ZTE and China Mobile Unveil Network Graph Model at DTW Ignite 2026, Targeting Level-4 Autonomous Networks

The telecom industry has had a definition of a fully autonomous network for years. It has been called Level 4. Until this week in Copenhagen, no major vendor had demonstrated a credible production path to it at scale.

ZTE Corporation and China Mobile unveiled a Network Graph Model at TM Forum’s DTW Ignite 2026 in Copenhagen on June 25 — a joint development that integrates knowledge graphs with large language models specifically to eliminate the hallucinations that have made LLM-based autonomous network management unreliable in production environments. The announcement anchors ZTE’s broader Level-4 autonomous network strategy at DTW Ignite, which the company is presenting as a shift from traditional task automation, where AI executes predefined procedures, to what it calls intrinsic intelligence: multi-agent systems capable of making end-to-end decisions across network domains without human intervention.

DTW Ignite 2026, TM Forum’s flagship global telecom event, is running this week in Copenhagen under the theme “The Future. Faster.” with three summits focused on Autonomous Networks, AI and Data, and Composable IT and Ecosystems. It is the event where the industry’s most ambitious claims about autonomous network operations get stress-tested against what operators say they actually need.

What Happened

According to ZTE’s official announcement from DTW Ignite 2026, the Network Graph Model was developed through China Mobile’s “Co-Innovation+” Autonomous Network Open Lab. The model integrates knowledge graphs — structured representations of domain-specific relationships and entities — with large language models in order to anchor LLM reasoning in verifiable network topology and configuration data. The goal is to prevent the kind of hallucinated outputs that make current LLM-based approaches unreliable for production network management, where an incorrect recommendation about a configuration change could degrade service for millions of users. The integration gives the LLM a structured knowledge base to reason against rather than relying solely on its training data.

In a Q&A with TM Forum Inform published ahead of DTW Ignite, Peng Zheng, ZTE’s General Manager of Service and Data Intelligence, outlined the company’s L4 autonomous network strategy in more operational detail than the formal announcement covers. The strategy rests on three interconnected elements: a foundation AI layer built around the Nebula Communication Large Model, which is trained on telecom-domain data to reason about network states and operations; large-scale deployment of high-value automation scenarios in operations and maintenance and user experience optimization, where ZTE is targeting more than 30 benchmark commercial scenarios for 2026; and intent-driven autonomy across the full network lifecycle, meaning operators specify desired outcomes in natural language rather than writing configuration rules.

The Mechanism: Why Agentic Multi-Agent Collaboration Is the Missing Piece

The TM Forum’s Autonomous Networks framework defines Level 4 as the capability for AI to proactively handle network issues and make decisions autonomously across multiple domains, with the network adapting to predicted conditions and optimizing for user experience rather than simply reacting to faults after they occur. The levels below L4 can be reached with single-domain automation: a system that monitors the RAN layer and automatically adjusts cell parameters to maintain coverage is performing valuable automation, but it is operating within a single network function domain with a limited and predictable action space.

Level 4 requires coordination across domains simultaneously: the RAN, core network, transport, operations, and business systems must be able to communicate and act in concert under AI direction. A network fault that originates in the transport layer can cascade into degraded RAN performance, which affects user experience KPIs, which trigger service-level penalty clauses, which require automated regulatory reporting. Handling that end-to-end workflow autonomously requires multiple specialized AI agents, each expert in its own domain, to collaborate under a coordination layer that can route tasks, resolve conflicts, and manage the overall closed loop without human intervention at any step. That is what ZTE means by multi-agent collaboration as a Level-4 enabler, and it is why the Network Graph Model matters: without a shared, hallucination-resistant knowledge representation, the agents in different domains cannot maintain a consistent and accurate shared understanding of what the network is actually doing.

The Backstory

ZTE’s trajectory at DTW Ignite illustrates how quickly the autonomous network conversation has moved in twelve months. At DTW Ignite 2025, held at the same Copenhagen venue a year ago, ZTE unveiled its AIR Net Solution built on a tri-engine architecture and open-sourced the Co-Sight Framework for cross-vendor interoperability in Level-4 autonomous networks. The announcement was ambitious but positioned primarily at the architectural level, describing what the components of L4 autonomy should look like rather than demonstrating it in a production network. At DTW Ignite 2026, the joint announcement with China Mobile is a production deployment story rather than an architecture proposal: the Network Graph Model was developed in a real Co-Innovation Lab and is being presented alongside ZTE’s plan for more than 30 commercial scenario deployments in 2026.

