SK hynix is widening its AI-memory roadmap beyond HBM, presenting a new set of technologies aimed at the data movement and memory bottlenecks inside large AI systems.

At AI Infra Summit 2026 in Santa Clara, the memory maker highlighted High Bandwidth Flash, processing-in-memory technology and a new SALT-KV architecture alongside enterprise SSDs. The common target is the same: keep accelerators supplied with data as AI models and inference workloads grow.

AI infrastructure is becoming a memory problem

Modern GPUs can process enormous workloads, but their performance depends on moving model weights, key-value cache data and other information fast enough to keep the processors busy. SK hynix says future AI infrastructure will need a wider spectrum of memory and storage rather than relying on faster compute alone.

The embedded English-language explainer provides background on the AI-memory capacity pressure affecting SK hynix, Micron and the broader semiconductor supply chain.

High Bandwidth Flash targets the gap between HBM and storage

SK hynix’s High Bandwidth Flash concept is designed to bring NAND flash closer to AI accelerators with substantially greater parallelism than conventional storage. It is one part of an effort to create additional tiers between extremely fast HBM and much larger but slower storage.

The company also highlighted PIM and AiMX approaches that move some computation toward memory itself. That can reduce repeated transfers between processors and memory, potentially lowering latency and energy consumption for selected AI workloads.

SALT-KV attacks the inference cache bottleneck

SK hynix also presented SALT-KV, a system aimed at managing the key-value cache created during large-language-model inference. KV cache can become a major memory-capacity and data-movement burden as context windows and concurrent users grow.

The company’s strategy shows how the semiconductor race is expanding beyond GPU arithmetic. HBM, flash, SSDs, networking and memory-management software increasingly determine how much useful AI work a data center can extract from its accelerators.

AI memory keeps moving deeper into the data center

SK hynix presented the technologies during the September 15–17 AI Infra Summit 2026. The company described HBF, PIM/AiMX, SALT-KV and enterprise SSD technology as parts of a broader “New Spectrum” for AI infrastructure.

Sources: SK hynix Newsroom, AI Infra Summit 2026 materials and the embedded English-language semiconductor explainer. Product performance and roadmap claims should be treated as manufacturer targets until independently benchmarked in production systems.

BitcoinVersus.Tech

Advertisement: Follow BitcoinVersus.Tech for independent coverage of Bitcoin, mining hardware, artificial intelligence, robotics, data centers, semiconductors, energy, gaming and emerging technology.

BitcoinVersus.Tech Editor’s Note & Disclaimer:
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. BitcoinVersus.tech provides news, technical analysis and educational information and does not provide financial or investment advice. Verify manufacturer specifications and roadmap claims independently before making purchasing, deployment or investment decisions. If you would like to support to help further secure the integrity of our research initiatives, please donate here: 3C9o19EH5HSiwEPyCTmEKzxhNCbo2X6TTb