SK hynix has built its own tiny processor for SSDs, and the company says it uses just one twenty-fifth of the power of a normal CPU core. The chip, called H-TPU, was revealed at the ISOCC 2026 conference and is not an AI chip: its job is to control how data moves inside a solid-state drive.
The details come from a post by tipster Abhishek Yadav, citing a report from ZDNet Korea. Here is what H-TPU does and why it matters for AI data centres. More hardware news is on our SK hynix tag page, and the company is at SK hynix.
SK hynix H-TPU: Key Facts
| Feature | Detail |
|---|---|
| Name | H-TPU |
| Revealed at | ISOCC 2026 |
| Purpose | Controls data movement inside an SSD (not an AI chip) |
| Size | About 1/25th the space of a regular CPU core |
| Power use | About 1/25th the power of a regular CPU core |
| Design | Dozens of small chips work together, each doing one job |
| Power saving | Unused parts go into low-power mode |
| Target | AI data centres and upcoming HBF memory |
What Is H-TPU?
Every SSD has a controller, a small processor that decides how data is read, written and moved. H-TPU is a controller-focused chip that takes about one twenty-fifth of the space and power of a normal CPU core. Instead of using a few large cores, SK hynix uses dozens of tiny units that each handle a single task. Parts that are not in use drop into a low-power mode.
Why SSD Controllers Are a Bottleneck
The post explains that GPUs and PCIe speeds are doubling with each generation, and SSD controllers are struggling to keep up. A fast GPU is useless if storage cannot feed it data quickly. By splitting the work across many small, specialised chips, SK hynix hopes to remove that bottleneck without raising power use.
The Link to HBF Memory
H-TPU is built for AI data centres and for the upcoming HBF memory, which is described as fast flash storage for AI. HBF is a new type of memory that aims to give AI systems far more capacity than current high-bandwidth memory. Controllers like H-TPU would help manage that flash efficiently.
Does It Matter for Everyday Users?
Not directly, at least for now. H-TPU is aimed at data centres, not consumer laptops or phones. Still, ideas from data centre chips often reach consumer products later, and lower power use is useful everywhere. In the long run, faster and more efficient storage could make AI services cheaper to run.
What Is Not Confirmed
The post does not give a release date, customer names, manufacturing process or performance numbers beyond the 1/25th size and power claims. These are company-reported figures from a conference presentation, and they have not been independently tested. We do not know if H-TPU will appear in consumer SSDs. The background on the company is on Wikipedia.
Details in early reports and social posts can change as companies publish official information, so please treat the figures above as reported at the time of writing. We will update this article when new or corrected details arrive. Check the official pages linked here before you buy or plan around anything mentioned, and remember that offers, dates and prices may differ by region, seller and time of purchase.
The takeaway: As AI grows, storage becomes as important as compute. A chip that cuts power and space in the storage path could reduce running costs in large data centres, where electricity is a major expense. We will follow up if SK hynix shares more about customers or release timing.
FAQs:
What Is H-TPU?
Every SSD has a controller, a small processor that decides how data is read, written and moved. H-TPU is a controller-focused chip that takes about one twenty-fifth of the space and power of a normal CPU core. Instead of using a few large cores, SK hynix uses dozens of tiny units that each handle a single task.
Why SSD Controllers Are a Bottleneck?
The post explains that GPUs and PCIe speeds are doubling with each generation, and SSD controllers are struggling to keep up. A fast GPU is useless if storage cannot feed it data quickly. By splitting the work across many small, specialised chips, SK hynix hopes to remove that bottleneck without raising power use.
Does It Matter for Everyday Users?
Not directly, at least for now. H-TPU is aimed at data centres, not consumer laptops or phones. Still, ideas from data centre chips often reach consumer products later, and lower power use is useful everywhere. In the long run, faster and more efficient storage could make AI services cheaper to run.
How efficient is it?
About 1/25th the power and space of a CPU core, according to SK hynix.
