At SEMICON Taiwan on September 1, Samsung Electronics laid out its next generation of memory products under the banner CUBE, disclosing specific performance targets for the first time: HBM5, the next version of its high-bandwidth memory, is designed to deliver twice the performance of HBM4E, while a new stacked memory technology called zHBM is planned for an eightfold improvement. The company also showed its roadmap for 3D integration and expanded capacity, in the clearest statement yet of how it intends to close the gap with SK hynix.
The presentation was aimed as much at the investment community as at the semiconductor engineers in the audience. Samsung has lost its long-held position as the leader in high-bandwidth memory, the chips that sit beside AI accelerators and have become the most valuable commodity in the semiconductor industry. SK hynix, its Korean rival, dominates the market, and the gap has been a persistent source of frustration inside Samsung, which has been late to successive generations of the technology. The CUBE strategy is an attempt to reset the competition on terms Samsung believes it can win.
HBM5 is the centerpiece. The company’s target, double the performance of HBM4E, implies a memory chip with significantly higher bandwidth, the measure that matters most for AI workloads, which move enormous volumes of data between memory and processors. Samsung has not disclosed when HBM5 will enter production, but the roadmap shown in Taipei suggests the company is positioning for the next wave of AI chips, which will demand faster memory to feed faster processors. The zHBM technology, with its eightfold target, points further ahead, toward a generation of memory that stacks chips in three dimensions to multiply capacity and speed.
The strategy reflects a broader shift in the memory business. For decades, memory makers competed on density and cost, selling interchangeable chips in a commodity market. The AI boom changed the rules: HBM is a differentiated product, engineered in close partnership with the chip designers who buy it, and the winner of each generation gains a durable advantage in both market share and pricing power. SK hynix understood this earlier than its rivals and has reaped the rewards, with its HBM business driving record profits. Samsung’s CUBE roadmap is an acknowledgment that the old playbook no longer works.
The competitive stakes extend beyond the two Korean companies. Micron, the third major memory maker, has also invested heavily in HBM, and the technology’s importance has made the memory industry a matter of national interest for both the United States, where Micron is based, and South Korea, where Samsung and SK hynix dominate. The AI boom has turned a commodity business into a strategic one, and the performance targets Samsung disclosed in Taipei are part of a contest that regulators and governments are watching as closely as investors.
The 3D integration plans Samsung showed are aimed at the same prize from a different direction. By stacking memory dies vertically, chipmakers can increase capacity and bandwidth without expanding the footprint of the package, a constraint that matters as AI systems demand ever larger amounts of memory in ever tighter spaces. Samsung has been developing the technology for years, and the CUBE presentation was an attempt to show that its work is advanced enough to matter in the race with SK hynix, which has its own 3D roadmap.
The timing of the announcement, at a trade show where the industry’s engineers and buyers gather, was deliberate. Samsung has spent the past year rebuilding credibility with the AI companies that buy HBM, and the CUBE presentation was a chance to show customers, in concrete numbers, what the company will deliver next. Whether the targets hold will depend on Samsung’s execution, which has been the company’s weakness in recent memory generations, but the specificity of the numbers was itself a statement of confidence.
The presentation also served a commercial purpose. Samsung’s memory division has been working to reassure the AI companies that buy HBM that it can supply the volumes they need, after supply constraints earlier in the cycle pushed some customers to lock in agreements with SK hynix. The CUBE strategy, with its clear targets and aggressive timeline, is part of that effort, a signal that Samsung intends to be a supplier of choice for the next generation of AI systems rather than a second source.
Analysts said the memory race has entered its most consequential phase. The next generation of HBM will be designed into the AI systems that define the rest of the decade, and the supplier that wins the design wins a multi-year revenue stream. SK hynix enters the phase as the leader; Samsung enters with the ambition of a challenger and the resources of a conglomerate. The gap between them, analysts said, will be closed not by speeches but by production yields, and the CUBE strategy will be judged by whether the performance targets Samsung announced in Taipei survive contact with the factory floor.
