Key points
- HBM demand for AI models outpaces current market supplies, draining wafer capacity.
High-bandwidth memory demand for artificial intelligence training is creating a severe supply bottleneck that is driving up GPU prices globally, Engadget reported.reported This shortage stems from manufacturers shifting resources toward AI data centers, leaving less capacity for consumer electronics and general-purpose memory production.Artificial intelligence training is using up so much computer memory that it is making video game cards much more expensive for regular people.
What it does
The industry is experiencing a shift where GPU makers prioritize AI infrastructure over consumer products. This reallocation has caused component prices to skyrocket and supplies to dwindle for everyday users. The phenomenon, known colloquially as chipflation, is outpacing general economic inflation levels significantly.
How to get it
Limits
HBM competes for the same wafer capacity as standard DRAM and other memory products. Because AI developers require HBM to both train and run models, this specialized memory demand far exceeds current market supplies, limiting the availability of chips for other sectors.
The takeaway
The shortage is expected to persist, with many chip producers having already sold much of their 2027 supplies.
Building on this
Data science teams must account for these supply constraints when planning infrastructure. With HBM demand far outpacing supply, developers cannot assume stable access to high-end training hardware. The competition for wafer capacity means that securing the necessary memory for model training will likely require long-term procurement strategies rather than spot purchases.
Why it matters
- Procurement budgets for AI training hardware will need to absorb significant price increases due to DRAM inflation.
- Long-term infrastructure planning must consider that chip producers are selling supplies years in advance.
- The shift from DDR4 to DDR5, combined with HBM demand, creates a dual-layer supply constraint for memory components.
Topics: hardware, AI infrastructure, supply chain, GPU
Updated
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
HBM
High-bandwidth memory, a specialized form of DRAM with faster connections and more bandwidth.
DRAM
Dynamic random access memory, a type of semiconductor memory used in computers and other devices.
Chipflation
The colloquial term for the rise in GPU and chip prices outpacing general inflation.
Questions readers ask
How much will DRAM prices increase?
JP Morgan reports that DRAM prices will have risen by 400 percent by the end of 2026 compared to 2024 levels.
Connected stories
Reading receipt
- HBM demand for AI models outpaces current market supplies, draining wafer capacity.
Still open: How long will the transition from DDR4 to DDR5 continue to impact supply chains?
Still open: Will reseller markets stabilize as chip producers focus on 2028 sales?