If you ask whether memory is oversupplied right now, the answer is "no." Demand still exists. But we have to examine whether this demand will actually turn into consumption.
Bruce Bateman, Omdia's Taiwan-based senior semiconductor analyst, answered this way when asked about the possibility of a memory semiconductor peak-out at Omdia Tech Forum Seoul 2026, held at EL Tower in Yangjae-dong, Seoul, on the 15th. His assessment is that while the semiconductor market has not reached its peak, this year could be the period of the fastest growth rate.
Bateman said, "The memory supply problem will not be resolved until new fabs are built and reach high utilization rates," and "Fabs are currently running at maximum, and manufacturers are raising prices."
Omdia assessed that supplies of high bandwidth memory (HBM) are effectively sold out through 2027. However, Bateman projected that HBM prices could fall in 2028–2030. That means the memory cycle is unlikely to turn down immediately, but prices could decline over the medium to long term.
◇ "This year is the peak of growth speed, not market size"… HBM supply tight through next year
Recently, as questions have been raised about the sustainability of artificial intelligence (AI) data center investment, the so-called "semiconductor peak theory" has come to the fore. Big tech corporations are competitively securing AI accelerators and HBM, but there are concerns that if data center completions are delayed due to power and water shortages, orders may not translate into actual use.
Omdia expects global semiconductor market revenue to expand from $687 billion (about 1,026.38 trillion won) in 2024 to $1.6 trillion (about 2,390.4 trillion won) in 2026 and $2.1 trillion (about 3,137.4 trillion won) in 2028. It forecast this year's growth rate at 87.5% year over year. Although the growth rate will slow thereafter, it projected that market size will grow faster than the historical average through 2028.
Bateman said of this, "This year, the market is at the peak of growth speed, not size." Since AI infrastructure build-out is continuing, he noted, growth deceleration should not be equated with market contraction. Omdia expects this year's Americas semiconductor market to grow 112% centered on AI data center investment, and the Asia-Pacific market, where Samsung Electronics, SK hynix, and TSMC have production bases, to grow 87%.
Omdia analyzed that memory will drive this semiconductor market growth. It projected this year's memory semiconductor market revenue to rise more than 250% from a year earlier to over $800 billion (about 1,195.2 trillion won).
Bateman cautioned, however, "The $1.5 trillion–$1.6 trillion (about 2,241–2,390.4 trillion won) figure is revenue, not the scale of wafers produced," saying revenue growth should not be viewed as the same as higher production volumes. The impact of price increases on market size expansion needs to be distinguished.
On HBM demand, he said it will be "tight at least through 2027," citing Samsung Electronics' Pyeongtaek P4 and SK hynix's Cheongju M15X expansions as key supply variables. The target timing for SK hynix's M15X to reach high utilization was presented as mid-2027.
Bateman said, "There is no way to solve the supply problem before new fabs are completed," and "Chinese DRAM corporations, instead of cutting prices to gain market share, have raised prices to just slightly below Samsung Electronics, SK hynix, and Micron." He added, "HBM is very expensive right now, but prices will come down in 2028 or 2029, or by 2030 at the latest."
Bateman said the scale of major U.S. data center projects at risk of delay or cancellation due to power and water issues amounts to 9–12 gigawatts (GW). If data center construction falls through, orders for Nvidia's AI accelerators and the HBM that goes into them could also decline.
He said, "When data center construction is halted, the question is whether corporations will cancel Nvidia orders or receive the products and store them in warehouses," adding, "AI accelerators have lead times of more than a year, and buyers must prepay $2 billion–$6 billion (about 2.988–8.964 trillion won) to place orders." He added, "There isn't too much memory right now," but also said, "With every corporation inflating the quantities they need, we have to confirm whether demand is real."
◇ Korea's opportunity as a 'third physical AI supply chain' outside the U.S. and China
Omdia held a tech forum in Korea for the first time since establishing its Korea office. Presentations were given by Bateman; ▲ Lian Ze Su, Omdia Singapore senior analyst for AI and Robotics ▲ Joanne Goh, Omdia Malaysia practice leader for manufacturing and industrial automation ▲ Shiraz Aziz, Omdia Malaysia analyst for cloud and data centers ▲ Chris Ryu, Omdia China senior analyst for new energy vehicles.
Five analysts active in the Asia-Pacific region agreed that AI, which has developed around data centers, is expanding into the real world, including robots, automobiles, and manufacturing equipment. They said the "physical AI" market is growing rapidly.
Su projected that shipments of embodied intelligence robots that combine quadrupeds and humanoids will increase from about 166,000 units this year to 1.1 million in 2030 and about 5 million in 2035. After 2032, adoption will expand across industries, and by 2035, the average selling price of high-end models is expected to fall below $20,000 (about 29.88 million won).
Su assessed that China leads in mass-production scale and training data, while the United States leads in AI models and foundational technologies. Su said Korea can supply an alternative to countries that do not want to rely entirely on the United States and China.
Su said, "If robots are treated as a sovereign technology going forward, countries will want to keep supply chains at home or at least avoid depending entirely on the United States and China," adding, "Korea can provide a third-country alternative outside the U.S. and China." Su cited Korea's semiconductor, battery, and auto parts supply chains, as well as its global distribution and service capabilities, as strengths.
Su pointed to ▲ talent development ▲ ecosystem collaboration among corporations ▲ securing high-quality data as elements the Korean government needs to prepare. Su said, "If you are developing robots for the Korean market, you must train models with data that reflects Korea's work environments and users," adding, "An ecosystem is also needed in which robot manufacturers and semiconductor, component, and cloud corporations can collaborate quickly."
In a welcome address, Kim Su-yeon, head of Omdia Korea, said, "If digital transformation determined corporate competitiveness in the past, from now on the key to new competitiveness will be how effectively AI and semiconductors, data, robots, and physical infrastructure are combined," adding, "Only when cross-industry cooperation and ecosystem building, talent development, and trustworthy technology development advance together can physical AI deliver tangible value."