Key Takeaways

  • Nvidia’s investments in cloud computing and inference providers extended its influence across the AI delivery chain.
  • A leadership crisis involving Jensen Huang would not, by itself, shift the market decisively toward closed AI models.
  • Enterprise AI is likely to remain divided between managed proprietary services and open models used for cost, control, and sovereign deployment.

Under Jensen Huang, Nvidia moved well beyond selling graphics processors. The company invested billions in cloud computing providers such as CoreWeave and expanded its exposure to AI inference, helping create an ecosystem in which Nvidia participates in more layers of the market it supplies.

That strategy matters because control over AI increasingly depends on distribution and infrastructure, not just who develops the strongest model. Nvidia’s GPUs and CUDA software ecosystem sit beneath many closed and open systems. Its investments in cloud capacity also connect the company more closely to the businesses renting that computing power to developers and enterprises.

The result is an unusually broad position. Nvidia supplies hardware, software tools, networking technology and ecosystem capital. In practical terms, a startup building an open model and a large vendor selling a proprietary AI service may rely on much of the same underlying Nvidia infrastructure.

That raises a provocative question: Would a personal downfall involving Huang alter the balance between open and closed AI?

Probably not on its own. Leadership stability can affect investor confidence, partnerships and corporate strategy, particularly when an executive is closely identified with a company. But the direction of open AI is being shaped by a wider set of forces, including inference costs, licensing rules, enterprise support requirements, national technology policies and access to computing capacity.

Open development is still expanding rapidly. Hugging Face recorded about 1.18 million new model repositories in 2025 and more than 3 million total repositories by August 2026. That volume points to a broad experimentation layer involving researchers, startups, enterprises and individual developers. It is not a movement that depends on a single executive or hardware supplier.

Enterprise deployment tells a less straightforward story. Menlo Ventures found that open-source LLM usage declined from 19% to 11% in its 2025 survey. However, the firm estimated in 2026 that open models still processed roughly one-fifth of enterprise tokens while capturing only about 4% of model spending.

Usage and revenue represent different contests. Open models can process meaningful workloads without generating spending comparable to managed proprietary services. Closed vendors tend to monetize operational simplicity, service guarantees, integrated security controls and technical support. Open models compete more directly on flexibility, deployment control and cost.

Geography adds another layer. An MIT and Hugging Face study reported in 2025 found that Chinese-developed models represented 17.1% of global downloads, compared with 15.8% for U.S. models. Alibaba’s Qwen and DeepSeek, alongside Meta’s Llama, illustrate how the open-model ecosystem is becoming more distributed across countries and organizations.

That diversity reduces dependence on any one model producer, but it does not eliminate infrastructure concentration. Many open models still run on Nvidia GPUs, rely on CUDA-compatible tools or reach customers through cloud providers with substantial Nvidia capacity. Open weights do not automatically mean an open computing stack.

For business buyers, the distinction is important. Organizations evaluating AI systems should examine model licensing, hosting options, switching costs, data controls and hardware dependencies separately. The NIST AI Risk Management Framework can also help enterprises consider governance and trust without treating open or closed development as inherently safer.

The likely market structure is bifurcation rather than outright victory for either camp. Closed vendors can continue leading managed enterprise deployments where support and convenience carry a premium. Open models can remain influential in experimentation, sovereign AI programs, edge systems and cost-sensitive inference. Nvidia’s leadership could affect how that infrastructure develops, but the open-versus-closed contest is already much larger than Jensen Huang.