NVIDIA launches 24/7 AI agents for autonomous telecoms

What happened?

NVIDIA has announced at DTW Ignite 2026 in Copenhagen a series of technologies that enable telecom operators to leap from task-based automation to full network autonomy. The company proposes a telecommunications autonomy platform based on three pillars: privacy-preserving synthetic data, language models specialized in the telco domain, and secure runtimes for AI agents.

According to NVIDIA's official blog, operators have already achieved significant returns from task automation in network management, customer service, and back-office operations. However, the company believes that "automation is no longer the goal, but the launchpad toward autonomy." This announcement comes amid growing pressure on operators: 5G, IoT, and edge computing networks have multiplied operational complexity, and margins are shrinking. Traditional rule-based automation is no longer sufficient. NVIDIA is betting on AI agents that "proactively monitor issues and coordinate changes across network, IT, and business systems," a qualitative leap toward truly autonomous operations.

Historically, the telco industry has been a pioneer in adopting automation: from early OSS/BSS systems to software-defined networking (SDN) and network functions virtualization (NFV). However, generative AI and autonomous agents represent a new frontier. In 2023, TM Forum already identified autonomy as one of the "strategic imperatives" for operators. Now, NVIDIA provides the concrete tools to make it happen.

Key components

Synthetic data to overcome the privacy barrier

One of the main obstacles to training AI models in telecommunications is the sensitivity of network and customer data. According to the NVIDIA report cited in the blog, 54% of operators cite data-related issues as their biggest barrier to adopting generative AI. To address this, NVIDIA introduces NeMo Safe Synthesizer and NeMo Anonymizer, tools that generate synthetic datasets that reflect the structure and distribution of real data without exposing sensitive information. These tools are part of the open-source NeMo framework, allowing operators to customize generation to their needs. SoftBank Corp. is already using these technologies to fine-tune its large language model for telecom and build specialized network agents. This approach is not new in other sectors (healthcare, finance), but in telco it is pioneering due to the scale and critical nature of the data.

Telco domain models

Autonomous agents require reasoning models that understand the telecommunications domain. NVIDIA proposes fine-tuning base models (such as Llama or Mistral) with the generated synthetic datasets, creating models specialized in telco terminology, protocols, and use cases. These models allow agents to interpret the operator's intent and act safely across multiple domains (network, IT, business). For example, an agent could understand a request like "optimize bandwidth in the northern zone during peak hours" and execute the necessary actions without human intervention. This level of semantic understanding goes beyond rule-based automation, which required programming each scenario.

Secure runtimes for agents

The platform includes a secure runtime that ensures agents act within policies set by humans, maintaining human control over critical decisions. This is essential in a sector where a mistake can affect millions of users. The runtime implements guardrails that verify each agent action against predefined policies and can stop or escalate actions that exceed certain thresholds. This "human-in-the-loop" approach is similar to that used in autonomous aviation or level 3 autonomous vehicles, where the machine operates autonomously but with human supervision.

Why is this important?

The telecommunications sector faces growing pressure to manage increasingly complex networks (5G, IoT, edge computing) efficiently without exposing sensitive data. NVIDIA's proposal directly addresses the main identified barrier: according to a company report, 54% of operators cite data-related issues as their biggest obstacle to adopting generative AI. Moreover, NVIDIA's vision goes beyond automating specific tasks. It aims to create systems that "proactively monitor issues and coordinate changes across network, IT, and business systems," representing a paradigm shift toward truly autonomous operations. This approach could reduce mean time to repair (MTTR) from hours to minutes and enable predictive capacity management that prevents congestion before it occurs.

Compared to previous events, this announcement recalls the launch of the first SDN controllers in 2011, which promised programmable and flexible networks. However, agent-based autonomy goes a step further: it not only programs the network but manages it intelligently and adaptively. It also parallels the evolution of virtual assistants in customer service, which moved from rule-based chatbots to generative AI assistants. Now, that same evolution is applied to network operations.

Market implications

For operators, this technology promises to reduce operational costs (advanced automation could cut OPEX by 20% to 30% according to analysts), improve network resilience, and accelerate the deployment of new services. SoftBank Corp., as the first real-world use case, will validate these promises in a production environment. For telecom software and equipment vendors (such as Ericsson, Nokia, Huawei), it poses a challenge: they must integrate these AI capabilities into their solutions or risk becoming obsolete. Companies like Amdocs or Netcracker are already incorporating generative AI, but NVIDIA's platform offers a more horizontal and open approach. For hyperscalers (AWS, Azure, Google Cloud), competition intensifies: all offer AI services for telco, but NVIDIA bets on a specific platform with synthetic data and domain models.

For end users, the impact will translate into more stable networks, faster incident response, and potentially new AI-based services, such as proactive customer care or personalized data plans. For example, an autonomous agent could detect a usage pattern indicating a user is about to exceed their data limit and offer them a higher plan before they complain. This improves experience and reduces churn.

What readers should know

• Synthetic data generation is a key trend for sectors with sensitive data, and NVIDIA is driving it with open-source tools like NeMo. This democratizes access to high-quality training data without privacy risks.
• SoftBank Corp. is the first announced real-world use case, but other operators are expected to follow, especially those already collaborating with NVIDIA on AI (such as AT&T, Verizon, or Deutsche Telekom).
• Security and human control are central: agents do not act without policy supervision. Secure runtimes ensure critical decisions are reviewed by humans, mitigating risks of bias or errors.
• This announcement takes place at DTW Ignite 2026, a benchmark event for telecom digital transformation organized by TM Forum. There, NVIDIA and its partners are demonstrating these components live, accelerating practical adoption.

"Automation is no longer the goal, but the launchpad toward autonomy." — NVIDIA Blog

In summary, NVIDIA is laying the groundwork for telecom networks to transition from being managed by humans with automation assistance to being autonomous, with AI agents making real-time decisions. This will not only improve operational efficiency but also enable new business models and services. However, the path requires overcoming challenges of integration, governance, and trust. The coming months will be crucial to see if operators massively adopt this vision or if competitive alternatives emerge.