MarketWatch flags that some of the most aggressive AI hardware trades — particularly optical interconnects and high-bandwidth memory — are rolling over, with at least one strategist warning that AI spending, a core engine of the 2023-2024 bull run, may be approaching a stall point. The concern is not that AI investment stops, but that the pace of incremental hardware ordering could slow enough to pressure the supply chain names that ran hardest on build-out expectations.

The three names most directly in the crosshairs are NVDA, AMD, and AVGO. NVDA remains the dominant AI chip franchise with $215.9B in revenue and a staggering 71.1% gross margin, but those numbers also set an impossibly high bar to beat or even match. AMD posted 34.3% revenue growth to $34.6B but carries a thin 12.5% net margin, leaving little cushion if data center orders soften. AVGO, with $63.9B in revenue and 67.8% gross margins, is more diversified into custom ASICs and networking, which may insulate it partially.

The core bear tension is valuation: all three trade at elevated multiples priced for sustained hypergrowth in AI capex. If optical and memory components — which are earlier in the AI build-out chain — are already stalling, that signal could propagate upstream to GPU and ASIC demand within one to two quarters. The bull counter is that hyperscaler capex guidance (Meta, Microsoft, Google, Amazon) remains publicly committed to multi-year AI infrastructure ramps, and any pause in optical/memory could simply be inventory digestion rather than demand destruction.

What to watch: hyperscaler earnings call commentary on capex pacing, any revision to NVDA's H2 2025 order book guidance, and whether the optical/memory weakness is confirmed by names like MRVL, MU, or COHR in their next prints. The NVDA earnings date is the clearest catalyst to resolve the bull/bear tension here.