Nvidia Started the AI Boom. These 2 Stocks Could Power the Next Phase.

Broadcom and Marvell are described as controlling roughly 95% of the custom AI ASIC market, with hyperscaler demand for custom silicon accelerating as custom ASICs are projected to reach 27.8% of AI server compute in 2026, up 44.6% year over year. Broadcom disclosed long-term TPU and networking supply agreements through 2031, while Marvell sees up to $11 billion in 2026 AI ASIC revenue and a pipeline of 50+ opportunities worth an estimated $75 billion in lifetime revenue. The article is broadly bullish on both stocks, despite Broadcom’s 15% post-earnings pullback and Marvell’s sharp 33% one-day jump after Jensen Huang’s comments.

Analysis

The market is still valuing AI infrastructure as if compute were the only bottleneck, but the next marginal dollar is shifting into orchestration: chip design, rack-level networking, and supply assurance contracts. That benefits AVGO and MRVL more than the pure-play GPU complex because they monetize the “picks-and-shovels of customization” rather than a single silicon SKU, which should make their revenue streams stickier and less cyclical than headline AI spend implies.

The second-order effect is that hyperscalers’ ASIC push is not just a cost-saving initiative; it is a bargaining strategy against Nvidia’s pricing power. If custom accelerators continue taking share, the real loser is not just NVDA unit growth, but the attach rate on premium interconnect, software optimization, and platform lock-in over the next 12-24 months. At the same time, the OEM and foundry ecosystem should see a deeper mix shift toward advanced packaging, custom networking, and long-duration capacity commitments, which tightens supply for slower-moving second-tier AI hardware vendors.

MRVL looks more interesting tactically because the market is pricing the ramp as optionality rather than durability. The asymmetry is that each incremental design win compounds into multi-year revenue visibility, while margins can expand sharply once the mix shifts from early-stage engineering programs to production silicon; that makes the next 2-4 quarters the key re-rating window. AVGO is the lower-volatility expression, but the recent drawdown creates a better entry point if investors want exposure to the AI capex stack with less execution dispersion.

Contrarianly, the consensus may be overestimating how quickly hyperscalers can fully displace merchant GPUs: custom silicon adoption usually expands total AI capex before it compresses it, because every new generation lowers unit cost and expands workload ambition. The better question is not whether ASICs win, but who captures the design, networking, and deployment margin along the way. On that basis, the trade is less bearish NVDA outright and more bullish the enablers of hybrid AI architectures.

Key Decisions for Investors

• Long MRVL on 3-6 month horizon into earnings/ramp updates; use a staggered entry after any post-rally consolidation. Risk/reward favors a re-rating if management can confirm design-win conversion, with downside capped by the stock already repricing some optimism.
• Buy AVGO on weakness as a lower-beta compounder versus the broader AI basket. Prefer shares or a call spread over outright calls, since the thesis is durable but the stock may grind rather than gap.
• Pair trade: long MRVL / short a basket of high-multiple AI hardware names most exposed to merchant GPU share loss. This isolates the custom-silicon win without paying full market beta for the entire AI stack.
• For traders seeking event risk, use 3-6 month AVGO or MRVL call spreads around next guidance cycle. The catalyst is not a single quarter, but evidence that pipeline converts into revenue and margin acceleration faster than consensus expects.
• Avoid chasing NVDA on the headline; if anything, use it as a relative-value short against custom-silicon beneficiaries only if channel checks show accelerating ASIC substitution over the next 2-3 quarters.