The Architect of VLC is Betting on the ‘Physical AI’ Infrastructure Layer
Table of Contents
From Media Playback to Machine Motion
For most of the internet, Jean-Baptiste Kempf is the man behind the orange traffic cone. As the lead developer of VLC Media Player—a piece of software with over 6 billion downloads—Kempf has spent decades mastering the art of making complex data streams play smoothly across virtually any device on earth. Now, he is applying that same obsession with latency and synchronization to a far more volatile environment: the physical world.
Kempf is the founder of Kyber, a Paris-based startup building an infrastructure layer designed to control remote devices in real time. While VLC managed the delivery of video and audio to a screen, Kyber manages the telemetry and control inputs required to operate drones, robots, and industrial machinery from thousands of miles away. It is a bid to solve the ‘last mile’ of connectivity for what is increasingly being termed ‘Physical AI’—the intersection of large language models and robotic actuation.
The project recently secured $5 million in funding led by Lightspeed, a venture firm with a high-conviction portfolio including Anthropic and Mistral AI. The investment signals a growing belief among VCs that the bottleneck for robotics isn’t just the AI models themselves, but the underlying plumbing that allows those models to interact with hardware without debilitating lag.
The Millisecond Problem
In the world of remote operation, latency isn’t just a nuisance; it’s a safety hazard. A delay of a few hundred milliseconds in a remote-driving or surgical application can be the difference between a successful operation and a catastrophic failure. Kyber’s core offering is an SDK that synchronizes video, audio, sensor data, and control inputs with surgical precision.
The name “Kyber” is a nod to the crystals that power lightsabers in Star Wars, reflecting Kempf’s focus on speed and energy. The technical foundation of the company draws heavily from Kempf’s tenure as CTO at Shadow, a cloud gaming startup. Cloud gaming is essentially the ultimate stress test for streaming technology—requiring high-fidelity video and near-instantaneous input response. Kyber adapts these principles for the Internet of Things (IoT), tuning performance to the specific compute constraints of the edge device.
According to Kempf, the current market is fragmented. While massive entities like Tesla or Waymo have built proprietary stacks for remote vehicle intervention, those solutions are designed for fleets of a few thousand. Kyber is aiming for a different scale entirely: a world where millions of drones and robots operate autonomously but require a standardized, scalable way to be monitored and overridden by humans or higher-level AI agents.
A Hybrid Strategy: Open Source and FDEs
True to the ethos that made VLC a global staple, Kyber is maintaining a dual-track distribution model. The core project is open source, allowing a wide community of developers to build upon the framework. This creates a massive top-of-funnel for adoption, while the company sells a productized, enterprise-grade version to corporate clients.
To bridge the gap between software and hardware, Kyber has adopted a deployment strategy reminiscent of Palantir. A significant portion of its 25-person team consists of Forward-Deployed Engineers (FDEs). These are not traditional sales reps or account managers, but engineers who embed themselves within a client’s operations to customize the deployment of the SDK into complex environments.
Currently, Kyber is focusing its commercial efforts on three primary verticals: robotics, diverse drone applications, and remote IT access. In the latter, Kempf sees an opportunity to disrupt the legacy remote-access market, which he suggests has been stifled by companies spending tens of millions on closed-source solutions that are rarely shared or modernized.
With offices in Paris, San Francisco, and Singapore, the company is positioning itself as the connective tissue for a global fleet of machines. Whether those machines are delivering packages, inspecting power lines, or managing server racks, the goal remains the same: making sure the signal reaches the machine before the moment of opportunity passes.
