NVIDIA’s reported decision to abandon plans to open-source LeptonAI has led to the exit of Caffe creator Yangqing Jia, raising fresh concerns over the company’s commitment to open-source AI infrastructure and developer trust.
NVIDIA’s reported decision to reverse its commitment to open-source the core LeptonAI platform has triggered the departure of AI pioneer and Caffe creator Yangqing Jia, putting the company’s open-source credentials under renewed scrutiny.
According to SemiAnalysis, NVIDIA had pledged to open-source LeptonAI by 2026 after acquiring the startup in April 2025, but later abandoned that plan. The reported reversal became a key point of conflict for Jia, whose career has been built around open-source projects including Caffe, ONNX, PyTorch 1.0 and the Apache 2.0-licensed leptonai Python package.
Jia left NVIDIA about 14 months after the acquisition and has since joined Hyperbolic as a Technical Advisor. Neither NVIDIA nor Jia had commented at the time of publication.
SemiAnalysis attributed the breakdown to three factors: the reported reversal of the open-source commitment, engineering resources being diverted towards interface changes instead of solving GPU multi-tenancy challenges, and Jia’s early departure before a typical post-acquisition integration period.
The report also questioned NVIDIA’s broader open-source credibility, citing examples such as NVIDIA Inference Microservices remaining largely closed and flashinfer reportedly moving towards closed-source binary distribution.
While NVIDIA continues to dominate AI hardware through its GPUs, CUDA ecosystem and NIM inference stack, the episode highlights how developer trust and open-source credibility are becoming increasingly important competitive advantages in AI infrastructure, where projects such as vLLM, SGLang, Kubernetes and SLURM continue to shape developer preferences.
