Nvidia Releases NeMo Microservices To Streamline AI Agent Development
Nemo Microservices
Nvidia unveiled the general availability of its NeMo microservices, equipping enterprises with tools to build AI agents that integrate with business systems and continually improve through data interactions. These microservices are being launched at a time when organizations are seeking concrete AI implementation strategies that can provide measurable returns on significant technology investments.
Enterprise AI adoption faces a critical challenge: building systems that remain accurate and useful by continuously learning from business data. The NeMo microservices address this by creating what Nvidia describes as a 'data flywheel,' allowing AI systems to maintain relevance through ongoing exposure to enterprise information and user interactions.
The newly available toolkit includes five key microservices:
These components work together to build AI agents that function as digital teammates, capable of performing tasks with minimal human supervision. Unlike standard chatbots, these agents can take autonomous actions and make decisions based on enterprise data. They connect to existing systems to access current information stored within organizational boundaries.
The distinction between NeMo and Nvidia's Inference Microservices, branded as NIMs, lies in their complementary functions. According to Joey Conway, Nvidia's senior director of generative AI software for enterprise, 'NIMs is used for inference deployments – running the model, questions-in, responses-out. NeMo is focused on how to improve that model: Data preparation, training techniques, evaluation.' When NeMo finishes optimizing a model, it can be deployed through a NIM for production use.
Early implementations demonstrate practical business impacts. Telecommunications software provider Amdocs developed three specialized agents using NeMo microservices. AT&T collaborated with Arize and Quantiphi to build an agent that processes nearly 10,000 documents updated weekly. Cisco's Outshift unit partnered with Galileo to create a coding assistant that delivers faster responses than comparable tools.
The microservices run as Docker containers orchestrated through Kubernetes, allowing deployment across various computing environments. They support multiple AI models including Meta's Llama, Microsoft's Phi family, Google's Gemma and Mistral. Nvidia's own Llama Nemotron Ultra, which focuses on reasoning capabilities, is also compatible with the system.
This release enters a competitive landscape where enterprises have numerous AI development options. Alternatives include Amazon's Bedrock, Microsoft's Azure AI Foundry, Google's Vertex AI, Mistral AI, Cohere and Meta's Llama stack. Nvidia differentiates its offering through integration with its hardware ecosystem and enterprise-grade support through the AI Enterprise software platform.
Nvidia Nemo and Enterprise AI Adoption
For technical teams, the microservices provide infrastructure that reduces implementation complexity. The containerized approach allows deployment on premises or in cloud environments with enterprise security and stability features. This flexibility addresses data sovereignty and regulatory compliance concerns that often accompany AI implementations.
Organizations evaluating these tools should consider their existing GPU infrastructure investments, data governance requirements and integration needs with current systems. The need for AI agents that maintain accuracy with changing business data will drive adoption of platforms that support continuous learning cycles.
The microservices approach reflects a broader industry shift toward modular AI systems that can be customized for specific business domains without rebuilding fundamental components. For technology decision makers, the release represents another step in the maturation of enterprise AI tooling that narrows the gap between research capabilities and practical business implementations.
As enterprises move beyond experimentation toward production AI systems, tools that simplify the creation of continually improving models become increasingly valuable. The data flywheel concept represents an architecture pattern where AI systems remain aligned with business needs through ongoing exposure to organizational information.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Forbes
an hour ago
- Forbes
Microsoft's Bad News—500 Million Windows Users Must Now Decide
Surprising bad news suddenly hits Microsoft. A new warning has been issued for Windows users, whose PCs have been described as 'magnets for security threats,' just as new data gives Microsoft a surprising bad news story ahead of the critical next few months. You can expect many more such warnings as 500-million Windows users face an increasingly urgent decision. The latest advice comes courtesy of PC maker Asus, pointing out that 'if you're still using Windows 10 or, dare we say it, something even older — your computer's days of regular updates and support are numbered.' As for upgrades, 'what makes Windows 11 different?," Asus says. "one word: Copilot," as it pushes the latest range of AI PCs. Clearly, you don't need to decide on a premium Copilot PC to benefit from Windows 11's future-proofing, ensuring your PC receives critical security updates after Windows 10's demise in October. AI PCs remain a niche, despite projections they will eventually dominate new PC sales. Right now, there's a more fundamental decision to make. Windows 10 versus Windows 11 globally. The latest Windows market data presents a painfully bleak picture with just over five months to run until free Windows 10 security updates end for all users. Paid extensions are available, but they're expensive for enterprises and restricted to just 12-months for home users who also must pay. Microsoft is pushing free upgrades not paid extensions. A month ago, it seemed Windows 11 had turned the tide against Windows 10. The newer OS already outanks its older sibling in the U.S. but not globally. Come the end of April, though, Windows 11 was within 10% of Windows 11 for the first time. 'Just over half (53%) of all users are still on Windows 10, but that's inching down month by month.' Not any more, it seems. While more directional than exact, Statcounter's data at the end of May shows a slight month-over-month increase for Windows 10, while Windows 11 dips. This after four months of steady progress the other way. Windows 10 is holding stubbornly above 50% while Windows 11 remains 10% behind. Windows 10 versus Windows 11 in U.S. This means there are around 750 million users are yet to upgrade to Windows 11, of which at least 240 million don't have an eligible PC. That still leaves around 500 million users who can take up Microsoft's offer for a free Windows 11 upgrade but have not. Even in the U.S., where Windows 11 has overtaken Windows 10, May's data suggests Windows 10 has grown its share from 41% in April to more than 43%, while Windows 11 drops a more worrying 3.5%, from 56.5% down to below 53%. All this makes June's data critical. Come the end of this month, there will be just three months until Windows 10 is shuttered. If Microsoft is to avoid a cybersecurity nightmare hitting mid-October, something need to change. For all those Windows 10 users with PCs eligible for a free upgrade, do not run out of time.
