1991: Linux version 0.01 is released

On 17 September 1991, a 21-year-old Finnish computer-science student named Linus Torvalds quietly posted the first public release of his hobby kernel, version 0.01, to an FTP server at the Helsinki University of Technology. The announcement was modest: a few hundred lines of assembly and C that could boot on an Intel 386, talk to a terminal, and not much else. Torvalds had written it largely because he wanted a free Unix-like environment he could tinker with on his own PC. He did not imagine he was launching one of the most consequential pieces of software in computing history.

The release sat at the intersection of several older intellectual currents. Unix itself, born at Bell Labs in the early 1970s, embodied the modular, tool-oriented philosophy of software design. The free-software movement, articulated by Richard Stallman a few years earlier, insisted that users must be free to study, modify and redistribute code. Torvalds combined these ideas with a pragmatic, “release early, release often” engineering style and an open invitation for others to contribute. Within months a small community of developers was sending patches; within a few years Linux had become a full operating system when combined with GNU tools.

That collaborative model proved astonishingly scalable. By the late 1990s Linux powered web servers and supercomputers. In the 2000s it became the default substrate for cloud computing. Today the connection to artificial intelligence is direct and nearly universal. The vast majority of machine-learning frameworks—TensorFlow, PyTorch, JAX—are developed and run on Linux. Training clusters at every major AI lab, from university basements to hyperscale data centres, boot Linux kernels. Container technologies such as Docker and orchestration systems such as Kubernetes, both native to the Linux ecosystem, make it possible to package and schedule the enormous workloads that modern neural networks demand.

Even the open-source culture that Linux popularised shapes how AI research itself is shared: model weights, training code and evaluation benchmarks circulate under permissive licences that descend intellectually from the same ethos Torvalds helped normalise. What began as one student’s weekend project on a particular September day has become the invisible operating layer beneath almost every contemporary AI system.