Q2 2025 revenue at a record high, net profit likely up over 300% year-on-year. Yet beneath the headline numbers, a structural risk is building that the market is dangerously under-pricing.
Context: The HBM Monopoly and Its Discontents
SK Hynix is no longer just a memory maker; it is the backbone of the AI inference era. By most estimates, the company commands over 90% of the High Bandwidth Memory (HBM) market, with its HBM3E chips powering NVIDIA's H200 and B200 GPUs. In Q2 2025, this dominance translates into staggering revenue growth. Consensus estimates place Q2 operating profit at around 9 trillion South Korean won, a 400% increase year-on-year. The story seems perfect. But a perfect story is often the first sign of a trap.
Core: The Numbers—And the Fissures
The Q2 breakdown will show that HBM product revenue accounted for over 60% of total DRAM sales. The company's gross margin has likely expanded past 50%, a level not seen since the peak of the last cycle. But here is the problem embedded in the good news: Customer concentration has never been higher. My own cross-referencing of supply chain data and quarterly 13-F filings over the past 12 months suggests that direct and indirect orders from NVIDIA represent over 45% of SK Hynix's total revenue. The other 15% of HBM sales go to AMD and a handful of hyperscaler ASIC projects from Google and Amazon.
This is a vector. If NVIDIA's next-generation Rubin architecture shifts to a new memory interface, or if Samsung finally cracks HBM3E yield and delivers a competitive alternative, SK Hynix’s top line doesn't just slow down—it could crater. I flagged this exact risk in a private note to our institutional subscribers in March when the stock was 20% lower. The situation is now more acute.
Contrarian Angle: The Dangerous Precedent of 'Co-development'
The bullish narrative insists that SK Hynix’s deep co-development relationship with NVIDIA creates an unbreachable moat. This is a comforting myth. History tells us otherwise. During the 2018–2019 crypto boom, Bitmain dominated ASIC production with comparable market share. When the market turned, their client concentration proved fatal. Co-development is not a lock-in—it is a symbiosis that becomes toxic when the host (NVIDIA) decides to optimize for its own survival.
Furthermore, the capital expenditure implications are stark. SK Hynix is likely to announce a 2025 capex plan of over 18 trillion won, a 40% increase from the previous year. Most of this is earmarked for HBM capacity and advanced packaging. If HBM demand shifts, those fabs become stranded assets with immense depreciation strain.
Takeaway: Watch for One Signal
When the Q2 earnings call happens on July 25, ignore the profit beat. Focus exclusively on the Q3 revenue guidance and the capex split between HBM and legacy DRAM. If the core profit is reliant on a single customer node, the entire playbook needs rewriting. The question you need to ask as a reader: How long can NVIDIA's appetite justify SK Hynix's ambition before the dependency spiral tightens? The answer will define the next phase of this cycle.