Axios
3 hours ago
- Axios
Scoop: Apple to appeal EU requirement to share info with tech rivals
Apple filed an appeal to the European Commission's specifications around how the company complies with its interoperability requirements, which compel it to share user information with outside developers as part of the Digital Markets Act, per a source familiar with the situation. Why it matters: Apple continues to push back forcefully against requirements imposed by the DMA, which the company has said forces it to give up its intellectual property and compromise user privacy. The deadline to appeal this particular requirement was May 30. How it works: The interoperability requirements are meant to give other device manufacturers and app developers access to Apple features that are normally exclusive to Apple products, including WiFi pairing and notifications on non-Apple smartwatches and headsets. Apple and Meta were the first companies to be hit with DMA fines in April, as Axios first reported. Apple put out a report last December arguing that the DMA's interoperability reports could be abused and expose private user information. What they're saying: "We design our technology to work seamlessly together, so it can deliver the unique experience our users love and expect from our products. The EU's interoperability requirements threaten that foundation, while creating a process that is unreasonable, costly, and stifles innovation," an Apple spokesperson told Axios. "These requirements will also hand data-hungry companies sensitive information, which poses massive privacy and security risks to our EU users." Per the Apple spokesperson, as allowed under the DMA's interoperability requirement, companies including Meta, Google, Garmin and Spotify have requested from Apple user notification content and stored WiFi networks, "giving them the ability to access personal information that even Apple doesn't see." The spokesperson added, "In the end, these deeply flawed rules that only target Apple—and no other company—will severely limit our ability to deliver innovative products and features to Europe, leading to an inferior user experience for our European customers." The other side: Companies that rely on the App Store to distribute their products have welcomed the DMA's requirements of Apple, saying they are long overdue for developers and other businesses to offer customers better features on their apps and devices.
Forbes
3 hours ago
- Forbes
The Strategy Behind Hugging Face's Acquisition Of Pollen Robotics
Reachy 2 In April 2025, Hugging Face acquired Pollen Robotics, a France-based company that develops humanoid robots, including Reachy 2. This marks a milestone in the convergence of generative AI and robotics, known as physical AI. This article analyzes the strategic reasons behind Hugging Face's acquisition of Pollen Robotics. Hugging Face's acquisition of Pollen Robotics demonstrates a long-term vision for the future of AI technology, as it evolves from digital intelligence to physical form. The strategy is based on three core pillars: 1) Vertical integration of the AI-to-robotics stack 2) Ecosystem leverage through their massive developer community 3) Timing advantage as foundation models become capable of controlling physical systems. The acquisition gives Hugging Face immediate access to Pollen's flagship Reachy 2 humanoid robot, a $70,000 research platform already deployed at prestigious institutions such as Cornell University and Carnegie Mellon. With 7 degrees of freedom, bio-inspired arms capable of handling 3 kg payloads, advanced VR teleoperation and fully open-source hardware designs, the Pollen Robotics Reachy 2 offers a proven platform for Hugging Face to build upon instead of starting from scratch. The timing proves particularly strategic given several converging factors. Nvidia recently chose Hugging Face as the preferred platform for its GR00T N1 humanoid robot foundation models, signalling industry recognition of Hugging Face's platform capabilities. Meanwhile, the remarkable growth of Hugging Face's LeRobot library to over 12,000 GitHub stars in just 12 months demonstrated strong developer demand for open robotics tools. The acquisition also follows Hugging Face's strategic hire of Remi Cadene, a former Tesla Optimus engineer, who now leads their robotics division. While Hugging Face has emerged as the largest collection of open-source and open-weight models, Pollen Robotics focuses on the vision of open hardware for robotics. This acquisition combines the strengths of open-source software with open hardware design in the field of robotics. The embodied AI market has reached an inflection point where theoretical capabilities meet practical applications, driven by breakthroughs in foundation models for robotics and dramatic cost reductions in hardware components. AI models such as Pi0 from Hugging Face, Nvidia's GR00T N1 and Google's Gemini Robotics extend the power of generative AI to robotics. Instead of generating text or media content, these models are trained to send commands directly to the robotic hardware. They leverage multimodal AI by combining video content and policies to generate commands that control robots. The combination of Hugging Face's AI infrastructure and Pollen's robotics expertise creates technical synergies that neither company could achieve independently. Hugging Face brings 1.5 million models and datasets hosted on its platform, proven infrastructure serving 12 petabytes of data, and deep expertise in transformer architectures and diffusion models. This AI foundation provides the intelligence layer essential for next-generation robotics. Hugging Face's acquisition of Pollen Robotics represents more than a product expansion. It marks a fundamental shift in how AI and robotics will evolve together. By combining proven AI infrastructure with capable robotics hardware under an open-source philosophy, Hugging Face creates unique value that neither pure software nor pure hardware companies can replicate. The broader implications extend beyond corporate strategy. If Hugging Face succeeds in democratizing robotics as they've democratized AI, we may see an explosion of innovation similar to what followed the open-sourcing of deep learning frameworks.
