Nvidia mostly sat out the chip sector's best quarter ever. What needs to change?

Semiconductor stocks are up strongly (SOX +80% in Q2) and memory makers led, but Nvidia lagged with only ~+12% for Apr–Jun despite accelerating data-center growth. In its May 20 report, Nvidia’s data-center revenue rose 92% YoY in the April quarter to $75.2B and guidance for May–July beat Street expectations, yet the stock is weighed by intensifying competition (AMD and Big Tech designing competing AI chips). Jim Cramer argues Nvidia should “open its checkbook” via larger buybacks, citing $13.65B R&D (+45% YoY) and projected FCF of $200B+ this fiscal year, while noting Nvidia has spent $40B on buybacks last fiscal year and $33.7B in the year ended Jan 2025, with an additional $80B repurchase authorization approved in May.

Analysis

The equity market is rotating from the original “GPU scarcity” trade into a broader AI supply chain where CPU content, memory, and tool intensity matter more. That usually favors AMD/ARM/INTC/TXN/AMAT/LRCX on a 1-3 month horizon because the market will pay up for the names with the cleanest incremental bottlenecks and the fastest operating leverage. The second-order effect is that hyperscaler custom silicon is no longer just a cost-control story; it is a margin-cap story for the incumbent platform leader if buyers can credibly substitute even part of their workload mix.

NVDA’s issue is mostly reflexivity: fundamentals remain strong, but the stock has become a funding source for newer AI exposures and is now fighting both crowded ownership and a lower implied scarcity premium. Buybacks can stabilize EPS and help absorb supply, but they do not fully offset a narrative shift if revenue growth normalizes while peers are still inflecting. In other words, capital returns are a support beam, not a growth substitute; the real catalyst is either a materially larger repurchase cadence or a re-acceleration in data-center growth relative to the rest of semis.

Contrarian view: the market may be overreading competitive threats and underestimating that Nvidia’s ecosystem still monetizes the highest-value layer of AI spend. If custom chips mostly expand total inference volume rather than displace training demand, NVDA can keep growing even as share gains migrate to adjacent names. The cleaner trade is relative value, not outright bearishness: long the beneficiaries of CPU/content expansion and AI equipment intensity versus a leader whose multiple is already doing less work than its earnings base.

Key Decisions for Investors

• Go long AMD / ARM / INTC vs short NVDA on a 1-3 month relative-value basis; thesis is mix shift toward CPUs and custom silicon while NVDA remains a flow-heavy laggard. Falsify if NVDA reclaims leadership versus SOX for two consecutive weeks or if the next guide implies another acceleration in data-center growth.
• Add AMAT and LRCX on pullbacks as a 3-6 month beneficiary of continued AI capex and memory-driven tool demand. Risk/reward is attractive if spending broadens; thesis breaks if foundry/tool order commentary rolls over or capex guidance is cut.
• Initiate a small long basket of TXN + ARM for 6-12 months to express increasing AI power-management and CPU orchestration content. Best entry is after a failed rally in NVDA, when the market is still paying for the next bottleneck rather than the old one.
• Avoid outright shorting NVDA here; if anything, use a conditional hedge and wait for a catalyst miss or lackluster repurchase update. If the company accelerates buybacks meaningfully, a short can squeeze hard even with flat fundamentals.
• Watch GOOGL and AMZN as indirect winners from custom silicon monetization; if either starts externally selling more of its chip stack, it is a negative for NVDA but positive for AVGO and the foundry/tool complex.