Qualcomm Unveils Mixed-Reality Platform for AI-Driven Spatial Computing

Qualcomm Technologies has announced Snapdragon Reality Elite, a next-gen extended reality (XR) platform designed to run generative AI directly on headsets and glasses. Unveiled at the Augmented World Expo in California, the chipset targets high-performance video-see-through headsets and lightweight, tethered optical-see-through devices. The device will first ship this fall in XREAL's Project Aura glasses.

Qualcomm, whose XR silicon powers most non-Apple standalone headsets, is pitching the platform to device makers seeking greater performance, more on-device intelligence, and improved power efficiency in wearable form factors.

The new platform runs large vision models (LVMs) and large language models (LLMs) directly on‑device. Image used courtesy of Qualcomm

The naming breaks with Qualcomm's usual XR2 convention, marking a purpose-built flagship rather than an iteration of the prior generation. The company says it designed the part from the ground up to span multiple form factors, whether the silicon sits in the headset itself or in a tethered compute puck, and across passthrough and see-through display systems.

On-Device AI For Generative XR

Snapdragon Reality Elite delivers up to 48 TOPS of AI processing, which Qualcomm says is a 160% increase in NPU performance over the Snapdragon XR2+ Gen 2. That headroom is aimed at large language models (LLMs) and large vision models (LVMs) running locally, enabling experiences such as photorealistic avatars built with Gaussian splatting, LLM-based agents, and real-time object generation. 

Qualcomm says a three-billion-parameter LLM can run on-device at 45 tokens per second, while a 512 x 512 large vision model returns results with roughly 1.7 seconds of latency. CPU performance rises to 30% and GPU performance up to 60% over the prior flagship—gains that give developers more rendering and compute headroom to pair high-resolution graphics with the larger on-device models.

Features of the platform's architecture. Image used courtesy of Qualcomm

A reworked Engine for Visual Analytics (EVA) block, hardened in silicon, offloads computer vision workloads, such as 3D environment reconstruction, from the main processing units. Qualcomm indicated that the expanded block could, from a chipset standpoint, support continuous scene meshing on headsets lacking a dedicated depth sensor, though it left specific implementations to developers. Moving those tasks to fixed-function hardware frees the GPU and NPU for rendering and AI inference, a trade that becomes more important as headsets attempt richer mixed-reality scenes without adding sensors or power budget.

Sharper Visuals and Faster Passthrough

The platform supports visuals up to 4.4K per eye at 90 frames per second, which Qualcomm says sharpens detail, improves color fidelity, and smooths motion across both passthrough and see-through displays. Enhancements to video see-through cut photon-to-photon latency by about 10% and reduce passthrough power draw by roughly a third, with added image noise reduction meant to blend digital content into the physical world. 

Running models locally also keeps latency-sensitive tracking and generation off the network, which matters for interactions that must respond within a single frame to avoid breaking immersion. The chipset also adds support for UFS 4.0 storage, 4.2-GHz memory (up from 3.2 GHz), up to two USB 3.1 ports, and Bluetooth 6.0, with the faster storage and memory feeding the larger models and higher-resolution buffers.

Power and Thermals for Wearables

Qualcomm claims up to 20% longer battery life at the same workload as the XR2+ Gen 2 and operation up to 12℃ cooler under load. The cooler, more efficient profile is what the company says makes the chip viable in lighter headsets and in tethered compute pucks a user might carry in a pocket. 

Thermals matter more in glasses than in bulky headsets, since a slimmer enclosure has less room for heat spreaders and active cooling and sits closer to the wearer's skin. Asked whether pocket placement would throttle performance compared with an actively cooled headset, Qualcomm declined to answer directly, saying thermal design is the responsibility of device makers.

XREAL's Project Aura, an optical-see-through glasses product running on Snapdragon Reality Elite, is confirmed to ship this fall, with the chipset housed in the device's tethered compute puck. Play For Dream has said it will use the platform in an upcoming flagship device, and Qualcomm expects additional products to follow. The company points to more than 60 million XR devices already in the market as the installed base its new platform is meant to build on.