---
title: "Samsung Unveils zHBM, AI Memory It Says Is 8x Faster. Here Is the Catch"
description: "Samsung has unveiled a new memory design called zHBM that it says can run about eight times faster than today's fastest AI memory. It showed the design at the Future of Memory and Storage (FMS) 2026 c..."
url: https://www.brandligo.com/blog/samsung-zhbm-ai-memory/
date: 2026-08-05
modified: 2026-08-05
author: "Brandligo Editor"
image: https://www.brandligo.com/blog/wp-content/uploads/2026/08/Samsung-zHBM-AI-memory-chip-at-FMS-2026.jpg
categories: ["AI News"]
type: post
lang: en
---

# Samsung Unveils zHBM, AI Memory It Says Is 8x Faster. Here Is the Catch

Samsung has unveiled a new memory design called zHBM that it says can run about eight times faster than today’s fastest AI memory. It showed the design at the Future of Memory and Storage (FMS) 2026 conference in Santa Clara on August 5, 2026, alongside a stack of other chips built for artificial intelligence.

Here is the honest version. zHBM is a concept model, not something you can buy and not a chip that ships this year. Samsung has not given a mass-production date for it.

So why does it matter? Because it shows where the whole AI industry is heading, and Samsung is one of the few companies that actually builds the memory every AI chip leans on. When the biggest memory maker on the planet says the next big jump comes from stacking memory in a new way, that is worth understanding, even if it is a couple of years out.

## What is zHBM, in plain English?

zHBM is a type of high bandwidth memory that Samsung wants to stack directly on top of an AI chip instead of beside it. High bandwidth memory, or HBM, is the fast memory that feeds the processor inside an AI accelerator, which is the specialized chip that runs AI models. Today those memory stacks usually sit next to the processor on the same slab of silicon. Samsung’s idea is to move the memory on top of the processor so data travels a much shorter distance, which is where a lot of the speed and the power savings come from. Samsung also says chipmakers can add their own custom circuitry into the design. If you want the wider picture, [Samsung Semiconductor](https://www.brandligo.com/brands/samsung-semiconductor) is the quiet part of the company that most people never see, and it makes the [DRAM and high-bandwidth stacks](https://www.brandligo.com/products/samsung-dram) that sit next to AI chips right now.

## How much faster is zHBM, actually?

Samsung says zHBM delivers roughly eight times the performance of HBM5, its next-generation memory, plus more than ten times the memory density. On top of that it claims about three times better energy efficiency and a cut in thermal resistance of more than half, which in plain terms means it should run cooler for the same work. Those are big numbers. They are also Samsung’s own figures for a concept, measured in Samsung’s own labs, and nobody outside the company has tested a shipping part because there is not one yet. Treat the eight-times claim as a direction of travel, not a benchmark you can rely on. The figures were reported by [Samsung’s own newsroom](https://news.samsung.com/global/samsung-unveils-next-gen-3d-memory-vision-at-fms-2026-charting-the-future-of-ai-infrastructure) and echoed by outlets including the [Korea JoongAng Daily](https://www.koreajoongangdaily.com/business/samsung-debuts-nextgen-memory-chips-for-ai-systemsnbsp/12809164).

## Why does AI need faster memory anyway?

AI chips spend a surprising amount of their time waiting for data to arrive from memory, so memory speed, not just raw computing power, now decides how fast an AI model actually runs. People call this the memory wall. The processor can do the math far quicker than the memory can hand it the numbers, and a large model has to move an enormous amount of data on every single step. That gap is the bottleneck. It is the reason HBM sits glued right next to the processor instead of across the board like the memory in a normal laptop. The models pushing this demand keep getting bigger and hungrier, from the ones behind [the latest frontier model launches](https://www.brandligo.com/blog/claude-opus-5-launch/) to [the everyday AI assistants](https://www.brandligo.com/blog/ai-assistant-pricing-2026/) millions of people now pay for. More capable models need more memory bandwidth. Simple as that.

## What is the catch with zHBM?

The catch is that zHBM is a concept, which means there is no release date, no price, and no independent benchmark yet. Samsung presented it as a preview of where its research is going, the same way carmakers show a concept car years before anything reaches a showroom. Stacking memory directly on top of a hot AI processor is genuinely hard. Heat has to escape somewhere, manufacturing yields have to hold up, and the whole thing has to survive being built at scale. Samsung showing zHBM now points to years of work ahead, not months. That is not a knock on the technology. It is just the difference between a roadmap and a product, and it is the part most quick summaries leave out.

