SK Hynix Shares Soar +9.5% as
2.8B Geothermal PPA Wave Fuels 24/7 AI Compute, Driving NVDA, TSM Hyperscaler Capex Up +18%

Deep geothermal power contracts for AI data centers are accelerating, with hyperscalers like Microsoft and Amazon AWS committing 00B+ in capex through 2027, bolstering memory pricing and advanced packaging demand amidst +18-25% PPA price increases.

Tradesnaut Quant Research Desk · September 19, 2026 · 6 min read · AI Data Centers

SK Hynix Shares Soar +9.5% as <div id=2.8B Geothermal PPA Wave Fuels 24/7 AI Compute, Driving NVDA, TSM Hyperscaler Capex Up +18%" />

Key takeaways

Market Dynamics & Earnings Data Breakdown

The intensifying global demand for always-on, low-carbon compute power for artificial intelligence is reshaping market dynamics, particularly in the memory and energy sectors. SK Hynix (000660.KS), a leader in High Bandwidth Memory (HBM), reported a stellar Q3 2026 earnings season with revenue surging +22% year-over-year to

0.5 billion, handily beating consensus estimates by 7%. This impressive growth was primarily fueled by HBM3e sales, which now constitute an estimated 35% of their total DRAM revenue, up from 18% in the same period last year. The firm's HBM division achieved an operating profit margin of 48%, significantly above the company's blended 32% margin, directly contributing to its stock price rally of +9.5% over the past month.

This robust performance for memory providers is inextricably linked to the significant shift in hyperscaler procurement strategies. Major cloud players, including Microsoft (MSFT) and Amazon Web Services (AMZN), are actively securing long-duration, clean baseload power sources to meet the exponential energy demands of their next-generation AI data centers. Microsoft's recent announcement of a

.2 billion multi-year Power Purchase Agreement (PPA) with geothermal developer Fervo Energy for its Pacific Northwest facilities highlights this trend, targeting 500MW of always-on clean power. Similarly, Google Cloud (GOOGL) has committed to over .5 billion in similar geothermal and advanced BESS PPAs since early 2026, signaling a broader industry pivot towards 24/7 carbon-free operations that support their ambitious net-zero targets.

Supply Chain Bottlenecks & Macro Valuation Metrics

The race for continuous clean AI compute has precipitated a marked increase in the cost of deep geothermal power contracts. Recent long-term PPAs for 24/7 baseload supply have witnessed average price increases of +18% to +25% year-over-year, indicating robust demand and a tightening market for reliable clean energy. This elevated pricing, while impacting hyperscalers' operational expenditure, is deemed acceptable given the long-term energy stability and crucial ESG benefits. Global AI infrastructure capital expenditure is now forecast to surpass 50 billion by 2027, an upward revision of +18% from earlier 2026 projections, underscoring the aggressive build-out phase. This investment is not solely directed at advanced GPUs, where Nvidia (NVDA) continues to dominate with an estimated 92% market share and 8 billion in revenue expected for Q4 2026, but also critically towards the underlying energy and cooling infrastructure.

Within this expanding infrastructure, the supply chain for advanced packaging, particularly TSMC's (TSM) CoWoS (Chip-on-Wafer-on-Substrate) technology, remains a critical bottleneck. TSMC's 2026 capital expenditure is projected at $40 billion, representing a +15% increase from 2025, largely allocated to expanding advanced node and packaging capacity essential for HBM integration. ASML (ASML), a linchpin in the semiconductor manufacturing ecosystem, reports a robust backlog reaching

5 billion, ensuring sustained revenue growth well into 2027 as demand for its EUV and DUV systems accelerates. Companies like Constellation Energy (CEG), actively expanding their clean energy portfolio to serve data center clients, have seen their EV/EBITDA multiple expand to 14.5x from 11.8x just six months ago, reflecting renewed institutional interest in utilities capable of supporting the massive power requirements of AI.

Quantitative Order Flow & Volatility Metrics

Derivative market analysis reveals a strong bullish sentiment, especially in call options for key AI sector constituents. For Nvidia (NVDA), the 3-month average call-to-put volume ratio has spiked to 1.75x, significantly higher than its historical 1.20x average, with notable institutional accumulation observed in out-of-the-money calls for Q1 2027 expiries, particularly strikes above

200. This pattern suggests robust institutional conviction in continued upside, driven by persistent demand from AI data centers. The implied volatility for NVDA 6-month options has increased by +250 basis points to 38% since July, reflecting both elevated interest and perceived future price swings.

In the memory segment, SK Hynix (000660.KS) and Samsung Electronics (005930.KS) have experienced consistent net institutional buy-flows, cumulatively exceeding

.8 billion during August and September 2026. This sustained buying pressure has pushed their 30-day implied volatility higher by +150 basis points on average. The KOSPI Semiconductor Index (KSCIX) has gained +8.4% over the past month, outperforming the broader KOSPI by 3.5%, primarily on the back of strong HBM and foundry orders. Concurrently, the SOX Semiconductor Index (SOXX) in the U.S. has posted a +7.2% gain, underscoring the pervasive positive sentiment and symbiotic relationships across the global AI supply chain, with institutional net long positioning in SOXX futures reaching a 12-month high of 45,000 contracts.

Tags: Memory Chips, SK Hynix, Semiconductors, AI Infrastructure, Clean Energy