The news dropped during SK hynix’s second-quarter earnings call, where the company bragged about better-than-expected DRAM and NAND shipments. Its 12-Hi HBM3e DRAM is shifting well, fuelling one of its strongest quarters in a while.

The real tease, though, is the new GDDR7. These 24 Gb (3 GB) chips offer 50 per cent more capacity than today’s 16 Gb (2 GB) dies. With speeds heading beyond 30 Gbps per pin and 40+ Gbps still some way off they’ll let GPU makers like Nvidia cram more memory onto fewer modules, easing bandwidth and density limits.

Samsung is pushing similar modules, which have been spotted floating around online. Nvidia's upcoming RTX 50 SUPER cards are expected to be among the first to use this higher-capacity GDDR7, likely landing later this year or early next. Expect AI workloads to hoover up this tech as well, since higher VRAM is practically mandatory for large models.

SK hynix is doubling down on AI and HPC, preparing to ship GDDR7 modules for AI GPUs while ensuring HBM4 stays on schedule for monster accelerators. Initial HBM4 supply started last month for evaluation, with Nvidia's Rubin and AMD’s MI400 expected to be the first adopters.

The company says it will double its HBM output year-on-year and keep pace with customer demand, making sure it doesn’t lose ground to Samsung or Micron in the high-stakes AI memory race.

With GDDR7 boosting mainstream GPUs and HBM4 pushing the bleeding edge of AI accelerators, the next 12 months will be a feeding frenzy for VRAM-hungry hardware.