Why it’s too early to call a top on semiconductor stocks, according to these highly regarded analysts

Nomura argues it’s too early to call a top in semiconductor stocks after SOX delivered an 80%+ gain in Q2. The firm cites sustained “go-big-or-go-home” hyperscaler spending through 2027, supported by rising memory-chip costs, plus ramping data-center build plans as key continuation drivers. Overall, the note is supportive of the chip-stock rally but is unlikely to be a major immediate market mover.

Analysis

The more durable implication is not "semis up," but a widening earnings gap inside the group. The names with direct exposure to AI infrastructure bottlenecks — leading-edge foundry capacity, advanced packaging, memory, and power delivery — should keep seeing estimate upgrades, while broad analog/industrial semi exposure is more vulnerable to being left behind as investors pay up for the narrow set of suppliers that actually sit on the critical path.

The second-order effect is margin pressure elsewhere in tech: hyperscalers can keep capex elevated, but that does not mean the economics stay benign. Rising memory content and data-center build intensity tend to leak into cloud gross margins before they show up in earnings guidance, so the near-term market reaction can stay constructive even as the 6-18 month setup becomes more selective. That creates a cleaner relative-value trade than a blanket index short: the cycle can remain healthy while breadth deteriorates.

The contrarian miss is that a lot of the bullish case is already in price at the index level. If the next few earnings cycles do not validate another step-up in capex, the multiple on the whole complex can compress quickly even if revenue keeps growing. The key falsifiers are any downward revision to hyperscaler capex plans, a rollover in memory pricing, or evidence that inventory is rebuilding in non-AI end markets — any of those would shift this from a momentum continuation into a narrow leadership trade.

Key Decisions for Investors

• Buy SMH or SOXX on pullbacks over the next 2-4 weeks rather than chasing strength; use it as liquid beta to a capex-durable cycle, but only if it holds the prior breakout range. If hyperscaler capex commentary softens, cut quickly — downside would likely be a sharp factor de-rating, not a slow grind.
• Relative-value: long MU vs. short ON into the next earnings window. MU has the cleanest direct leverage to memory pricing and AI storage intensity, while ON is more exposed to auto/industrial cyclicals and is less likely to participate if the rally narrows to AI bottlenecks.
• Add AMAT/LRCX on any 5-7% pullback for a 3-6 month horizon; these are the cleaner second-order beneficiaries of sustained data-center build plans. Falsifier: if order commentary or backlog commentary stops improving despite strong chip ASPs, the market will start discounting a future capex air pocket.
• For investors already long the space, hedge index exposure with a short ON or a partial call-spread structure on SMH to keep upside but limit damage if the move becomes too crowded. The risk/reward is better in relative positioning than in outright chasing after an 80% quarter.