[Edaily Reporter Park Sun-Yeop ] An analysis suggests that improvements in the efficiency of artificial intelligence (AI) models will expand the range of applications rather than reduce semiconductor investment and memory demand, thereby boosting overall demand. SK hynix is expected to transition to a business structure with lower earnings volatility than in the past, driven by expanded production of High Bandwidth Memory (HBM) and an increase in long-term supply contracts.
Kim Dong-won, Head of Research at KB Securities, maintained a “Buy” rating and a target price of 4.2 million won for SK hynix(000660)in a report released on the 22nd. This represents an upside potential of 129% compared to the closing price of 1.836 million won on the 21st.

Recently, concerns were raised that the efficiency improvements in algorithms—following the unveiling of the new “Kimi K3” AI model by China’s Moonshot AI—could lead to a decline in AI investment and a slowdown in memory demand. However, KB Securities assessed that similar concerns arose when DeepSeek and TurboQuant were introduced in the past, but these proved to be short-lived phenomena.
Division Head Kim predicted that as AI inference costs decrease, companies will integrate multiple AI agents to perform more tasks. He explained that even if the memory capacity required for a single inference is reduced to one-third of the current level, if the frequency of AI usage increases 20-fold, overall memory demand will rise approximately sevenfold.
He explained, “Improvements in the efficiency of AI models and semiconductors are not factors that reduce investment; rather, they further stimulate AI demand by lowering service prices and expanding applications,” adding, “This is a phenomenon similar to how improved fuel efficiency in cars lowers operating costs but can actually increase overall gasoline consumption.”
He also predicted that the competition among global Big Tech companies to invest in AI would continue. KB Securities estimated that Google and Meta’s AI investments this year would total 270 trillion won and 220 trillion won, respectively. The forecast indicates that these two companies’ investments alone will account for 46% of total AI investment by U.S. Big Tech firms.
Google is expected to secure a leading position in AI infrastructure by guaranteeing rental fees for external data center operators through next year, while also pursuing the development of next-generation server chips optimized for its proprietary AI model, Gemini. Meta plans to secure a total of 14 gigawatts (GW) of AI computing infrastructure by adding another 7 GW next year, following the construction of a 7-GW AI data center this year.
SK hynix’s revenue structure is also projected to change significantly for the first time in a decade. The share of business-to-business (B2B) revenue—centered on Big Tech and AI data centers—is expected to expand from approximately 30% in 2017 to 70% by 2027.
Analysts suggest that as the share of HBM production increases starting next year—limiting the capacity to expand general-purpose memory supply—a higher proportion of long-term supply contracts could reduce profit volatility and improve the predictability of earnings. Unlike the memory bull cycle of 2017–2018, the shift toward a structure with stable customers and supply periods could also pave the way for a revaluation of the company’s stock.
KB Securities estimated SK hynix’s second-quarter revenue this year at 82.4 trillion won and operating profit at 62 trillion won. Operating profit represents a 573.1% increase compared to the same period last year. The firm projected full-year operating profit for this year to reach 258.1 trillion won, rising to 394.1 trillion won by 2027.