# WiCi: Connecting GPUs over WiFi > Source: > Published: 2026-11-01 01:29:51+00:00 # Frontier AI belongs to everyone. WiCi is a systems research lab. We build systems for wireless GPU and local AI. WiCi is a systems research lab. We focus on wireless GPU computing and local inference systems for frontier AI. We develop systems for sharing compute across devices while keeping models and data under local control. Local compute. Direct control. Running AI locally gives individuals and organizations control over their models, data, and usage. WiCi makes local deployment more practical through wireless GPU access, efficient inference, and shared compute. ## Systems for Personal Frontier AI. Our research addresses the constraints that make local AI difficult: network latency, limited memory, competing workloads, and real-time response. GPU software makes thousands of small calls that become expensive over Wi-Fi. Our runtime reduces network overhead through caching, data reuse, and asynchronous execution, keeping applications on the device while GPU computation runs nearby. [Making a Nearby GPU Feel Local (opens in a new tab)](article/nearby-gpu-feel-local) Large models can exceed both GPU memory and system RAM. InferSSD, our open-source engine, keeps expert weights on SSD and streams the working set to the GPU, so larger models run on personal hardware. [Running 800B-Parameter LLMs on Consumer Hardware with InferSSD (opens in a new tab)](article/running-800b-models-with-inferssd) Independent applications compete for GPU memory and often load duplicate model weights. Our scheduling research focuses on reusing those weights and managing memory across workloads while keeping application state separate. [The GPU Never Learned to Share (opens in a new tab)](article/gpu-never-learned-to-share) Responsive AI requires coordinating input, inference, and response timing. Our voice research forms candidate answers while speech is arriving and uses a learned timing policy to decide when to respond. [Answering Is Not Buzzing (opens in a new tab)](article/answering-is-not-buzzing) ## Research & Ideas [View all research & writing](research) Research and perspectives on the systems behind local AI. ### The General and the Particular Why interaction, memory, and adaptation belong on local infrastructure. (opens in a new tab) - [**Making a Nearby GPU Feel Local** How the wireless runtime reduces network overhead when applications use a nearby GPU. (opens in a new tab)](article/nearby-gpu-feel-local) - [**Running 800B-Parameter LLMs on Consumer Hardware with InferSSD** How InferSSD streams expert weights from SSD, with results on Apple Silicon and NVIDIA systems. (opens in a new tab)](article/running-800b-models-with-inferssd) - [**Answering Is Not Buzzing** How early listening, answer generation and speech timing reduce voice interaction latency. (opens in a new tab)](article/answering-is-not-buzzing) ## The first wireless GPU Introducing WiCi One Built for homes, studios, and workplaces. **From US$2,599.** **Powerful local AI for everyday devices.** WiCi One combines wireless GPU computing and SSD inference in one local system. Run advanced models, share compute across applications and devices, and keep your data on your network. Demo films In the press ## Work on the systems behind personal AI. Join researchers and engineers working across networking, GPU systems, and AI inference. Develop new methods and turn them into working systems. [Explore open roles (opens in a new tab)](careers) ### Build with WiCi. Access local GPU compute through familiar APIs, SDKs, and supported driver integrations. Bring AI inference into assistants, creative tools, and connected devices. [Explore developer tools (opens in a new tab)](developers)