Supply and demand fluctuations are nothing new for the memory market, but artificial intelligence has raised the stakes for high-bandwidth memory chips, says Shawn DuBravac, chief economist for the Global Electronics Association.
As AI hyperscalers ramp up workloads and build new data centers, they’re purchasing many of the high-bandwidth memory silicon chips that are typically available for PCs, smartphones and other devices. That’s pushing some chipmakers to prioritize HBM over other types of memory, including the workhorses Dynamic Random-Access Memory and NAND flash memory.
“It’s not that the demand per se for DRAM and NAND has increased, it’s that the capacity to produce DRAM and NAND has been reallocated to other areas, so that’s really the main story here,” says DuBravac.
A 2026 report from the Semiconductor Industry Association highlights just how much AI data centers depend on chips, with one AI server rack containing more than 4,500 packaged chips.
Compared with previous disruptions in the memory market, the Global Electronics Association characterizes the current crunch as more structural and, most likely, sustained.
“We’re probably looking at least a year of continued constraint, and that could easily go into 2028,” says DuBravac.
Click the banner below to find expert advice about IT supply chain concerns.
Device Manufacturers Face Higher Costs, While Chipmakers Scale With Caution
One outcome of the silicon grab is that manufacturers’ usual methods of navigating supply constraints, such as high volume commitments and strong supplier relationships, have lost some of their edge. They’re still sourcing memory, but with higher costs and longer lead times.
In a Global Electronics Association survey, 83% of North American manufacturers say prices have increased, with 39% reporting a “significant” jump.
READ MORE: How does quantum computing shift business forward?
Semiconductor manufacturers are scaling production capacity, but that expansion won’t happen overnight, says DuBravac. And, with AI’s trajectory unclear, chipmakers are cautious about rightsizing those investments.
“They need to be really confident that they’re going to be able to monetize that capacity in the short term and over the longer term,” he says.
For Businesses, Memory-Intensive Compute May Be Reserved for Select Users
Non-silicon materials do have a role to play in the memory market. For example, gallium nitride suits small, energy-efficient devices that require fast charging, while silicon carbide works well for electric vehicles, industrial power and other applications requiring heat tolerance and durability.
But alternative materials have their own challenges in cost and sourcing, and DuBravac says that widespread adoption of non-silicon components is a longer-term strategy.
“The most straightforward way to address this is to just put less memory in the device. Rather than a materials shift, there could be a shift in architecture and how we configure memory,” he says, pointing to memory pooling as one tactic.
$2.8 trillion
The projected global spending on semiconductors for new AI data center infrastructure through 2028
Source: Source: Semiconductor Industry Association, “Powering AI: The Semiconductor Ecosystem at the Foundation of Data Centers,” June 2026
Device buyers are likely to see lower memory specs, paused upgrade cycles and higher prices, according to a 2025 TrendForce analysis. That means businesses planning PC upgrades must be strategic about those purchases, says DuBravac.
“The question they should be asking is how much memory they need on the device. At the end of the day, they will most likely be paying higher price points for computers, especially if they want memory-intensive machines for AI compute,” he says.
DISCOVER: It's time to rethink data center security in the age of AI.
With IDC projecting that pricing and other ripple effects will last at least through 2027, businesses aren’t likely to see a return to normal anytime soon. Small businesses looking to adapt to these trends can take some steps to mitigate their risks.
For example, some vendors offer options such as “buy and hold,” in which a business purchases hardware early to lock in pricing and availability, storing it in secure warehouses until they needed for deployment. Other small to medium-sized businesses may explore switching to devices that require less memory or use recycling programs to sell back old equipment and use the proceeds to purchase new equipment. Ultimately, the challenges SMBs face in processor and memory procurement will require creative solutions to overcome them.
