.
• Avoid chasing ALAB on strength; use it as a relative-value long only against a faster-growing but higher-beta networking competitor if you want exposure to PCIe scale-up adoption through 2027.
• Watch for a 2026-2027 correction opportunity in the group if hyperscaler capex pauses: the best entry will likely be on a reliability/qualification scare, not on clean fundamental data, because that is where premium compression tends to overshoot.