NVIDIA Collabs with SK Hynix For Design SSD That AI Focused

NVIDIA is teaming up with SK Hynix from South Korea to create a specialized SSD designed specifically for AI tasks. Phison Electronics is joining them in this development effort, too.

These new AI-focused SSDs are projected to deliver performance that’s over ten times better than what’s currently being used in AI servers – way beyond what today’s enterprise SSDs can handle.

Here’s why this matters: AI models keep getting bigger, and when they’re running (the inference stage), they need instant access to huge amounts of data. Regular high-bandwidth memory and DRAM just can’t keep up anymore with the speed and volume requirements. These AI SSDs would basically sit in a sweet spot between memory and storage, acting like an extra memory layer to help AI systems work better.

SK Hynix is aiming to show off a working prototype by late 2025, with plans to bring it to market sometime around 2027. This is being viewed as an important milestone in improving AI processing architecture.

It’s worth noting that NVIDIA has already started integrating GDDR7 memory into the pre-fill phase of its upcoming RubinC PX GPUs, showing how they’re rethinking the entire AI computing setup.

But there’s a potential problem brewing. Right now, cloud providers and major AI companies are already ordering storage devices like crazy, which is putting serious strain on NAND flash memory supplies.

If these AI SSDs take off and get widely adopted, we might see NAND shortages similar to what happened with DRAM.

Experts are warning that the infrastructure costs for AI are climbing fast, which is shaking up the semiconductor supply chain. DRAM prices are already going up steadily, and many in the industry think NAND supply and demand could be the next big concern to watch.