- Orange and France’s CEA have launched a five-year research lab focused on “semantic communications”
- The idea is that transmissions are successful once meaning — not exact data — is understood
- The approach aims to cut data volumes moving across networks, with potential knock-on effects for energy use and infrastructure costs
Orange is testing the limits of AI in network operations by launching a joint research lab with France’s CEA to trial semantic communications. Their goal is to determine whether networks can move beyond simply transporting bits toward transmitting meaning.
The five-year initiative, called “AI-Native Communications,” is focused on what both organizations say could be a fundamental shift in how data is handled across telecom infrastructure.
Today’s networks are built on strict bit-level accuracy — any dropped or altered packet (or data) is treated as an error. The model Orange and CEA are exploring flips that approach. Instead of requiring perfect replication, transmissions would be considered successful once the intended meaning is understood by the receiver, according to the press release about the news.
"Today's networks obsess over transmitting every bit perfectly," Eric Hardouin, Orange fellow and VP, Networks and Infrastructures Research at Orange, told Fierce via email from the Vivatech show in Paris. "Semantic communications ask a different question: did the meaning get through? If you can shorten a text, transform an image into a single word, or strip out everything except what actually matters to the receiver, and nothing important is lost, why send the rest? It's a fundamental rethink of what a network is actually for. Not all communications will work this way, of course. Identifying exactly where it makes sense is one of the core goals of the lab we've just launched," he said.
That change could have real implications for how much data networks need to carry. By transmitting only information that contributes to meaning, rather than every byte, semantic systems could sharply reduce data traffic and the overhead required to move it, Orange said.
Efficiency gains will depend enormously on the application, according to Hardouin. "We're not yet at a point where we can put a single number on the network-wide impact. What we can say is that the potential is significant, and pinning down the most valuable use cases is exactly what we're working on."
The rise of AI-driven networks
Orange and CEA said this approach also aligns with the rise of AI-driven networks, where machines — not humans — are increasingly the primary communicators. The lab will focus in part on developing shared “semantic representations” between AI models, effectively creating a common language for machine-to-machine interactions inside future networks.
The operator says this work ties semantic communications directly to the emerging 6G conversation, where AI-native architectures are expected to be a core design principle. Industry research already frames semantic communication as a potential foundation for 6G, shifting performance metrics away from raw throughput toward task completion and context awareness.
However, the project is still early-stage research, so it's not on the table for the first wave of 6G standardization, Hardouin told Fierce. "Realistically, we're looking at standardization discussions picking up around 2030, with commercial deployments potentially from 2035 - a later evolution of 6G rather than its launch. But I'd caveat that heavily: the timeline could shift in either direction depending on how fast the science moves."
The role of semantics communications in 6G standards
Semantic communications, if adopted, would likely be part of that standards debate, particularly as operators and vendors compete to define what “AI-native” networks actually look like.
"Semantic communications and AI are deeply intertwined. You need AI to extract and manipulate meaning from data in the first place," he said. "Native AI in the network isn't just a nice-to-have; it's the foundation that makes semantic communications possible. We see this as genuinely disruptive technology with enormous potential, and that's precisely why Orange and CEA are joining forces now: to help shape both the technology and the standards that will follow."
The concept remains early-stage. Semantic systems require shared context and understanding between sender and receiver—something that’s straightforward in controlled AI environments but significantly harder across heterogeneous telecom networks at scale. Academic work also highlights open challenges around standardizing semantic metrics and ensuring consistent interpretation across systems.
That said, the challenges are "real and layered," Hardouin told Fierce. "How do you define protocols for meaning rather than bits? Where does the computing power sit, and how do you balance it against bandwidth? And critically, how do you make this a universal network feature rather than something each app has to build for itself? Those are hard, open questions. But we're doing the foundational work that has to come first."