NVIDIA and DigitalOcean Collaborate on Always-On Agentic Technology Stack

NVIDIA and DigitalOcean Collaborate to Advance Open-Source AI Development

NVIDIA and DigitalOcean are joining forces to enhance the open-source AI landscape, with a focus on building production-ready agentic AI systems. During a recent DigitalOcean Deploy session titled “Open by Design,” Kari Briski, NVIDIA’s VP of Generative AI, and Salman Paracha, SVP of AI at DigitalOcean, discussed the growing demand for openness and flexibility in AI development. This collaboration aims to support developers in creating robust AI applications that can operate continuously.

The Need for Open Models in AI Development

The rise of generative AI is not solely attributed to proprietary models from established tech companies; open-source models are increasingly playing a crucial role in shaping the developer ecosystem. However, having access to these models does not guarantee their continuous improvement or regular updates. NVIDIA has identified a gap in the market where enterprise customers seek reliable access to open-source models that often remain stagnant after their initial release.

This realization led to the introduction of NVIDIA’s Nemotron, a family of multi-modal models designed specifically for agentic AI applications. Released in March 2026, Nemotron enables developers to create applications that require advanced reasoning capabilities and high computational efficiency while adhering to open-source standards. By utilizing these models alongside NVIDIA’s software libraries, developers can ensure their projects evolve over time with regular updates and expanded support.

Enhancing Control and Customization with Local Deployments

Running open-weight large language models (LLMs) locally offers developers greater control over performance, privacy, and customization options. A tutorial on deploying NVIDIA’s Nemotron 3 Nano on DigitalOcean’s GPU Droplets provides guidance for developers wishing to experiment with efficient open models without relying entirely on hosted AI APIs. This hands-on approach fosters an environment where developers can fine-tune their applications according to specific needs.

Briski emphasized the commitment of NVIDIA to treat these models as a library that will be continually improved upon, similar to their GPU technologies and CUDA libraries. The discussion also highlighted the importance of orchestration frameworks—tools that manage the lifecycle, memory, tool calling, and scaling of agentic systems. These frameworks are essential for enabling developers to build sophisticated AI-native applications.

Overcoming Challenges in Evaluation and Observability

Despite advancements in technology, many developers still face significant hurdles when attempting to create durable AI solutions comparable to established systems like OpenClaw or Claude Code. A major challenge lies in evaluating performance accurately; while some test cases exist for specific use cases such as coding, others lack standardized benchmarks. This absence makes it difficult for developers to assess the viability of their ideas effectively.

NVIDIA is actively working with partners like Synopsys and Cadence to develop test cases and benchmarks tailored for industries such as electronic automation. By creating more comprehensive evaluation frameworks, NVIDIA aims to lower barriers for entry into AI development and bolster developer confidence.

As part of this evolution, many developers have adopted sub-agent workflows that break complex problems into manageable subtasks assigned to individual agents. This approach allows for better testing and verification processes while maintaining transparency throughout the system’s architecture.

Token Economics: Balancing Cost and Value

The scalability of AI systems introduces new challenges related to token usage—the units that measure computational tasks performed by models. Developers must find ways to run consistently generating token systems while ensuring they can build effective business models around them. Briski pointed out that as model architectures evolve, so too will the methods used for counting tokens generated by different types of models.

Organizations must remain mindful of costs associated with larger models while also focusing on delivering value through enhanced workflows. To address this issue, NVIDIA is implementing technical measures aimed at improving token efficiency within its latest Nemotron model by utilizing hybrid-state-space transformers rather than traditional dense model architectures.

This shift from dense megamodels towards sparser mixtures of experts (MoEs) reflects a broader trend within the industry toward optimizing resource usage without sacrificing performance. As Briski noted, organizations need to adapt their thinking around token economics as they navigate these changes.

The Future: Evolving Architectures and Developer Empowerment

NVIDIA’s applied research team continues its commitment to exploring new architectures through collaboration with the open-source community. By reviewing academic research and testing innovative models, they aim to stay at the forefront of technological advancements in AI development.

DigitalOcean is also focused on expanding its ecosystem around open-source projects through initiatives like Plano—its data-plane technology—and research into small action models (SAMs). These SAMs are designed for task completion using context compression techniques that enhance efficiency without requiring extensive reasoning tokens.

Paracha emphasized the importance of providing developers with choices regarding harnesses (frameworks used for managing model lifecycles) so they can select tools best suited for their needs without being constrained by proprietary solutions. This approach fosters an environment conducive to innovation across diverse applications.

What This Means

The collaboration between NVIDIA and DigitalOcean marks a significant step toward democratizing access to advanced AI technologies through open-source initiatives. As organizations shift from traditional generative approaches toward more flexible open-source frameworks, developers will need to adapt their skills accordingly. Understanding evolving architectures and managing long-horizon tasks will be essential as they navigate this dynamic landscape.
Ultimately, this partnership aims not only at enhancing technological capabilities but also at empowering individual developers—enabling them to create innovative solutions that integrate seamlessly into existing software ecosystems across various industries.

For more information, read the original report here.