Lexar unveiled an AI‑focused SSD that offloads portions of large language models onto NAND flash, trimming the DRAM a PC needs. CTO Daniel Guo notes that DRAM costs roughly six times more than flash, so shifting work to storage cuts expense and power. The drive pairs custom storage‑processing hardware with software to act as a DRAM‑less AI accelerator, making it attractive for gamers and creators alike.

In internal benchmarks Lexar ran the 122‑billion‑parameter Qwen 3.5 model on a system with only 32 GB of RAM, a task that normally demands a $4,500 build with 128 GB. The AI suite delivered 15.6 tokens per second versus 5.2 using conventional frameworks, and maintained performance even with 256K‑token contexts where standard stacks failed. The test also showed stable throughput across varying batch sizes.

Lexar’s hardware lives in a hot‑swappable M.2 module that fits a 25 mm front‑panel slot, available in PCIe Gen 5 and Gen 4. Its custom SPU controller manages data movement without DRAM, shaving up to 40% of memory demand in certain workloads. While latency rises on very large models, the solution offers a viable path for AI‑centric PCs without costly memory upgrades. Early adopters can expect firmware updates to broaden model compatibility.