Jia Yangqing Reportedly Leaves NVIDIA Just One Year After Acquisition

According to SemiAnalysis, Lepton AI founder and CEO Jia Yangqing has left NVIDIA, just one year and two months after NVIDIA acquired Lepton AI for approximately $700 million. Jia joined NVIDIA in April 2025 as a vice president.

Acquisition Background and Product Struggles

Lepton AI, founded by Jia in 2023, offered a Python-native GPU cloud platform with core components including the Photon abstraction layer, Tuna inference engine, and open-source monitoring tool gpud. In April 2025, NVIDIA completed the acquisition, and Jia along with co-founder Bai Junjie joined the company. Post-acquisition, the Lepton AI platform was quickly shut down, blocking new user registrations and deleting its official Twitter account.

NVIDIA integrated Lepton AI into DGX Cloud Lepton, downgrading its role from a full cloud platform to a GPU workload scheduler interfacing with over 25 cloud partners. However, the product underperformed. DGX Cloud itself faced challenges: initially priced at $36,999 per month, it lost its price advantage after AWS cut prices by 45%. NVIDIA has since repurposed it as an internal R&D tool and stopped separately disclosing cloud spending commitments in Q2 of fiscal 2026.

Open-Source Promise Unfulfilled

NVIDIA had pledged to open-source Lepton's core software platform by 2026, but as of the end of June 2026, the promise remained unfulfilled. SemiAnalysis suggests this may be the direct cause of Jia's departure. Nearly all of Jia's major career works (e.g., Caffe, early TensorFlow, PyTorch 1.0, ONNX) were open-source, while NVIDIA's core business model is selling hardware. Open-sourcing Lepton could allow competitors to use its scheduler to manage non-NVIDIA GPUs, conflicting with NVIDIA's commercial interests.

Jia's Next Move

Public information shows that on June 9, 2026, Jia joined Hyperbolic Labs, a decentralized GPU compute marketplace, as an advisor. This aligns with his previous direction of promoting open AI infrastructure.

Industry Perspective

SemiAnalysis notes that Agentic Coding is changing software economics. For example, AMD engineer Anush Elangovan used 10-12 Claude Code agents to rewrite a Slurm replacement called Spur in Rust overnight, consuming 6.5 billion tokens per week. This suggests that the software capabilities NVIDIA acquired may be rapidly recreated by AI.