SK Hynix posts record first-quarter profit, in line with estimates as memory prices climb

SK Hynix posted record first-quarter revenue of 52.58 trillion won and operating profit of 37.61 trillion won, with revenue topping 50 trillion won for the first time and profit nearly doubling from the prior quarter. Results were slightly below LSEG smart estimates, but the company is benefiting from surging AI-driven HBM demand and a broader memory shortage that is supporting pricing. Counterpoint said SK Hynix still leads HBM with a 57% market share, reinforcing its competitive position in AI semiconductors.

Analysis

This is not just a memory upswing; it is a capacity bottleneck being repriced into the entire AI stack. The most important second-order effect is that HBM scarcity keeps pricing power with the few suppliers that can actually ship qualified product, while forcing hyperscalers and GPU OEMs to over-order and forward-build inventory, which can create a self-reinforcing demand loop for another few quarters.

For NVDA, the near-term impact is mixed: scarcity supports the value of its platform because it validates AI capex urgency, but it also raises bill-of-materials costs and creates the risk that some customers delay cluster deployments if memory lead times stretch further. The real beneficiaries beyond the obvious memory names are the integrated supply-chain chokepoints around advanced packaging, test, and substrate capacity, where utilization can stay tight even if wafer starts eventually catch up.

The market is probably underestimating the duration of the shortage. If incremental wafer capacity takes 4-5 years to arrive, then the earnings cycle is not a one-year spike but a multi-year regime shift; that argues for buying on any pullback rather than chasing headlines. The main reversal risk is not demand but supply normalization through aggressive capex, yield improvements, or product substitution away from the most memory-intensive AI architectures, which would likely show up first as margin compression before unit demand rolls over.

Consensus is still treating this as a clean NVDA positive, but the better setup may be relative value within semis: long the constrained enablers of AI deployment, short the names most exposed to eventual memory price mean reversion. The overdone part is assuming every AI beneficiary rises equally; the underappreciated part is that elevated memory costs can compress system-level gross margins even as unit demand remains strong.

Key Decisions for Investors

• Maintain/add to NVDA on 1-3 month dips, but hedge with a small short basket of server/OEM names most exposed to component-cost inflation; risk/reward favors staying long the AI platform while limiting margin-compression risk.
• Long HBM-capable memory suppliers vs. legacy DRAM exposure in a 6-12 month pair trade; thesis is that pricing power persists until meaningful new capacity comes online, while commoditized DRAM becomes the eventual mean-reversion loser.
• Buy medium-dated call spreads on NVDA into any memory-led pullback in AI semis; structure for upside continuation over the next 2-4 quarters while reducing premium bleed if the market pauses.
• Watch for signs of customer inventory build and order pushouts over the next 1-2 quarters; if those appear, trim cyclical semiconductor beta and rotate toward names with contractual pricing or bottlenecked supply.