Hi Dotikers!

Open, but not for everyone

The old open versus closed fight came back this week, and this time with signatures at the bottom of the page. Jensen Huang published an open letter on open weights and American AI leadership, Microsoft, Meta and IBM signed, Google and OpenAI followed over the weekend. Nvidia then launched an Open Secure AI Alliance built around open cybersecurity tooling. Anthropic did not sign, got accused of wanting to ban open weights, and Dario Amodei answered by saying the opposite: no ban, but chip export controls, a crackdown on industrial scale distillation, and mandatory safety testing for any sufficiently capable model, open or closed.

The trigger is called Kimi K3: 2.8 trillion parameters published on Hugging Face, the largest open model ever released.

This debate is dressed up as a clash of principles, but it is mostly a clash of business models. Nvidia sells GPUs whichever side wins, while negotiating 250 billion in financing for OpenAI data centres. Meta needs the open ecosystem to stay alive. Anthropic sells closed access. Nobody here is neutral, and the word "open" makes for excellent makeup.

And let us look at the bill: K3 needs roughly 1.4 TB of fast memory and at least 64 accelerators. Open the way a yacht is open to the public, nothing stops you from coming aboard, you just need the yacht first.

The only measure that really holds up is the one that does not depend on whether the weights are published: test before shipping. The rest is positioning.

Convenient timing too, since yesterday we covered Gemini Robotics and dexterity. With Gemini Robotics ER 2 landing the same week, the real moat is shifting toward physical data, sensors and hardware. Publishing weights hands nobody the factory.

Alex.

Most support teams are experimenting with AI. Few are transforming because of it. The gap is where the most consequential decisions in support are being made, and where most teams get stuck. Hear from industry leaders on moving from pilot to production.

Watch now