Samsung Agrees to Use Taiwanese Technology for US Chipmaker NVIDIA

Samsung Electronics is reportedly set to use the 3-nanometre production process of Taiwan’s TSMC to manufacture base dies for the latest generation of high-bandwidth memory (HBM) chips for NVIDIA.

The move would mark the first production collaboration between Samsung and TSMC in memory, according to a report by the Seoul Economic Daily cited by Reuters on Monday (19 October 2026).

The request came directly from NVIDIA. The graphics-chip maker asked Samsung, SK Hynix and Micron Technology to use TSMC’s manufacturing process for the base die of NVHBM, alongside sixth-generation DRAM technology for the core.

Introduced in August, NVHBM incorporates a memory controller designed by NVIDIA into the base die, allowing the company to standardise its implementation across suppliers.

Samsung has agreed to the terms following an internal review. However, the South Korean company will retain its own production lines for standard HBM and products tailored to other customers. It is also assessing whether its own foundry facilities could produce NVHBM base dies if demand exceeds TSMC’s available capacity.

The decision highlights the pressure on memory-component suppliers to meet standards set by major customers as demand for AI chips surges. Even a fully integrated manufacturer such as Samsung is having to work with a rival to avoid losing customers’ confidence.

What is HBM?

HBM, or high-bandwidth memory, is an advanced type of memory designed to transfer data at exceptionally high speeds, far beyond the memory used in ordinary phones or computers. If conventional memory such as DDR is like a standard water pipe, HBM is more like a large conduit carrying thousands of streams at once.

HBM’s distinguishing feature is that memory chips are stacked vertically in several layers and connected directly to the main chip in a tightly integrated package. This shortens the distance data must travel, increasing speed, reducing power consumption and keeping the overall form factor compact.

Why is HBM important for AI?

AI technology processes enormous quantities of data, including billions of parameters, images, text and patterns that must be calculated almost instantly. The limiting factor is no longer how quickly a chip can compute, but how quickly data can be delivered to it. This is where HBM is crucial:

• Dozens of times faster: Current HBM can transfer hundreds of gigabytes of data per second, far exceeding standard memory. Greater bandwidth allows AI systems to process and respond to more information in an instant.

• More energy-efficient: Because data does not have to travel long distances back and forth, power consumption falls significantly. This matters because large AI chips can use as much electricity as hundreds of households.

• Helps AI chips reach their full potential: Without adequate HBM, even powerful chips are left waiting for data, like a racing car stuck in traffic. The performance of chips from NVIDIA, AMD and Google depends on sufficient HBM supplies.

HBM has become one of the most sought-after and valuable components in the technology supply chain because supply is falling short of demand from the AI industry. Major companies are holding back data-centre expansion because they cannot secure enough HBM. The latest AI chips cannot be purchased without guaranteed accompanying HBM supplies.

As a result, HBM prices have risen sharply. Demand far outstrips manufacturing capacity, pushing prices up several-fold and making the component a major share of the total cost of an AI chip.

Only a handful of companies can make HBM. The technology for stacking memory chips with extreme precision is difficult to master. Samsung, SK Hynix and Micron are currently the only companies capable of producing HBM at commercial scale. Dependence on these three companies is so great that even a minor disruption at one of their factories could halt AI projects around the world.

This is why NVIDIA is setting production standards for NVHBM and specifying the technology its suppliers must use. Whoever controls HBM supply has influence over the pace of global AI development.