MemryX AI unveiled its next‑gen accelerator, MX3+, on Nov 10, 2026. The chip promises up to 10 × higher peak AI compute and nearly 10 × more on‑chip memory than the existing MX3, without increasing power use. It can be linked with up to 15 other MX3+ chips, forming a 16‑chip system that exceeds one peta‑operation per second (1 POPS) of compute. Engineering samples are expected in late 2027, with volume production aimed for mid‑2028.
Why it matters
Customers in robotics, industrial automation and edge video can run larger vision and transformer models without raising power budgets.
