WiCi One Puts an RTX 5060 Ti Behind Wi-Fi 7

WiCi has detailed an unusual external GPU system called WiCi One that replaces the usual Thunderbolt, USB4 or OCuLink cable with a Wi-Fi 7 connection. The base configuration combines an NVIDIA GeForce RTX 5060 Ti with 16GB of VRAM, a 4TB PCIe 5.0 NVMe SSD and an Intel Core Ultra 7 255H inside a standalone compute box.

The important part is the software model. WiCi says its Driver mode presents the remote hardware as a virtual GPU to supported workloads, while separate SDK and API modes target AI applications. The company lists Windows, Linux and macOS for its virtual-GPU software and says a Developer Preview is planned for Q4 2026.

A GPU does not behave like ordinary network storage. CPU and GPU software can generate large numbers of short, latency-sensitive interactions, while a directly attached PCIe device operates on a radically lower-latency link than Wi-Fi. Simply putting GPU traffic onto a fast wireless network would therefore be a poor substitute for PCIe.

WiCi’s proposed answer is to reduce how often those interactions become network round trips. Its technical material describes caching, deduplication, batching, streaming and trace replay, and says an internal experiment reduced roughly 20,000 driver calls to about 800 network round trips. WiCi cites around 1.5 ms round-trip latency over Wi-Fi 6 in that work. Those techniques make the architecture technically interesting, but they also mean performance can vary substantially with workload behavior rather than scaling from Wi-Fi bandwidth alone.

WiCi is positioning the device primarily as shared local AI compute. Its TurboStream design uses the onboard NVMe SSD as a backing tier for model weights that do not fit in GPU VRAM. That does not turn SSD capacity into VRAM in the conventional performance sense: moving weights from storage is far slower than accessing GDDR memory, so useful throughput depends on how effectively the runtime can stage and reuse model data.

The networked design also allows one box to serve multiple nearby devices without putting a discrete GPU into every laptop or workstation. For inference and other jobs that can keep most GPU-side state resident, that is a more plausible fit than workloads that constantly transfer large amounts of data back and forth.

WiCi says games run on the client machine rather than through conventional video streaming and lists Cyberpunk 2077 and Black Myth: Wukong among its tested workloads. That distinction matters: if Driver mode really exposes a usable GPU interface to an unmodified game, the system is closer to GPU remoting than services such as cloud gaming or Steam Remote Play.

It is also where independent testing will matter most. Games combine rendering, resource uploads, shader compilation, synchronization and CPU-GPU communication in ways that can expose latency quickly. Compatibility across anti-cheat systems, graphics APIs, frame pacing and different Wi-Fi environments cannot be inferred from a small vendor benchmark set. Until the Q4 preview and third-party tests arrive, WiCi One is best understood as an ambitious local-compute architecture rather than a proven wireless replacement for a directly attached gaming GPU.