AI is having it's Linux moment
Linux was a turning point in an otherwise closed-war between operating systems in the 90s, and it unequivocally made the world a better place. Although cutting edge research might still be done in American labs, Chinese labs have proven to be equally innovative and, at the same time, unbelievably open and generous with their research. They will be looked upon as the real heroes of this time, Deepseek being the pioneer.
With the arrival of Kimi K2, agentic AIs have reached near-parity with closed-source frontier lab models. A refreshingly open model, open-weights and MIT-license (well, almost)
What does this mean for the Software industry?
- It is now possible to entirely run a Claude Code equivalent inside your own rack of GPUs
- It might be possible to run a quantized version on a consumer GPU which is "good enough", but customizable
- It will cost a lot less to do inference
- There's a RedHat type of company to be formed which duct-tapes open-source software and packages it nicely for corporations
As performance gaps continue narrowing and deployment costs plummet, competitive advantage will increasingly derive from the execution around the model - rather than having access to the latest model.