Kevin S. Xu on X: "Prediction: Nvidia Nemotron's market share will 5-10x from now until end of this year. Open source AI is having a real moment. Nemotron is best set up to capture it, especially inside large enterprises. It won't show up cleanly on OpenRouter stats or similar leaderboards, because the install will live within on-prem environments. This is not because Nemotron is the "smartest" model per se. Many other open/close models are "smarter". But it is the most open! Weights are open. So are pre and post-training data, plus how the model is built. Only other model that is as open is the K2 models from @mbzuai, an academic institution. (See @ArtificialAnlys openness index, an increasingly important chart.) None of the open source Chinese labs open up beyond just the weights. Many do write great papers to share methodology and innovation. None share the datasets that go into pre/post training. This data transparency part is becoming increasingly important for large companies, who want to verify the models they deploy are not trained on data that could present security vulnerabilities, then post-train them further with its own data and IP. Only Nemotron fits the bill. It also has all the hyperscalers plus Palantir as partners to make deploying, post-training, plus continuous improvement on-prem in perpetuity manageable. This is huge undertaking!" / X

Prediction: Nvidia Nemotron's market share will 5-10x from now until end of this year. Open source AI is having a real moment. Nemotron is best set up to capture it, especially inside large enterprises. It won't show up cleanly on OpenRouter stats or similar leaderboards, because the install will live within on-prem environments. This is not because Nemotron is the "smartest" model per se. Many other open/close models are "smarter". But it is the most open! Weights are open. So are pre and post-training data, plus how the model is built. Only other model that is as open is the K2 models from @mbzuai, an academic institution. (See @ArtificialAnlys openness index, an increasingly important chart.) None of the open source Chinese labs open up beyond just the weights. Many do write great papers to share methodology and innovation. None share the datasets that go into pre/post training. This data transparency part is becoming increasingly important for large companies, who want to verify the models they deploy are not trained on data that could present security vulnerabilities, then post-train them further with its own data and IP. Only Nemotron fits the bill. It also has all the hyperscalers plus Palantir as partners to make deploying, post-training, plus continuous improvement on-prem in perpetuity manageable. This is huge undertaking!

6:53 PM · Jul 6, 202668.1KViews

Post

Post