Ragu Sriramaneni, a Micron HBM design architecture expert, said the memory wall still exists and may be getting worse. According to Sina Finance, he made the remarks on the 23rd at the Hot Chips 2026 semiconductor conference held at Stanford University in the United States.

Sriramaneni said AI chip computing performance has been rising about threefold every two years, while HBM performance has increased by less than twofold over the same period. He also said Micron's recently mass-produced HBM4 reflects these challenges, with a single Cube bandwidth of up to 2.8TB per second, and that Micron is supplying the product for Nvidia's next-generation AI chip, Vera Rubin.