[Edaily Reporter JAEMIN SONG ] Contrary to concerns that the memory market may have peaked due to a slowdown in artificial intelligence (AI) infrastructure investment, global Big Tech companies are successively raising their capital expenditure plans. As they continue to compete to secure data centers and servers—even while accepting deteriorating cash flow and financial burdens—this lends credence to the outlook that demand for high-bandwidth memory (HBM) and server DRAM from SamsungElectronics and SK hynix will persist for the long term.

According to industry sources on the 24th, Alphabet, Google’s parent company, has raised its capital expenditure plan for this year from the previously projected $180–190 billion to $195–205 billion. The market expects that capital expenditures by major cloud service providers (CSPs)—including Google, Amazon, Microsoft (MS), and Meta—will reach $1.23 trillion (approximately 1,800 trillion won) next year, an increase of about 55% compared to this year.

AI model companies have also joined the investment wave. OpenAI is reported to have raised its projected spending on computing infrastructure through 2030 from $600 billion to $750 billion. In Georgia, a 3.2-gigawatt (GW) data center is being built at a total cost of over $30 billion, and SpaceX AI is also considering plans to construct a new large-scale data center in Texas. Tesla, too, plans to invest over $25 billion in AI infrastructure and other areas this year.

This is viewed as a direct boon for the domestic memory industry. Alphabet allocated approximately 60% of its $44.9 billion in second-quarter capital expenditures to server purchases. This means that even as the burden of investment costs increases, it is difficult to cut spending on servers and core semiconductors, which determine competitiveness in AI services. As competition in AI inference intensifies, the importance of not only computational performance but also memory capacity and bandwidth is growing even further.

The amount of HBM required per AI accelerator is also increasing rapidly. NVIDIA’s next-generation graphics processing unit (GPU), “Rubin,” is expected to feature 288 gigabytes (GB) of HBM4, while AMD’s “MI455X” will be equipped with 432 GB of HBM4. As the memory requirements per server increase, cloud service providers (CSPs) and AI semiconductor companies are moving to sign long-term supply contracts spanning three to five years with SamsungElectronics and SK hynix. With prices for general-purpose DRAM also on the rise, the boom that began with HBM appears to be spreading to the entire memory market. As a supplier-dominated market takes shape, this structure puts suppliers in a favorable position to raise memory prices in subsequent contracts.

Kim Dong-won, Head of Research at KB Securities, stated, “As of the third quarter of this year, SamsungElectronics’ memory supply to Google is estimated to account for 50% of server DRAM demand and 80% of HBM demand,” adding, “Currently holding the top market share in memory supply to Google, SamsungElectronics is the biggest beneficiary of Google’s AI ecosystem.” SamsungElectronics plans to expand production of 1c DRAM-based HBM4 next year and begin mass production of HBM4E in the first half of 2028. SK hynix, a leader in the HBM market, is also expected to see improved profitability due to supply shortages and the expansion of long-term contracts.

SK Group Chairman Choi Tae-won also predicted at a recent press conference, “Since AI accounts for more than half of the semiconductor market, overall demand is expected to increase by 50–60% or more,” adding, “With virtually no increase in supply next year, the gap between supply and demand is bound to widen further.”

However, the massive investment costs are a key variable that will determine the sustainability of future infrastructure expansion. With Alphabet and Tesla’s second-quarter free cash flow turning negative, and major companies even expanding their corporate bond issuances, pressure is mounting to prove the profitability of AI businesses. An industry official said, “For now, Big Tech is prioritizing securing infrastructure over cost-cutting, so it is unlikely that memory demand will decline,” adding, “Going forward, we need to monitor not only the total investment amount but also how consistently funds continue to flow into essential areas such as servers and semiconductors.”