SK Hynix Posts Record Q3 Profit as Revenue Crosses 100 Trillion Won for First Time on AI Demand

The Quick Take

• SK Hynix Q3 2026 revenue surged 327.9% year-on-year to 104.62 trillion won ($78.86B), the first time quarterly sales crossed 100 trillion won
• Record quarterly profit exceeded analyst expectations of 98.34 trillion won, driven by AI memory chip demand
• The result confirms that AI infrastructure buildout continues to generate exceptional earnings power for leading memory suppliers

SK Hynix delivered a record-shattering Q3 2026 result, with revenue surging 327.9% year-on-year to 104.62 trillion won — equivalent to $78.86 billion — marking the first time the company has ever crossed the 100 trillion won threshold in a single quarter. The result beat analyst consensus of 98.34 trillion won and was driven almost entirely by demand for high-bandwidth memory chips used in AI accelerators, particularly for large language model training and inference infrastructure. The scale of the year-on-year growth reflects both how depressed the memory market was in Q3 2025 and how dramatically the AI capex cycle has reshaped demand for leading-edge memory products.

SK Hynix's record result has significant implications for the entire semiconductor supply chain. NVIDIA's AI chip dominance drives HBM demand directly to SK Hynix as its primary HBM3e supplier, meaning a read-across to NVIDIA's data center revenue trajectory is immediate. Samsung Electronics, competing in the HBM market but behind SK Hynix in qualification timelines, faces ongoing margin pressure. For memory sector peers including Micron Technology, the result validates that the AI-driven memory upcycle has further to run, supporting elevated ASP expectations and capex investment decisions. Indian technology companies with semiconductor exposure — particularly fabless design firms and IT services companies serving hyperscaler clients — stand to benefit from sustained AI infrastructure spending.

Watch SK Hynix's Q4 2026 earnings guidance as the clearest forward signal: if management raises HBM output targets or announces new customer qualifications beyond NVIDIA, it confirms that the AI memory upcycle is broadening beyond a single hyperscaler. The macro variable that could disrupt this thesis is a slowdown in hyperscaler AI capex — if Alphabet, Microsoft, or Meta signal capex moderation in their Q3 earnings reports, HBM demand assumptions would need to be revised downward. Additionally, any US export restriction tightening affecting AI chips to China would indirectly reduce total AI training demand and with it the HBM consumption trajectory that underpins SK Hynix's record revenue run.

Synthesized from 1 source.