Qualcomm Technologies has unveiled its latest AI inference-focused solutions for data centres in the form of the AI200 and AI250 accelerator cards and corresponding rack systems.

The company targets commercial availability of the AI200 in 2026, followed by the AI250 in 2027.

Both systems are intended for large language model (LLM) and multimodal inference as well as broader AI workloads.

The AI200 system features a rack-level design and utilises individual cards supporting 768GB of low-power double data rate synchronous dynamic random-access (LPDDR) memory, designed to enable larger models and increased workload flexibility.

Qualcomm reported that the AI250 will incorporate near-memory computing, offering more than ten times the effective memory bandwidth compared to prior generations while lowering power draw.

Both platforms employ direct liquid cooling, PCIe-based scale-up, and Ethernet-based scale-out connectivity.

The racks are specified to operate at up to 160kW. Confidential computing is also included to meet secure workload requirements.

Qualcomm’s software stack is engineered to support integration with standard machine learning (ML) frameworks and inference engines, as well as generative AI (gen AI) and large language model serving methods such as disaggregated inference.

Developers can onboard models from sources such as Hugging Face and deploy them using Qualcomm’s tools, including its Efficient Transformers Library and Inference Suite.

Qualcomm senior vice president and technology planning, edge solutions and data centre general manager Durga Malladi said: “With Qualcomm AI200 and AI250, we’re redefining what’s possible for rack-scale AI inference.

“These innovative new AI infrastructure solutions empower customers to deploy generative AI at unprecedented TCO, while maintaining the flexibility and security modern data centers demand.”

In parallel to these product launches, Qualcomm Technologies has announced a partnership with HUMAIN to deploy its data centre hardware in Saudi Arabia.

The project aims for an initial deployment of 200 megawatts of Qualcomm AI200 and AI250 rack capacity from 2026 onwards.

According to both firms, this effort is intended to establish edge-to-cloud hybrid AI inference infrastructure ahead of the ninth Future Investment Initiative conference and follows a May 2025 announcement made at the US–Saudi Investment Forum.

The collaboration combines HUMAIN’s regional infrastructure capabilities with Qualcomm’s semiconductor technology stack to deliver inference services for enterprises and government agencies within Saudi Arabia and globally.

HUMAIN CEO Tareq Amin said: “With Qualcomm’s world-class AI infrastructure solutions, we’re shaping the foundation of the Kingdom’s AI future.

“This collaboration unites HUMAIN’s deep regional insight and unique full AI stack capabilities with Qualcomm’s unmatched semiconductors technology and product leadership.”