SK Hynix Q1 profit surges over 400%, beats expectations on stellar AI demand

SK Hynix reported first-quarter operating profit of 37.61 trillion won, up 405% year-on-year and well above expectations, while revenue surged nearly 200% to 52.58 trillion won. The company said it is benefiting from strong AI-driven memory demand, with high-bandwidth memory and broader memory pricing remaining key tailwinds. Management expects memory prices to keep rising in coming quarters, a positive read-through for AI hardware suppliers.

Analysis

The key read-through is not just that AI demand is strong, but that the bottleneck has shifted from compute to memory intensity. As model sizes and inference loads rise, HBM becomes a structural tax on every AI buildout, which means suppliers with capacity discipline can keep extracting pricing power even if GPU order growth normalizes. That puts NVDA in a mixed spot: near-term demand remains intact, but the market may be underestimating how much of the AI value chain’s margin expansion is migrating upstream to memory and packaging.

Second-order, this is bearish for any AI hardware ecosystem names that rely on cheap component costs or fast price deflation to sustain unit growth. If memory inflation persists into the next 2-3 quarters, it can squeeze system integrators, delay some enterprise deployments, and create a temporary cap on gross-margin expansion for server OEMs and cloud capex-heavy users. The implication is a wider dispersion trade within semis: memory suppliers with HBM mix and pricing power should outperform GPU-adjacent beneficiaries that are still valued on accelerating top-line growth.

The contrarian risk is that consensus may be extrapolating peak pricing too far. Memory is historically prone to sharp supply responses once lead times and utilization stay elevated for multiple quarters; a production conversion wave could unwind pricing faster than the market expects, especially if AI capex growth decelerates in late 2025. For NVDA, that would not hit near-term units first—it would hit sentiment via concerns that AI infrastructure economics are becoming less elastic as input costs rise.

From a trading standpoint, the cleaner expression is not a naked long on AI beta, but a relative-value long of memory leaders versus broader AI hardware. The asymmetry is strongest over the next 1-2 quarters while pricing is still firm and guidance revisions are likely to keep drifting higher; beyond that, the risk shifts to margin normalization and cyclicality.

Key Decisions for Investors

• Long NVDA / short a basket of AI hardware beneficiaries with more cost pass-through risk (e.g. SMCI, DELL) over the next 1-2 quarters; thesis is memory inflation compresses downstream margins before it impacts unit demand.
• Add on pullbacks to HBM-exposed memory leaders rather than chasing broad semi beta: prefer a basket approach to SK Hynix-adjacent names (or U.S. memory proxies if liquidity is required) for a 3-6 month trade with asymmetric pricing-power upside.
• Use call spreads rather than outright longs on NVDA for the next earnings cycle; upside remains intact, but the better risk/reward may be capped if market starts re-rating who captures AI capex margin.
• Avoid being long low-margin server/OEM names into the next 1-2 quarters unless paired with memory exposure; rising component costs create a negative gross-margin surprise risk.