Samsung’s record earnings come with a warning: the AI memory shortage could get worse in 2027. The South Korean company expects demand to pull further ahead of supply.
During its Oct. 29 earnings call, Samsung said its high-bandwidth memory (HBM) supply for 2027 was already sold out and indicated tight conditions could last through 2028, according to Reuters.
The pressure extends beyond AI accelerators. Tight memory supply could raise server and PC costs, making component availability a constraint on how quickly businesses can expand their AI infrastructure.
Samsung's 2027 HBM supply is already committed
Samsung reported a record third-quarter operating profit of 107.4 trillion won ($80 billion). In its earnings release, the company said that "increasing AI-driven demand will far exceed the industry supply in comparison with 2026."
Reuters reported that Samsung signed long-term contracts with 10 major customers and was negotiating with 10 more. New chip factories take years to build, limiting near-term supply relief.
South Korean chipmakers expand as APAC hardware costs rise
Samsung and SK hynix together control about two-thirds of the global memory market, according to the Associated Press. They plan a combined 800 trillion won ($599 billion) investment in a chipmaking hub in southwestern South Korea.
SK hynix projected shortages through 2030 and said long-term contracts cover more than half its planned 2027–2028 capacity. The company is also investing heavily in domestic production, including a $64 billion chipmaking expansion in Cheongju, South Korea.
TrendForce estimates that the share of global notebook production outside China could fall below 20% in 2027 as brands reassess manufacturing costs and supply-chain efficiency. Meanwhile, AI hardware demand is boosting electronics export forecasts in the Philippines, highlighting how the AI boom is affecting manufacturers across the region.
What eWeek found: DRAM could stay tight as NAND eases
TrendForce's research points to different supply conditions in 2027. HBM and conventional server and PC DRAM face sustained demand pressure, while NAND supply could loosen later in the year.
Memory type | 2027 outlook | Potential buyer impact |
|---|---|---|
HBM | Samsung's supply is committed | AI accelerator orders may face allocation limits |
Server and PC DRAM | Capacity remains tight | Higher system prices or reduced memory specifications |
NAND flash | Supply may ease in the second half | SSD prices could stabilize sooner |
Mike Howard, director of DRAM and memory markets at TechInsights, told EE Times that "Memory has become one of the most important components in the AI hardware stack, often the gating semiconductor component."
For businesses investing in AI infrastructure, memory and packaging bottlenecks could delay hardware deployments. Greater SSD availability would offer only partial relief if HBM and conventional server DRAM remain constrained.
AI server and PC buyers face three procurement risks
IT leaders setting 2027 budgets need to know whether the hardware they order can ship with enough memory to support their workloads. A quoted server price offers little certainty if its accelerators or DRAM remain unallocated. Three checks can help:
- Supply and delivery: Ask vendors to specify the GPU/HBM and DRAM configurations they can actually supply, confirm component allocations, and distinguish firm delivery commitments from estimates. For AI projects, align hardware arrival dates with data center capacity and deployment schedules.
- Budget and performance: Obtain fully configured quotes with expiration dates, and model higher memory costs before approving purchases. Check whether political pressure on memory sourcing could also affect vendors’ component choices. Compare any lower-memory alternative against workload requirements and the possibility of replacing PCs earlier than planned.
- Fallback options: Identify compatible alternative systems, confirm they meet performance needs, and prioritize critical rollouts. If equipment arrives late, decide which deployments can be phased without delaying essential operations.
The broader implication is that access to memory could help determine which AI projects scale and when. Long-term supply agreements give manufacturers greater confidence to invest, but they also show how much future output customers are already reserving. A larger AI budget does not, by itself, establish that the required hardware will be available.
For businesses, that makes confirmed configurations and delivery commitments part of the investment decision. A cheaper system with less memory may fail to support the intended workload, while a delayed server can leave an otherwise funded project waiting for infrastructure.
The next signal to watch is whether new manufacturing capacity narrows the supply gap. TrendForce expects substantial output contributions from planned DRAM expansion to materialize in 2028. Until then, improving NAND availability could help storage buyers without removing the memory constraints facing AI servers.
Read more: Memory supply is only one factor shaping AI deployment schedules — HPE’s latest private cloud AI update shows how hardware options and software availability can arrive on different timelines.