Samsung and SK hynix develop CXL memory to accelerate AI

Technology Jul 20, 2026

Samsung and SK hynix are developing new memory solutions designed to make AI run faster and smoother.

Using Compute Express Link (CXL), their tech lets servers expand memory without needing more processors, so CPUs and GPUs aren't overloaded.

This means bigger, more efficient memory pools for real-time AI tasks: no pricey upgrades needed.

Samsung 1TB CXL hits 92% DRAM

SK hynix showed off a 256GB CXL module in June, doubling its previous capacity, while Samsung tested a massive 1TB CXL memory pool that hit 92% of traditional DRAM performance across eight GPUs.

Experts say this could seriously boost speed for demanding AI jobs.

With AI getting bigger every year, analysts expect CXL-powered servers to take off by 2026 and become the norm by 2028.