[Edaily Marketin Won Jae-yeon Reporter] As the AI market expands in earnest, venture capital is spreading beyond graphics processing units (GPUs) and high-bandwidth memory (HBM) to memory expansion and data transmission technologies. This is because, as AI models grow larger, a data center’s competitiveness depends not only on computational performance but also on how quickly it can store and transfer vast amounts of data.

According to The VC and the investment industry on the 16th, Exina, a South Korean Computational Express Link (CXL)-based computational memory company, recently secured approximately 200 billion won in Series B funding. The company’s valuation is reportedly around 800 billion won. Point Two Technology, an AI data center interconnect company, also secured $76 million (approximately 112.1 billion won) last April. These two companies alone have attracted over 300 billion won in funding this year.

Neither company manufactures GPUs directly. Exina develops technology to expand and share memory within servers and reduce data transfer volumes, while Point Two Technologies develops technology to increase data transfer speeds between GPUs and servers. In essence, the investment industry’s focus is shifting toward technologies that maximize the utilization of existing equipment rather than acquiring additional high-cost GPUs.

As AI models grow in size, memory shortages and data transmission delays become factors that degrade the performance of the entire system. If the necessary data is not supplied in a timely manner, GPUs are put on standby, increasing the burden of power consumption and operating costs. In particular, as the focus of the AI industry shifts from model training to inference—the process of providing actual services—the importance of cost efficiency is growing even further.

Overseas, the valuations of related companies are also rising rapidly. U.S.-based optical interconnect company Ayar Labs raised $500 million (approximately 742.7 billion won) last March, achieving a valuation of $3.75 billion (approximately 5.57 trillion won). Optical switch startup NI also raised $80 million (approximately 118.8 billion won) in April. Both companies are developing technologies to increase the transmission speed of data flowing between AI semiconductors and reduce power consumption.

The government has also recently been promoting AI data centers as a core industry. Earlier, in its second-half work plan reported to President Lee Jae-myung on the 16th, the Ministry of Science and ICT designated AI data centers, physical AI, and K-semiconductors as the “three major megaprojects” and announced plans to support private investment. The AI data centers that SK, GS, and Naver plan to build by 2029 will have a total capacity of 8.4 gigawatts (GW), and the three companies’ investment plans—including external funding—amount to approximately 550 trillion won.

The Ministry also plans to form an inter-ministerial task force (TF) and support domestic companies’ entry into the supply chain—including domestically produced AI semiconductors, cloud services, and power and cooling equipment—during the data center expansion process. This reflects expectations that data center construction will not be limited to GPU purchases but will lead to actual demand from infrastructure companies for memory, storage, and interconnects.

An official from an investment firm that invested in Exina stated, “As demand for AI inference grows, it becomes increasingly important not only to have the absolute computational performance of GPUs but also to deliver data quickly and utilize memory resources efficiently,” adding, “Future investments will gradually expand beyond GPUs themselves to infrastructure companies that solve memory bottlenecks and data transfer issues.”