The outfit insists this makes it the first in the world to move the next-generation tech into full production.

The new chips boast more than 10 Gbps operating speed, outstripping the JEDEC standard of 8 Gbps. They double the input-output count from the previous generation to 2,048 terminals, while SK hynix claims a 40 per cent improvement in power efficiency.

The firm reckons performance gains of up to 69 per cent for AI services are possible when HBM4 is deployed, which it says will ease data bottlenecks and cut data centre power bills.

Joohwan Cho, head of HBM development at SK hynix said the new memory was “a new milestone for the industry” and promised that the company would “fulfill time to market and maintain a competitive position” by delivering promptly on performance and reliability.

SK hynix president and head of AI infra Justin Kim said the launch was “a symbolic turning point beyond the AI infrastructure limitations” and insisted HBM4 would be a “core product for overcoming technological challenges.”

He added that the company aims to grow into “a full-stack AI memory provider” as demand continues to surge.

SK hynix has paired the chips with its advanced MR-MUF packaging process, which it says has already proven reliable, and its fifth generation 10-nanometre-class process, labelled 1bnm, to limit risk in high-volume manufacturing.

The company is pitching the launch as proof it can lead the AI memory market. What remains to be seen is whether customers agree that HBM4 really is the breakthrough that SK hynix claims.