## What else did Samsung announce at FMS 2026?

Besides zHBM, Samsung showed zNAND-O flash built for on-device AI, a V10 BV-NAND chip with more than 400 layers, and updates to its HBM4E and HBM5 memory. zNAND-O comes in four-layer and eight-layer versions aimed at edge AI, meaning AI that runs on the device in your hand rather than in a data center. V10 BV-NAND is the tenth generation of Samsung’s V-NAND flash, the storage tech inside SSDs and phones, and it packs roughly 58 percent more density than the previous V9 generation. HBM4E has been sampling to customers since May 2026 with a new cooling design, HBM5 appeared as a working model, and a low-power chip called LPDDR5X-PIM can even do simple calculations inside the memory itself. Here is how the pieces line up.

| Technology | What it is | Headline claim | Status |
| --- | --- | --- | --- |
| zHBM | HBM stacked on the AI chip | About 8x HBM5 performance | Concept |
| zNAND-O | Flash for on-device AI | 4 and 8 layer versions | In development |
| V10 BV-NAND | 10th-gen flash storage | 400+ layers, about 58% denser than V9 | Announced |
| HBM4E | Current AI memory, new cooling | Faster, cooler than HBM4 | Sampling since May 2026 |
| LPDDR5X-PIM | Low-power memory that computes | Runs math inside the memory | Shown at FMS 2026 |

You can browse Samsung’s wider hardware, from [V-NAND flash](https://www.brandligo.com/products/samsung-v-nand) to its [consumer and data-center SSDs](https://www.brandligo.com/products/samsung-ssds), on its directory listings.

## Where does this leave Samsung in the memory race?

Samsung held about 39 percent of the global DRAM market in the second quarter of 2026, and this roadmap is its pitch to stay in front as AI rewrites what memory has to do. That share figure comes from the Korea JoongAng Daily’s coverage of the event. The company is not alone in chasing AI memory, though. SK hynix and Micron are pushing hard on the same prize, and the lead has changed hands before. What FMS 2026 tells you is that Samsung intends to compete on architecture, not just capacity, by rethinking where the memory physically sits. Whether zHBM ends up first, second, or nowhere depends on execution over the next few years. You can see the full spread of what the company builds on [Samsung’s directory profile](https://www.brandligo.com/companies/samsung).

## Should regular buyers care?

Most of this is data-center technology, so you will not find zHBM inside a phone or a laptop, but the same research quietly trickles down to the devices you actually buy. zNAND-O is aimed squarely at on-device AI, the kind of feature that runs a translation or a photo edit on your handset without pinging a server. Faster, cooler, lower-power memory tends to show up later in phones, wearables, and cars, which is where a lot of Samsung’s other work lives too. The company that builds this memory is the same one that makes [the QD-OLED screen](https://www.brandligo.com/blog/samsungs-next-gen-display-qd-oled-technology/) you might be reading this on. So no, you do not need to act on this news. It is a signal about the plumbing behind the AI features creeping into everything you own.

## Quick questions

**Is zHBM available to buy now?** No. It is a concept model Samsung previewed at FMS 2026, with no price and no ship date.

**Is it really eight times faster than HBM5?** That is Samsung’s own figure for a concept compared with HBM5, taken in its labs. No independent test exists yet, so treat it as a target rather than a proven result.

**How is zHBM different from normal HBM?** It stacks the memory on top of the AI chip instead of next to it, which shortens the distance data has to travel and, Samsung says, saves power.

**When will it ship?** Samsung has not said. Concept memory like this usually sits years away from mass production.

## The takeaway

The number to watch is not eight times faster. It is the mass-production date Samsung has not given yet. Until the company commits to one, zHBM is a lab result and a statement of intent, both of which matter but neither of which puts a chip in a server. The near-term parts that will actually touch the AI tools you use are the less flashy ones on the roadmap, HBM4E and HBM5, which are close to real. Keep an eye on those. And remember that in AI, the race is increasingly being won or lost on memory, not just on the processors that get all the headlines.

**About this article.** Written and fact-checked by the Brandligo editorial desk. AI tooling was used to gather and cross-check sources; every fact and figure here was verified against the primary sources linked above, including Samsung’s own newsroom, before publication. Published August 5, 2026. If you spot something out of date, tell us at info@brandligo.com.