It's a critical component for the AI industry, you're unlikely to ever own one yourself, and yet it's the reason computers will be more expensive over the coming year.
-
It's called HBM — and although early versions have existed on the market for many years, it only rose to prominence recently, driven by the AI revolution.
The component is built from layers of RAM stacked on top of one another.
Its operating frequency is lower than that of standard RAM, but it features an extremely high number of data channels for transferring information to the processor, allowing it to move a massive amount of data every second.
-
AI servers are built from standard components — an Intel or AMD CPU, and a large amount of server-grade RAM.
The main processor manages communication with the GPU subsystem, which consists of multiple high-powered processing units with HBM modules attached directly to them, serving as high-capacity RAM.
The need for this component stems from the fact that an AI model is built on billions of individual parameters, all of which must be stored simultaneously in RAM in order to perform computational operations on them.
-
Manufacturing such a component is an extremely complex and delicate process.
The memory layers must be placed with precision, one on top of the other, with a vertical data channel running between the layers and a large number of communication channels leading out to the processor.
Any defect during the manufacturing process means the component is discarded and production starts over.
Although RAM is not typically expensive, the fact that the compute engine simply cannot function without this component has turned it into a significant bottleneck — and tech giants are willing to pay a premium for it.
-
The result is that traditional RAM production lines are now running at full capacity manufacturing the more profitable HBM memory, creating a shortage in the conventional PC market that is expected to persist through the end of 2026.
The fact that RAM has a unique architecture — incorporating a capacitor in addition to transistors — means that other production lines cannot be retooled for RAM manufacturing.
Lenovo has already announced that it secured sufficient inventory for all of 2026 in advance; however, smaller manufacturers, or those that failed to anticipate this trend, may face serious difficulties producing computers without this relatively affordable yet critical memory.
Pictured: the HBM3E chip, manufactured by Micron and used in NVIDIA's powerful GPU accelerators.
It delivers high capacity per unit (up to 36 GB), eliminating the need to split a model's memory across multiple components, a remarkable bandwidth of 1.2 TB/s, and 30% greater energy efficiency compared to competing models.
--
👋 Hi, I'm Shlomo Strauss — follow me for more interesting content on science and technology.