Qualcomm has officially announced the rollout of its new - generation AI inference optimization solutions tailored for data centers. These encompass accelerator cards and rack systems that are built upon the Qualcomm AI200 and AI250 chips. These cutting - edge solutions deliver rack - scale performance and boast remarkable memory capacity, specifically optimized for generative AI inference in data centers, all while keeping the total cost of ownership at a low level.

The AI200 is purpose - built for rack - level AI inference. Its accelerator cards are capable of supporting a massive 768GB of LPDDR memory, providing ample space for complex AI computations.

On the other hand, the AI250 features an innovative near - memory computing memory architecture. This unique design significantly boosts memory bandwidth and, at the same time, cuts down on power consumption, making it an energy - efficient option for data center operations.

Both of these rack solutions are compatible with direct liquid cooling technology. In terms of power consumption, the entire rack can draw up to 160 kilowatts of power, ensuring high - performance operations without excessive energy use.

Looking ahead, the AI200 is anticipated to hit the commercial market in 2026, while the AI250 is expected to follow suit in 2027.