A report from semiconductor research firm SemiAnalysis has revealed that Yangqing Jia, Vice President of System Software at NVIDIA and founder of Lepton AI, has left the company.
This departure comes just one year after NVIDIA CEO Jensen Huang orchestrated a major acquisition of the small startup for a sum reported to be around $700 million.
SemiAnalysis described this development as "shocking." As of the time of reporting, neither NVIDIA nor Jia himself has issued an official statement regarding the matter.
Understanding the Strategic Context
To grasp the significance of this exit, it is necessary to examine NVIDIA's broader strategy in AI infrastructure. Huang's vision extends beyond merely selling chips.
His plan involves building a proprietary software platform that operates atop cloud services like AWS and Azure, aiming to centralize access to critical hardware resources for developers.
In April 2025, NVIDIA completed its acquisition of Lepton AI, a company co-founded by Jia and Junjie Bai, another Meta AI alumnus.
The company, founded in 2023, specialized in helping developers access and utilize GPU resources. Reports indicated the deal was valued in the hundreds of millions, with industry sources later specifying a figure around $700 million.
Combined with a prior acquisition of the AI infrastructure platform Run:ai for approximately $1.2 billion, NVIDIA's total investment in its GPU cloud software stack approached $2 billion.
Following the acquisition, Lepton AI was rebranded as DGX Cloud Lepton, positioned as a marketplace connecting global GPU cloud providers with AI developers.
Jia was seen as the key executive tasked with driving this ambitious initiative forward.
Reasons for the Swift Departure
The report from SemiAnalysis points to several contributing factors behind the split after just one year.
The most immediate catalyst was reportedly underperformance. The DGX Lepton platform failed to meet external and internal expectations, with reports suggesting it had largely ceased external operations by mid-2025.
A deeper, more fundamental conflict arose over commitments to open-source software. NVIDIA had reportedly pledged to open-source the core Lepton platform by 2026 but later reversed this decision.
SemiAnalysis speculates that Jia's departure may be linked to Huang's ultimate refusal to approve the open-source plan. As a foundational contributor to major open-source AI frameworks like Caffe, PyTorch, and ONNX, Jia has been deeply embedded in the open-source ecosystem for decades, making this reversal a significant clash of values.
Additionally, it is reported that NVIDIA's product management culture consumed substantial team resources on superficial tasks like UI adjustments, while core challenges such as multi-tenancy remained unresolved, leaving the platform ill-equipped to meet the demanding needs of AI-native developers.
Immediate Transition to a New Role
Earlier this month, GPU cloud services provider Hyperbolic announced that it had appointed Jia as a company advisor.
In its announcement, Hyperbolic stated that Jia would provide expertise in areas including AI systems, cloud infrastructure, GPU marketplaces, multi-cloud orchestration, and GPU utilization optimization.
The focus of his new advisory role aligns closely with his previous responsibilities overseeing the DGX Lepton platform at NVIDIA, indicating an almost seamless transition to another player in the same competitive field.
Broader Industry Implications
Jia's exit is viewed by the industry as a strong signal with wider ramifications.
Firstly, it suggests that dominance in GPU hardware cannot be easily extended into the software layer. As the hardware leader attempts to expand upward, it encounters greater resistance, indicating that AI infrastructure cannot be monopolized in the same way as GPU hardware.
Secondly, middleware platforms are facing a crisis of value. The rise of agentic coding tools, such as Cursor, Claude Code, and Codex, which can automatically generate infrastructure code from natural language, bypasses the need for platforms designed primarily to lower engineering barriers.
This trend forces such platforms to re-prove their indispensable value in solving fundamental resource scarcity issues.
Comments