Most mobile carriers still treat radio access network planning like a construction project: survey the site, dispatch an engineer, manually tune the antenna, repeat a few hundred thousand times. Rakuten Mobile skipped most of that — and cut its 5G buildout cost by 40% in the process.
Building and running a national mobile network has always meant throwing people at the problem: RF engineers tuning individual cell sites, NOC technicians watching dashboards for outages, capacity planners running spreadsheets to predict next quarter's traffic. That model doesn't scale once a network has hundreds of thousands of cells and traffic patterns that shift hour by hour. There's a faster way to run a network — and it doesn't involve hiring more engineers.
Rakuten Mobile and its infrastructure arm, Rakuten Symphony, deployed an AI-powered RAN Intelligent Controller (RIC) across Rakuten's entire 4G and 5G Open RAN network in Japan — more than 200,000 cells covering 96% of the population. The result: roughly 40% lower capital expenditure, about 30% lower operating expenditure, and up to 25% energy savings on optimized base stations. Here's what the system actually does, and why every carrier watching from the sidelines should care.
What a "Self-Tuning" Network Actually Means
A RAN Intelligent Controller sits between the radio hardware and the software that decides how it behaves. Rakuten's RIC platform — paired with RAFT (Rakuten AI for Telecom), described internally as the industry's first AI-native, enterprise-grade generative AI platform for telecom — ingests live network telemetry and runs small AI applications called rApps on top of it. Those rApps analyze subscriber traffic patterns in real time, predict short-term demand, and adjust base station behavior automatically — shifting power, reallocating spectrum, and rebalancing load without a human approving each change.
This matters most for Network Operations Managers and OSS/BSS engineers running networks that have outgrown manual capacity planning — typically carriers above a few thousand cell sites, where the volume of micro-decisions (which sector to power down at 3am, which cell needs more spectrum during a stadium event) is already beyond what a NOC team can review in real time.
Inside the AI Loop That Runs the Network
The architecture behind Rakuten's deployment breaks down into a repeatable loop rather than a one-time project:
- Telemetry ingestion — the Non-RT and near-RT RIC continuously pull traffic, signal quality, and energy data from every cell site.
- rApp analysis — third-party and in-house rApps (including partners like AirHop Communications) process that telemetry for patterns: congestion, interference, idle capacity.
- Autonomous adjustment — the RIC pushes configuration changes back to the network — power levels, sectorization, spectrum allocation — without manual ticketing.
- Zero-touch site commissioning — new cell sites are provisioned against a standard template instead of being hand-configured, cutting commissioning from weeks to minutes.
The pattern repeats continuously, every few minutes, across all 200,000+ cells — something no human NOC team could do at that frequency or scale.
The Results: 40% Less CapEx, 30% Less OpEx
According to Rakuten Symphony's own published case study, the AI-powered RIC deployment delivered roughly 40% savings on capital expenditure and around 30% on operating expenditure across Rakuten's Japanese network, alongside up to 25% energy savings on optimized base stations under ideal conditions.
Rakuten isn't an isolated case. NVIDIA's 2026 State of AI in Telecom survey found that 48% of telecom enterprises have now deployed agentic AI in at least one core business function — nearly double the cross-industry average of 26% — with network automation overtaking customer experience as the leading area for AI investment and ROI.
The gap is widening for operators that wait. Carriers still running manual capacity planning are competing against networks that reconfigure themselves every few minutes; the lag isn't just operational cost, it's the time-to-market on every new coverage area or capacity upgrade.
The AI Stack Behind Rakuten's Network
This isn't off-the-shelf software a single engineer installs — it's carrier-grade infrastructure bundled into an operator's network deployment.
| Tool | Role in This Deployment | Access Model | Notes |
|---|---|---|---|
| Rakuten Symphony RIC Platform | Core RAN Intelligent Controller — Non-RT and near-RT control loops | Bundled free with Symworld app purchase | Multi-vendor Open RAN compatible |
| RAFT (Rakuten AI for Telecom) | Generative AI layer for autonomous operations and OSS/BSS decisioning | Enterprise / carrier license | [REQUIERE VERIFICACIÓN] — standalone pricing not publicly disclosed |
| Symworld | Underlying cloud-native telecom platform (data lake, AI platform, CI/CD) | Enterprise / carrier license | RIC is bundled on top of a Symworld app purchase |
None of this is a weekend integration — but the architecture pattern (telemetry in, AI decisioning, autonomous action out) is the same blueprint operators of any size are now applying with vendor-agnostic Open RAN stacks.
Who Should Be Watching This Case Study
This is most directly relevant to Network Operations Directors, OSS/BSS architects, and CTOs at mid-size to large mobile carriers evaluating Open RAN or planning a multi-year 5G buildout. It's less relevant if you're running a small regional network where manual planning still covers your cell count comfortably, or if your infrastructure is locked into a single-vendor RAN stack that doesn't support third-party rApps yet.
Explore More AI Case Studies on Sityos
This case sits alongside a growing library of real-world AI automation deployments on sityos.com. If Rakuten's approach to autonomous networks is relevant to your stack, you'll also want to look at:
- Vodafone/Fastweb's multi-agent AI — how a different carrier automated customer service instead of network operations, hitting 82% autonomous resolution.
- Duke Energy's self-healing grid — the same "autonomous, self-correcting infrastructure" pattern applied to utilities instead of telecom.
- Mastercard's real-time fraud detection — another case of AI making thousands of autonomous decisions per second at infrastructure scale.
Get the Next Case Study Before It's Published
New, fully-sourced AI automation breakdowns — Tutorials and Case Studies — go live on sityos.com every week.
Frequently Asked Questions
Is Rakuten's RIC platform available to other carriers, or is it Rakuten-only?
It's commercially available — Rakuten Symphony sells the underlying Symworld platform to other operators, with the RIC bundled in. Rakuten Mobile is the first and largest production deployment, but the company has operations and customers across Japan, the US, Singapore, India, South Korea, Europe, and the Middle East/Africa region.
Does this replace network engineers, or just change what they do?
It changes the job more than it eliminates it. Routine reconfiguration and capacity tuning move to the AI loop; engineers shift toward exception handling, rApp tuning, and the small share of decisions the system flags as low-confidence.
How long does a deployment like this take?
Rakuten's own nationwide rollout took years to reach 200,000+ cells, but that includes building the physical network itself. The AI/RIC layer is what made commissioning each new site take minutes instead of weeks once the platform was in place — the AI deployment timeline is closer to weeks-to-months for an operator adding it onto existing infrastructure, not a years-long project on its own.
Autonomous networks aren't a future-state slide in a vendor deck anymore — they're running production traffic for 96% of a country's population today. The TM Forum still puts most operators at autonomy levels 1-3 out of 5; the carriers closing that gap first are the ones setting the cost baseline everyone else gets compared against.