VLC Creator Launches Kyber to Solve the Latency Bottleneck in Physical AI

Artificial intelligence has spent the last decade mastering the digital realm, generating text, writing code, and rendering images with unprecedented speed. But as the technology industry attempts to push these capabilities into the physical world—powering autonomous drones, industrial robots, and self-driving vehicles—it has collided with a fundamental law of physics. Between an AI model capable of making a complex decision and a machine capable of executing it, there is a network. When the "brain" is located in a remote data center and the "body" is navigating a busy warehouse, the time it takes for data to travel back and forth becomes a critical bottleneck. This delay, known as latency, is the invisible ceiling currently limiting the deployment of physical AI at scale.[1][2]

To solve this infrastructure gap, the technology sector is turning to an unlikely source: the architect behind one of the most ubiquitous pieces of consumer software in history. Jean-Baptiste Kempf, the lead developer of the VLC Media Player—famous for its orange traffic-cone icon and over six billion downloads—has emerged from stealth with a new venture. His Paris-based startup, Kyber, is building a dedicated infrastructure layer designed specifically for the real-time remote control of physical devices.[3][4]

Kyber recently secured a $5 million seed funding round led by Lightspeed Venture Partners, with participation from OVNI Capital and Kima Ventures. The investment underscores a growing realization among venture capitalists that the AI boom has disproportionately focused on models and compute power, while neglecting the "last mile" of physical execution. As Lightspeed noted in its investment thesis, physical AI is ultimately only as effective as the underlying networking systems that run it.[1][6]

To understand why a new infrastructure layer is necessary, one must look at how remote machines are currently operated. Most existing systems rely on communication protocols originally developed for the consumer internet, particularly WebRTC. WebRTC is the foundational technology behind modern video conferencing, designed to facilitate communication between humans. For a video call, a latency of 200 to 300 milliseconds is entirely acceptable; the human brain easily smooths over the slight delay in conversation.[2][5]

However, translating human-centric latency to machine operations yields catastrophic results. In the physical world, a few hundred milliseconds of lag can represent several meters of travel for a drone or an autonomous vehicle before a steering command is actually executed. In industrial manufacturing, robotic surgery, or military applications, these delays are not just inconvenient—they are operationally unacceptable and potentially dangerous. If a remote operator or an AI agent cannot see what a robot is doing in true real-time, they cannot safely control it.[2][3]

Kyber addresses this by abandoning legacy video-conferencing protocols in favor of a unified software development kit (SDK) engineered from the ground up for machine control. The platform synchronizes high-definition video, spatial audio, complex sensor data, and control inputs into a single, cohesive data stream. By adapting highly optimized transmission protocols originally designed for cloud gaming and high-performance media, Kyber claims it can reduce end-to-end latency to as little as eight milliseconds.[2][5]

This architectural decoupling is a significant shift for the robotics industry. Historically, robotics companies have been forced to build their own proprietary, custom communication stacks from scratch. This fragmented approach is expensive, difficult to scale, and often results in subpar performance when deployed across diverse geographic regions with varying network conditions. Kyber's solution separates the operator, the compute environment, and the physical action, allowing them to exist in entirely different locations without sacrificing responsiveness.[4][5]

Kempf’s strategy for Kyber borrows heavily from the playbook that made VLC and FFmpeg foundational pillars of the digital video ecosystem. Neither of those projects became major businesses through direct consumer monetization. Instead, they were released as open-source software, which encouraged widespread, frictionless integration by developers around the world. Over time, they became the de facto technical standards embedded in millions of applications, streaming services, and connected devices.[2][3]

Kyber is adopting a similar "open-core" business model to drive rapid adoption across the highly fragmented robotics sector. The foundational software development kit and core transmission protocols will be released as open-source software, allowing engineers, researchers, and startups to freely integrate the low-latency technology into their prototypes and commercial hardware. Meanwhile, Kyber plans to monetize the platform by offering advanced enterprise features, comprehensive fleet management tools, and dedicated commercial support for massive multinational companies operating at scale. This ensures the technology remains accessible while building a sustainable revenue stream.[2][5]

This dual approach is particularly attractive to investors looking to capture value in the early stages of a new technology cycle. In software markets, the companies that establish the underlying technical standards often become the most valuable entities in the ecosystem. By providing a free, highly performant baseline, Kyber aims to become the default communication layer for the entire physical AI industry, regardless of which hardware manufacturers ultimately sell the most robots.[2][4]

The startup, which maintains a lean team of roughly twenty-five professionals across hubs in Paris, San Francisco, and Singapore, is not waiting for the future to arrive. Kyber's infrastructure is already being deployed in active commercial environments. Early adopters include organizations in the defense sector, telecommunications, and enterprise IT, where the software is being used to supersede legacy virtualization tools for remote access.[3][5]

Ultimately, the success of Kyber will serve as a bellwether for the broader robotics industry. If autonomous systems are to scale from isolated pilot programs featuring a few thousand units to global networks of hundreds of millions of connected devices, they will require a standardized, ultra-reliable nervous system. By bridging the gap between digital intelligence and physical action, Kyber is attempting to build the exact infrastructure that will allow the next generation of machines to safely navigate the real world.[1][5]