Nvidia's CFO Just Explained Why the AI Boom Is Eating Its Gross Margin -- and It's a Green Light for Micron

Nvidia reported fiscal Q2 revenue of $96.2B, more than doubling YoY and up 18% QoQ, with data center sales surging 117% YoY to $89B (hyperscalers $48.7B, +102% YoY). Despite the blowout, Nvidia guided gross margin down, expecting GAAP/adjusted gross margin of 74%±50 bps (vs. a 75% print) and citing rising high-bandwidth memory costs; memory commitments rose from $119B last quarter to $279B. The article frames this as a positive read-through for Micron, as higher memory pricing power and demand from Nvidia’s AI capex cycle should benefit HBM/DRAM suppliers.

Analysis

The key market mechanism is a transfer of surplus inside the AI stack: compute remains scarce, but memory is becoming the tighter bottleneck and is now capturing more of the economic rent. That favors MU and the broader HBM/DRAM supply chain, but it also means the next leg of AI spend is less accretive for system integrators and hyperscalers if memory inflation persists. In relative terms, the trade is stronger for memory than for GPUs over the next 1-3 months, even if the structural AI capex trend stays intact.

The risk is that this is still a cyclical pricing event dressed up as structural permanence. If HBM supply ramps faster than expected or if buyers redesign around lower memory intensity, the pricing power could fade within 2-4 quarters, especially after a demand-front-loaded surge. That would hit MU’s multiple first, because the market tends to extrapolate peak HBM margins too early; the falsifier is evidence of flattening DRAM/HBM ASPs or a guide that shows commitments expanding without corresponding gross margin lift.

Consensus is probably underweighting the second-order loser: not NVDA’s unit demand, but downstream AI platform economics. If component inflation keeps rising, hyperscalers may slow non-core deployments, push harder on custom silicon, or squeeze server supply-chain margins before they touch GPU orders. The other contrarian angle is that MU may be the cleaner trade than NVDA: the former gets direct pricing leverage, while the latter absorbs cost inflation and can only partially pass it through.

Key Decisions for Investors

• Long MU vs short NVDA for 1-3 months: relative-value bet that memory pricing captures incremental AI spend faster than GPU margins. Use a tight stop if NVDA guides back to margin stability or if MU commentary on HBM pricing turns incremental rather than tight.
• Add MU on pullbacks into the next earnings window, targeting a 6-12 month re-rate if HBM ASPs keep rising. Risk/reward is attractive if the market is still pricing MU as a normal cyclical despite AI-driven supply scarcity.
• Avoid outright shorts in NVDA here; if anything, use the print to trim or hedge into strength with short-dated calls against existing longs. The upside remains intact, but margin compression makes forward multiple expansion harder than the headline growth rate suggests.
• Watch SK Hynix and Samsung memory commentary as the key falsifier for the MU bull case. If both announce faster-than-expected HBM capacity additions, reduce MU exposure because the pricing tailwind can unwind faster than revenue models imply.
• Alert on hyperscaler capex guidance over the next 1-2 quarters: if memory inflation starts forcing order rationalization, consider a short basket of AI infrastructure beneficiaries with the highest operating leverage to component costs.