VLC Creator Raises $5M for Kyber, an Ultra-Low Latency Network for Robots

Artificial intelligence is rapidly graduating from generating text on digital screens to navigating the physical world. Yet, as autonomous drones, industrial robots, and self-driving vehicles become more capable, a critical technological bottleneck has emerged: the infrastructure required to control them. Between an AI model capable of making a decision and a physical machine capable of executing it, there is a massive need for real-time, bidirectional communication. Today's internet infrastructure, designed primarily for asynchronous communication and buffered content delivery, was never optimized to handle the millisecond-level demands of physical machines.[3]

Enter Jean-Baptiste Kempf, a French serial entrepreneur and open-source legend best known as the lead developer of VLC Media Player. Having built the ubiquitous orange traffic-cone software that has been downloaded over six billion times, Kempf is now turning his attention from digital media to physical robotics. His new startup, Kyber, aims to provide the missing infrastructure layer for controlling remote devices in real time, ensuring that distant machines can be operated as if they were sitting right in front of the user.[1][2][7]

The technology industry is taking notice of this pivot. Kyber recently emerged from stealth with a $5 million seed funding round led by Lightspeed Venture Partners, a prominent venture capital firm that has previously backed AI heavyweights like Anthropic and Mistral AI. The round, which also saw participation from OVNI Capital and Kima Ventures, underscores a growing consensus among investors: physical AI is only as good as the underlying network systems running it.[2][3][4][7]

The core problem Kyber aims to solve is the fragmented and latency-prone nature of remote machine control. As the number of robots and drones operating in public and industrial spaces scales up, managing them becomes exponentially complex. These machines require constant, reliable connections to transmit high-definition video, audio, and sensor data back to a central operator or an AI control system. In environments like remote surgery, industrial manufacturing, or autonomous driving, even a tiny delay in data transmission can result in catastrophic real-world consequences.[3][5]

Currently, many companies in the robotics and drone space rely on custom-built, proprietary systems to manage this communication. However, these bespoke solutions often struggle to scale when fleet sizes grow from a few dozen experimental units to thousands of deployed machines. Kempf noted that the largest remote driving fleets today manage perhaps 2,000 to 3,000 vehicles, but scaling to millions of concurrent devices requires a fundamentally different kind of platform—one that standardizes the communication layer across the entire industry.[2][5]

To address this, Kyber has developed a Software Development Kit (SDK) that synchronizes video, audio, sensor data (such as GPS and IoT metrics), and control inputs (like mouse, keyboard, or gamepad commands) over a single unified link. The goal is to provide developers with a cohesive software layer capable of simultaneously managing all these streams while ensuring temporal consistency. By synchronizing these inputs with minimal delay, Kyber allows human operators or AI agents to control machines located thousands of miles away with near-zero perceived lag.[3][5][7]

The mechanism behind Kyber’s speed relies heavily on Kempf’s two decades of experience in video streaming. The platform is built on top of FFmpeg and VLC, the foundational open-source projects that quietly power much of the world's multimedia infrastructure. However, Kyber fundamentally rewires how these tools operate over a network. Instead of relying on traditional web protocols, Kyber utilizes QUIC, a modern transport-layer network protocol designed to reduce latency and accelerate connection speeds.[2][3][4]

The ultimate metric for this technology is "glass-to-glass" latency—the total time it takes for a video frame to be captured by a camera lens, encoded, transmitted across a network, decoded, and displayed on an operator's screen. During a demonstration at the Mile High Video conference, Kempf showcased Kyber achieving an astonishing 8 milliseconds of glass-to-glass latency. To put that into perspective, the blink of a human eye takes about 100 to 400 milliseconds.[2][4]

Achieving this level of speed requires a radical departure from conventional streaming philosophies. Traditional protocols like WebRTC or HTTP Live Streaming (HLS) prioritize maintaining consistent audio-video synchronization and high image quality, often buffering data to smooth out network hiccups. Kyber, conversely, prioritizes absolute speed above all else. As Kempf explained, the system is not concerned with maintaining perfect sync in the traditional sense; it is designed to push latency as low as physically possible, dropping frames if necessary to ensure the control inputs remain instantaneous.[4]

The origins of Kyber trace back to Kempf’s tenure as the Chief Technology Officer at Shadow, a French cloud gaming company. Cloud gaming requires incredibly low latency to ensure that a player's button press translates instantly to an on-screen action, making it a perfect testing ground for remote control technologies. Kempf began building Kyber as a side project to solve these exact bottlenecks before eventually spinning it out into a dedicated infrastructure company.[2][7]

The startup’s name itself is a nod to the absolute limits of speed and energy. It is inspired by the Kyber crystals from the Star Wars universe, which power lightsabers and channel immense energy. The branding reflects the company's core ethos: when controlling objects in the real world, every single millisecond matters, and the technology must operate as close to the speed of light as network physics will allow.[6][7]

In a move that mirrors the success of VLC, Kyber is adopting an "open-core" business model. The foundational technology is open-source and available under the AGPL license for non-commercial and open-source projects. This strategy is designed to accelerate widespread adoption among developers, researchers, and hobbyists. For large corporations and enterprise deployments that require proprietary integration, Kyber generates revenue through commercial licenses and specialized engineering services.[3][4][6]

By making the core technology open and accessible, Kempf hopes to establish Kyber as the de facto standard for machine communication, much like VLC and FFmpeg became the standards for video playback. If successful, communication protocols and real-time control layers could become as strategically important to the emerging machine economy as cloud computing platforms like AWS and Azure became to the software economy over the last decade.[3]

While robotics and drones are the primary focus, Kyber’s potential applications extend far beyond physical machines. The technology is designed for any use case where the operator, the compute power, and the action are in three different locations. This includes cloud rendering for virtual and augmented reality, remote IT access—where Kyber aims to challenge established enterprise providers like Citrix—and telemedicine, where surgeons could operate robotic arms remotely with absolute precision.[2][4][5][7]

Ultimately, Kyber is positioning itself as the nervous system for the next generation of artificial intelligence. As AI models move out of data centers and into the physical world, they will need to continuously ingest high-fidelity sensor data and output rapid mechanical commands. By solving the latency bottleneck, Kyber is building the invisible infrastructure that will allow hundreds of millions of autonomous agents to interact with our world safely, reliably, and in true real time.[3][7]