Chip Stocks Are Tumbling. Is It Time to Sell Cerebras, Broadcom, and Nvidia?

Cerebras has fallen about 18% since its May 15 IPO despite strong AI adoption signals, including OpenAI’s planned $20 billion spend and AWS integration, but its valuation remains steep at 97x sales. Broadcom reported Q2 AI semiconductor revenue up 143% to $10.8 billion and net income up 88% to $9.3 billion, though the stock sold off after management kept its $100 billion AI chip sales outlook unchanged and total revenue came in slightly below consensus. Nvidia remains the dominant AI chip name with 88% data center GPU share and a relatively low 30x P/E, even as investors worry about competition and AI spending durability.

Analysis

The market is increasingly sorting AI hardware into three buckets: platform leader, custom-chip enabler, and speculative architecture bet. That matters because the next leg of spend is likely to migrate from blanket accelerator purchases toward workload-specific deployments, which structurally favors vendors with software lock-in and high switching costs over pure hardware narratives. In that regime, the real competitive pressure is not just between NVDA and the rest; it is between premium multiple durability and the pace at which hyperscalers internalize more silicon design in-house.

Broadcom is the cleanest expression of that shift because its customer base is essentially buying engineering capacity as much as chips. The recent de-rating looks more like a multiple reset than a thesis break, and that is important: if AI capex growth slows from hypergrowth to merely strong growth, names with already-embedded expectations can compress another 15-20%, but Broadcom's cash generation gives it a higher floor than most peers. The hidden risk is customer concentration — if even one hyperscaler delays a platform transition, near-term revenue can wobble despite strong annual guidance.

Cerebras is the opposite: the market is paying for optionality before the business model has proven repeatability. The second-order effect is that any success there could pressure inference economics across the stack, especially for cloud providers trying to lower unit costs on deployed models, but that upside is still highly path-dependent and likely to show up over years, not quarters. The near-term risk is that the stock can remain disconnected from fundamentals because the float is small and narrative-driven flows can dominate price discovery for months.

Key Decisions for Investors

• Overweight AVGO vs NVDA on a 3-6 month horizon via a pair trade; long AVGO / short NVDA in roughly dollar-neutral size. Thesis: AVGO has already repriced more of the upside risk while retaining direct AI exposure and stronger cash flow visibility; downside is that NVDA keeps compounding faster if AI capex re-accelerates.
• Buy a starter long AVGO position on further weakness over the next 1-2 weeks; target 12-18 month hold. Risk/reward improves if the stock stays 10-15% below recent highs, because the multiple reset gives you a better entry on the highest-quality custom-silicon name.
• Avoid chasing CBRS after IPO volatility; if you want exposure, use a small options-defined position only. Structure: call spread 6-12 months out rather than common equity, since the binary adoption risk is high and the valuation leaves little room for execution misses.
• Hold NVDA as a core position, but consider trimming incremental adds until the next capex read-through from hyperscalers. If cloud spend data decelerates for 1-2 quarters, NVDA is the most vulnerable to another leg of multiple compression.
• For a relative-value hedge, pair long AVGO with short AMD over the next quarter. Rationale: the article's regime favors established custom/ASIC monetization, while AMD still has to prove it can convert AI interest into durable share gains.