SK Hynix's $130 billion return plan meets a 6.4x forward multiple as AI demand accelerates
The AI-driven memory cycle is rewriting the earnings picture for semiconductor producers, and the latest evidence arrived from Seoul on July 29. SK Hynix (SKHY), with a market capitalisation of $1.19 trillion, reported…
Key takeaways
- SK Hynix reported second-quarter revenue of $56.9 billion, a 257% year-on-year increase, with operating profit up 557% to $43.4 billion (reported July 29).
- The company announced a $29 billion buyback and a plan to return over 50% of cash flow from 2025 through 2027 to shareholders, which JPMorgan estimated could total at least $130 billion.
- SK Hynix trades at a compressed forward price-earnings ratio of 6.4x despite its strong earnings trajectory.
- SKHY shares fell roughly 9% over the five trading sessions before publication, with Micron and Sandisk showing similar declines.
- Alphabet's TurboQuant compression method could reduce the memory needed to run large language models by six times, posing a longer-run structural risk to memory demand.
The AI-driven memory cycle is rewriting the earnings picture for semiconductor producers, and the latest evidence arrived from Seoul on July 29. SK Hynix (SKHY), with a market capitalisation of $1.19 trillion, reported second-quarter revenue of $56.9 billion, a 257% year-on-year advance, with operating profit up 557% to $43.4 billion. The company has also announced a $29 billion buyback plan and a commitment to return more than 50% of cash flow from 2025 through 2027 to shareholders; JPMorgan estimated the total could reach at least $130 billion.
The result fits a sector-wide pattern. Micron (MU) and Sandisk (SNDK) are moving through the same AI-driven demand environment, each tracking the proliferation of AI infrastructure that has lifted earnings across the memory complex. The broader cycle may have more room: wider adoption of cheaper Chinese AI models in Western markets could accelerate total AI utilization over the next 12 to 18 months, extending the current demand cycle further than consensus pricing implies.
Against that backdrop, the forward valuation looks compressed. SK Hynix trades at a forward price-earnings ratio of 6.4x. Analyst sentiment on flash-memory names has stayed cautious despite the earnings trajectory, and SKHY has shed roughly 9% across the five trading sessions before publication, with Micron and Sandisk tracking similar declines. Nvidia (NVDA), scheduled to report its financial results on August 26, is one named catalyst that could reset the near-term read-through for the memory complex.
The memory-efficiency question
The longer-run risk is structural. Alphabet has developed TurboQuant, a compression method that CNBC reported could reduce the memory required to run large language models by six times. SK Hynix CFO Kim Woo-hyun has argued that memory-efficiency technologies are evolving to maximise context processed per unit of memory, implying that aggregate unit demand may not fall in step with per-task efficiency gains. The harder calculation is whether demand from a growing number of AI-enabled devices will outpace per-device compression over time. That question remains open, and at 6.4x forward earnings, the price reflects the difficulty of answering it.
Source · 來源