The DTW Ignite event itself has become a relevant benchmark for how seriously the telecom industry takes autonomous network claims. NVIDIA is demonstrating AI tools for telco autonomy at the same event, including synthetic data generation, telecom-domain models, and secure agent runtimes. Ericsson is showcasing agentic AI in network fulfillment, Day 2 operations, and 5G dynamic slicing. Blue Planet is demonstrating end-to-end Level-4 architecture for service lifecycle management. The competitive density at DTW Ignite 2026 means ZTE’s Network Graph Model announcement is being made in front of an audience that can immediately compare it against rival approaches from the other booths, which is a more rigorous public test of the claim than a standalone press release would face. The same autonomous network evolution is also driving Nvidia’s sweeping AI infrastructure deals across South Korean carriers and chipmakers — a deployment that represents a different but complementary approach to the same L4 target that ZTE is pursuing in Copenhagen.

Reactions

Peng Zheng’s framing of L4 autonomous networks emphasized the commercial case alongside the technical one: the network operations systems ZTE is targeting — zero-wait and zero-touch across service creation, fulfillment, and Day 2 O&M — are designed to advance the industry’s shift from “network-centric” to “user-experience-centric” operations, which means the value metric is user-visible service quality rather than internal network efficiency metrics. That framing is strategically significant: it positions autonomous network investment not as a cost-reduction exercise but as a competitive differentiator that could allow carriers to charge for guaranteed experience rather than simply competing on coverage and price per gigabyte.

China Mobile’s Co-Innovation Lab partnership gives the ZTE announcement institutional weight beyond a single vendor’s product claim. China Mobile is the world’s largest mobile operator by subscriber count, and a joint announcement from its dedicated autonomous network research programme carries a different credibility weight than a demo from a vendor’s internal engineering team alone. The fact that China Mobile chose to announce this at DTW Ignite rather than at a China-domestic event also signals an intent to establish international standards-body credibility for the approach, rather than treating it as a China-only capability.

The Dispute: L4 in the Lab Versus L4 in Production

The autonomous network community has a well-established scepticism of Level-4 claims that is worth applying to ZTE’s DTW Ignite announcement. The TM Forum’s own maturity assessments have consistently found that the gap between what vendors demonstrate at conferences and what is actually running in production networks at certified Level-4 maturity is significantly wider than press releases suggest. The Network Graph Model addresses one specific failure mode — LLM hallucinations in network management — but Level-4 autonomy requires solving dozens of interdependent problems simultaneously, from intent translation accuracy to cross-vendor API standardisation to liability frameworks for automated decisions that cause service degradation.

ZTE’s target of 30 benchmark commercial scenarios for 2026 is measurable and would represent genuine progress if achieved, but “benchmark commercial scenario” is not the same as “Level-4 certified network operation.” The TM Forum’s formal certification pathway for autonomous network maturity levels requires independent assessment against standardised test cases rather than vendor-defined benchmarks. Whether ZTE’s Co-Sight Framework and Nebula Communication Large Model produce deployments that pass that independent bar — rather than deployments that perform well against ZTE’s own test cases — is the real test. Research on how enterprises misunderstand and underestimate their AI control gaps suggests that accountability for AI-driven decisions is a broader unsolved problem, not specific to telecoms — and Level-4 network autonomy raises it in one of the most consequential operational environments possible.

What Happens Next

ZTE’s 30 benchmark commercial scenario target for 2026 is the clearest near-term marker to watch. The company has said it will implement these scenarios with operator partners, which means there should be verifiable carrier-side disclosures of L4 autonomy deployments over the next six to twelve months rather than only ZTE-side product announcements. The Co-Sight Framework’s open-source release also creates an external validation pathway: if the framework actually supports cross-vendor interoperability as claimed, third-party deployments using it alongside non-ZTE network equipment should begin appearing in the TM Forum’s Innovation Hub trials over the same period, providing an independent measure of whether the architecture delivers what the Copenhagen announcement promised.

Why It Matters

The Level-4 autonomous network target is not primarily a technology aspiration; it is an economic one. Carriers worldwide are under pressure to maintain or grow revenue while managing networks that are growing more complex and more expensive to operate in direct proportion to the AI-driven traffic growth they’re handling. Autonomous network management is the most credible path to breaking that cost spiral, and the vendors that can demonstrate verified L4 deployments first will have a significant commercial advantage as carriers make their next-generation network management platform decisions. ZTE’s DTW Ignite 2026 announcements — combining the Network Graph Model, the Nebula LLM, and a 30-scenario commercial deployment roadmap — represent its most concrete claim yet that L4 is within reach in production rather than only on a conference stage. The question is not whether anyone believes it in principle; it is whether the 30 scenarios actually deploy and whether their performance meets the TM Forum’s independent standard for Level 4, rather than the vendor’s own. That answer will come from operator disclosures over the next year. The stakes are high: AI companies are now valued in the hundreds of billions to trillions of dollars in large part because their services require the kind of reliable, always-on network infrastructure that Level-4 autonomous management is supposed to deliver.

Sources

ZTE Corporation (zte.com.cn newsroom); TM Forum Inform; Converge Digest; NVIDIA Blog (DTW 2026); Blue Planet / The Fast Mode.