Intel wants Linux to move data by cable without the network stack

Intel’s USB4STREAM belongs to the class of Linux changes that will not sell laptops from a keynote slide, but could remove real friction from repair benches, labs and developer setups. According to Phoronix, Intel is preparing a new protocol for the upcoming Linux 7.2 kernel that would use USB4 and Thunderbolt links as a simple way to move raw packets from one host to another.

The important distinction is that USB4STREAM is not just another networking feature with a different label. The supplied description frames it as a low-level channel that can avoid the traditional Linux networking stack when that stack is unnecessary, awkward to configure, or undesirable for the job. If two machines are physically connected with a USB4/Thunderbolt cable, the protocol can act as a direct packet stream between the hosts.

That makes the feature more practical than glamorous. The named use cases are specific: quickly backing up a system from one host to another, sharing a webcam or other peripherals across systems, and working in environments where avoiding full network configuration is part of the value. Since USB4 is already designed as a multi-protocol transport layer over the same connector family, a host-to-host path is not decorative marketing. It is a direct use of transport capability already present on modern machines.

Linux already documents its USB4 and Thunderbolt support, but USB4STREAM narrows the job to one exact operation: move raw packets between two hosts without putting the network stack in the middle. That is plain engineering, but plain engineering often becomes valuable once somebody turns it into a usable layer. Cable-based backup, recovery workflows and controlled peripheral sharing make more sense on a service bench than in a consumer demo reel.

The feature should not be inflated beyond the available source material. There is no supplied claim of a polished desktop application, consumer interface, benchmark result, speed record or broad industry standard around USB4STREAM. For now, this is kernel-level plumbing. It will matter first to people who already understand why normal networking is the wrong shape for a particular task: administrators, service technicians, developers and data recovery tool authors.

Intel’s role still matters. The company has been central to the Thunderbolt ecosystem, and the official Thunderbolt technology site has long positioned the connector family as a combined channel for data, displays and peripherals. USB4STREAM fits that direction, but with a Linux tone: less branding, more explicit control over what actually moves through the cable.

If USB4STREAM lands in Linux 7.2, the useful question will not stop at whether the protocol can move packets. The real test is who turns it into a workflow above the kernel: USB4 cable backup, webcam sharing without network setup, or forensic transfer when higher system layers are inconvenient. The kernel can open the channel. Practical value arrives only when someone builds a tool that does not require every step to be explained by hand.