Memory Costs Rise: Can NVIDIA Protect Its 70%+ Gross Margin?

NVIDIA expects non-GAAP gross margin to fall to 74% ±50 bps in Q3 FY2027 (from 75% in Q2) due to sharply higher memory costs, before dropping to 71%-72% in Q4 FY2027 and recovering to 72%-73% in FY2028. Despite the near-term margin pressure, Q2 revenue rose 106% YoY to $96.2B and Data Center revenue jumped 117% to $89B, while non-GAAP gross margin improved 250 bps YoY to 75% on strong Blackwell demand. NVIDIA is increasing supply/capacity commitments to $279B (mainly for memory) and plans price increases beginning in FY2028 to help defend margins above 70%.

Analysis

This is less an NVIDIA demand problem than a margin-transfer event up the AI supply chain. The first-order loser is the system integrator that has to pre-buy scarce memory to keep shipments flowing; the second-order winners are the memory vendors and any upstream capacity owner able to reprice into shortage. That dynamic matters because NVDA is valued like a premium compounder, so even a manageable gross-margin air pocket can compress the multiple if investors start treating FY27/FY28 as peak economics rather than a glide path.

Near term, the stock reaction is likely driven by margin slope, not revenue durability. Over the next 1-3 months, watch whether memory vendors push another pricing round or whether NVIDIA offsets with mix/price faster than expected; if management can hold margins above ~73% despite procurement pressure, this becomes noise. Over 6-18 months, HBM/DRAM capacity additions should ease the constraint, which would shift the debate from "can NVDA protect margins" to "who captured the most pricing power during the shortage."

Relative losers are AMD and INTC because thinner gross margins leave less cushion if AI server bill-of-materials costs keep rising. AMD has the cleaner AI growth profile but less pricing leverage; INTC has the additional burden of foundry/CapEx intensity, so any cost inflation hits operating leverage harder. The contrarian point is that rising memory prices are bullish evidence for AI demand elasticity: the market is still supply constrained, not demand exhausted.

Key Decisions for Investors

• Long MU vs short NVDA on a 1-3 month horizon: best expression of memory inflation transferring value from the platform owner to the component supplier. Falsify if NVDA holds margins above 73% into the next quarter or MU guides to easing HBM/DRAM pricing.
• Short AMD on rallies or buy put spreads into the next earnings window: AMD has improving AI mix but a much thinner gross-margin buffer, so any persistent memory inflation should hit its earnings revisions harder than NVDA’s. Cover if Data Center gross margin keeps expanding or if management explicitly passes through costs.
• Avoid chasing NVDA on strength; look to add only on a 3-5% post-print drawdown. This is still a quality compounder if margins stabilize above 70%, but upside is better if the market overreacts and then re-rates the name back toward premium growth once pricing power is confirmed.
• Watch MU and SKHYV as the cleaner beneficiaries of AI scarcity economics over the next 2-4 quarters. If their pricing/guidance keeps tightening, it supports a longer-duration long in memory suppliers versus semicap broad market exposure.