The tokenomics of telecom – how AT&T is re-engineering OSS/BSS with AI

RCR’s free virtual Telco AI Forum is taking place on June 16th. Register here to explore AT&T’s telco AI strategy in more depth.

Imagine for a second that a telecom network is like this massive, high-speed automated highway. Now, the Operation Support Systems (OSS) are the engineers making sure the asphalt stays smooth and the traffic lights stay perfectly synced. Meanwhile, the Business Support Systems (BSS) are kind of like toll booths, customer service desks, and subscription counters.

Now, historically, these two systems talked to each other by sending huge, clunky dumps of data—basically, the engineering equivalent of sending a carrier pigeon with a physical encyclopedia strapped to its leg. It was slow, expensive, and honestly, a huge bottleneck for any real innovation.

Enter tokenomics.

In today’s telecom world, tokens—both lightweight API security tokens and the data tokens used by Large Language Models (LLMs)—are like digital VIP passes. They condense complicated data, secure transactions, and streamline AI processes. For giants like AT&T, using tokens effectively isn’t just some experimental tech trend; it’s a multi-billion-dollar survival strategy.

The financial reality of telco AI

Every telecom exec wants to cash in on AI, but running generative AI models? It’s crazy expensive. LLMs charge by the token—which are basically pieces of words. If your BSS or OSS feeds raw, unoptimized network data into AI, well, you’ll go broke faster than a teenager on an international roaming plan.

AT&T cracked this nut. By rolling out an enterprise-wide token and AI strategy, they handle a staggering 27 billion tokens daily. Instead of getting stuck in endless development loops, their AI generates a 5-fold return in free cash flow within the first year. This smart strategy helped them save 1billion in total costs in 2025 and they’re on track to hit 4 billion by 2028.

The blueprint: Three methods of token optimization

To make these numbers real, operators must ditch monolithic data pipelines and adopt a decoupled, event-driven architecture.

Method A: OSS lightweight telemetry tokenization
OSS manages a huge network infrastructure. Usually, when a fiber line goes down, systems flood the central data center with verbose logs, driving up cloud storage and compute costs.
Here’s the tokenized fix: parsing moves to the edge. Local software intercepts raw logs and turns long, wordy strings into small, standardized integer keys before sending them to a central message broker. By shrinking heavy data hashes into simple codes, this method cuts data payload size by 80-85%, slashing cloud fees and saving millions.

Method B: BSS contextual security state tokens
BSS deals with billing and customer care—the place where your patience usually runs out. We’ve all been there: calling support, authenticating, getting transferred, and repeating details again. Every second an agent spends re-verifying costs money.
The solution? A cryptographically signed security token that follows the user’s digital footprint. When you log into a mobile app, this secure token carries your identity and recent billing glitches through to the live agent without a database re-check. This shaves off 52 seconds per call, saving about $4.3 million annually in a big call center.

Method C: AI Cost optimization via token pruning
Raw customer billing and network data are stuffed with useless padding. If you feed this unfiltered text directly to an LLM, it can cost upwards of 0.05per subscriber query. Telecoms build a text−filtering layer between their BSS database and AI Gateway to strip out boilerplate, repetitive formatting, and empty headers. This local pruning cuts input size by over 650.002 per query. For 10 million subscribers, that means going from 500,000 down to 20,000, turning an expensive experiment into a money-maker.

How the core network powers the transformation

Software can’t run faster than the network plumbing beneath it. To support token-driven wins, AT&T completely revamped its Core Network, moving from rigid, proprietary hardware to a cloud-native, software-defined 5G Standalone (5G SA) core. Their network became a distributed AI engine.

• VIP lanes via network slicing: Using the User Plane Function (UPF) in the 5G core, AT&T creates isolated, logical sub-networks on the same physical infrastructure. Critical OSS telemetry tokens and real-time BSS fraud data travel on these dedicated slices, guaranteeing ultra-low latency and zero packet loss, bypassing regular consumer traffic entirely.
• Trimming data fat at the edge: Pushing UPF directly to the network edge via Multi-access Edge Computing (MEC), AT&T’s local edge core intercepts traffic at the cellular switching site. This is where their custom Network Foundation Model (NFM) lives. The edge core filters and tokenizes data locally, sending only lightweight summaries to the cloud. This enables their radio propagation tools to run 4,000 times faster than old systems.
• Eliminating silos with control plane harmonization: AT&T moved its core network control plane onto a standardized, software-defined API framework, funneling clean data into a unified lake house. Because the data streams are uniform, AT&T can skip expensive commercial LLMs for routine tasks and route them to highly efficient internal Small Language Models (SLMs). This cut generative AI operating costs by roughly 90%.

The token balance sheet

By treating tokens as a core financial and architectural metric, modern telecom operators are cutting structural waste. AT&T’s blueprint shows that when you turn your core network into an open, programmable, token-friendly asset, you don’t just build a faster network, you build a seriously profitable enterprise AI engine.

RCR’s free virtual Telco AI Forum is taking place on June 16th. Register here to explore AT&T’s telco AI strategy in more depth.