On July 26, Moonshot AI released Kimi K3's open weights — a 2.8-trillion-parameter model, the largest open-weight release to date, at roughly 1.4 terabytes in MXFP4 quantization. Unlike last week's API launch (which raised market doubts about AI capex), this release shifts a different debate: data sovereignty.
What Changed
According to TechTimes and Moonshot's own technical blog, Kimi K3's full weights can now be freely downloaded under a Modified MIT license and run on a company's own infrastructure — without depending on the hosted API in China. That resolves, at least technically, the main objection companies had to using Kimi K3 until now: sending sensitive data to servers outside their control and outside local jurisdiction.
Why It Matters
Until last week, using Kimi K3's capability meant accepting that data would flow through Moonshot's infrastructure in China. With open weights, a company can run the same model inside its own data center or a local cloud provider, keeping full control over where data lives and who can access it. That doesn't eliminate cost — running a 1.4TB model requires heavy infrastructure, realistic today mainly for large enterprises or specialized inference providers — but it removes the legal and trust barrier that blocked the model's evaluation in regulated sectors.
The Impact for Brazil
For banks, insurers, and healthcare companies in Brazil — sectors with strict LGPD requirements and sensitive data — this is the first time they can evaluate a model at Kimi K3's level without the dilemma of sending data abroad. In practice, this should accelerate internal evaluations at companies that until now ruled out Chinese models outright over sovereignty concerns alone — even without yet having the infrastructure to run it themselves.
Entercast's Take
This is another example of how a model's access architecture matters as much as its capability. Companies that build their AI strategy around deployment options — managed API, private cloud, on-premise — rather than around a single vendor, gain real flexibility when the data sovereignty conversation reaches the compliance table, which is happening faster and faster in Brazil.