This 40%-a-Year Fund Manager Says Semis Are STILL the Place to Be After a 100% Run

T. Rowe Price’s Dom Rizzo says the compute layer remains the “bottleneck,” citing semiconductors up ~100% while software is down ~14% and pointing to strong YTD moves (AMD +142%, Intel +222%, Micron +227%). He expects ~75% (possibly >80%) hyperscaler capex growth next year, with NVIDIA reporting April-quarter non-GAAP EPS of $1.87 on $81.61B revenue, data center +92% and networking +199%, Broadcom guiding next-quarter AI semiconductor revenue to $16B (+200% YoY), and Micron Q3 revenue up 345.7% YoY with gross margins expanding to 84.6%. The main near-term risk is optical, after a one-day selloff (AMD -5.7%, Micron -9.74%, TSMC -6.8%), but the article frames a valuation-supported, cycle-driven bullish stance into the “Vera Rubin”/Rubin-era earnings power.

Analysis

The clean read-through is breadth, not just beta: if AI demand is shifting toward inference and agentic workloads, the scarce assets are memory, packaging, networking, and leading-edge foundry capacity. That makes MU, AVGO, NVDA, and TSM the closest toll collectors, while application software faces budget leakage as customers redirect spend to infrastructure with faster ROI. AMD can participate, but it is still more of a share-gain story than a primary bottleneck beneficiary.

The consensus still looks too anchored to a software-led AI narrative. Relative value should keep favoring semis over IGV/XLK software baskets, and the market may be underpricing how much sustained capex can support pricing power in memory and networking even if headline chip multiples stay rich. The less obvious risk is that INTC becomes a late-cycle beneficiary only if enterprise CPU demand stays tight; without execution, it remains a cycle trade, not a thesis.

Near term, the setup is vulnerable to sharp de-risking after any hyperscaler capex miss or memory-price wobble, which would hit high-beta names within days. The 1-3 month catalyst is the next round of cloud guidance; the 6-18 month risk is that agentic usage does not monetize fast enough and capex normalizes. A sustained rise in DRAM pricing or another upside revision to capex would validate the thesis; a drop in cloud capex growth below roughly 50% y/y would falsify it.

Key Decisions for Investors

• Long MU / short IGV for 1-3 months: best expression of infrastructure spend crowding out software budgets; target a widening relative-performance gap into the next hyperscaler earnings cycle.
• Long AVGO / short NVDA as a valuation-aware pair trade: if capex broadens beyond GPUs into networking and custom silicon, AVGO should compound with less multiple risk than NVDA; cut if NVDA keeps re-accelerating gross margin and guide.
• Buy SOXX on weakness, not strength, with a 4-8 week horizon: the setup favors basket exposure to the hardware cycle, but entry should wait for post-selloff stabilization after hyperscaler commentary.
• Set a watch item on TROW rather than a position: only consider a long if active tech-fund flows inflect positively for 2 consecutive quarters; otherwise the performance halo is too indirect to matter.