Photo by Brett Sayles on Pexels

Rakuten Mobile’s Open Architecture Becomes the Engine for AI-Driven Network Intelligence

When Rakuten Mobile launched its fully virtualized, cloud-native mobile network in Japan in 2020, the telecom world watched with a mixture of admiration and skepticism. Building a greenfield carrier entirely on open, software-defined principles was an audacious bet. Four years on, that bet is paying dividends in ways that go far beyond cost savings — Rakuten is now using its Open Radio Access Network (O-RAN) foundation as the launching pad for an ambitious artificial intelligence strategy that could redefine how mobile networks are designed, managed, and optimized.

Unlike incumbent operators who must retrofit AI capabilities onto decades-old proprietary hardware and siloed network domains, Rakuten Mobile’s architecture was purpose-built for programmability. That distinction, industry analysts say, is not a minor operational detail — it is a fundamental competitive advantage in the age of AI-native telecommunications.

Why O-RAN Is the Ideal Foundation for AI Integration

Traditional RAN deployments have long been defined by proprietary vendor lock-in, where hardware and software from a single supplier operate as a black box. This model is deeply hostile to AI integration: machine learning models require access to granular, real-time data streams, and they need the freedom to act on insights by dynamically reconfiguring network parameters. In a closed system, that level of access and agility is simply not available.

O-RAN changes the equation entirely. By disaggregating the RAN into open, standardized components — the Radio Unit (RU), Distributed Unit (DU), and Centralized Unit (CU) — and introducing the RAN Intelligent Controller (RIC) framework, O-RAN creates explicit “hooks” where AI and machine learning applications can observe network conditions and execute real-time or near-real-time interventions.

Rakuten’s network is built on exactly this architecture. Its software-defined, cloud-native stack means that every network function generates accessible telemetry, and every parameter is, in principle, tunable through software. This is the raw material that AI systems need to function effectively — and Rakuten already has it baked into its infrastructure by design.

Autonomy with Guardrails: Rakuten’s Measured AI Philosophy

Despite the excitement surrounding fully autonomous, self-healing networks, Rakuten Mobile has adopted what it describes as a “guardrailed autonomy” approach to AI deployment. Rather than handing complete control to machine learning algorithms, the operator is implementing AI-driven decision-making within carefully defined operational boundaries — a philosophy that reflects both technical prudence and regulatory awareness.

This approach mirrors broader industry sentiment. While the vision of a zero-touch network is compelling, operators globally have been cautious about allowing AI systems to make high-impact changes — such as modifying handover parameters or reallocating spectrum — without human oversight or hard limits on intervention scope. The consequences of an unconstrained AI making erroneous decisions in a live network serving millions of subscribers are simply too significant to ignore.

Rakuten’s framework appears to position AI as a powerful co-pilot rather than a fully autonomous agent, at least in the near term. xApps and rApps deployed on the RIC platform can analyze KPIs, predict congestion, and recommend — or automatically execute — optimization actions, but always within predefined policy envelopes set by network engineers.

Technical Building Blocks: RIC, xApps, and Cloud-Native AI Pipelines

At the technical heart of Rakuten’s AI strategy is its RIC platform, which supports both the Near-Real-Time RIC (operating on 10ms–1s latency loops) and the Non-Real-Time RIC for slower, policy-level intelligence. These platforms host a growing ecosystem of AI-powered applications targeting specific network challenges: interference management, traffic steering, energy efficiency optimization, and predictive maintenance.

Rakuten’s cloud-native underpinning — built on Kubernetes orchestration and microservices architecture — means AI inference workloads can be containerized, scaled elastically, and deployed across distributed edge compute nodes where latency-sensitive decisions need to be made close to the radio edge. This is a critical capability as AI use cases evolve from centralized analytics toward real-time, distributed intelligence.

The operator has also invested heavily in its internal data platform, recognizing that high-quality, labeled training data is as important as the AI models themselves. Network AI is only as good as the data pipelines that feed it, and Rakuten’s fully digital, software-driven infrastructure simplifies the process of collecting, normalizing, and ingesting the massive telemetry volumes that modern ML systems require.

Implications for the Broader Telecom Industry

Rakuten Mobile’s trajectory carries important lessons for the global telecom industry. For operators still running traditional, vendor-locked RAN infrastructure, the path to AI-native networking is considerably steeper. Migrating to open, disaggregated architectures requires significant capital investment, organizational transformation, and a willingness to work with a more complex, multi-vendor ecosystem.

Yet the competitive pressure to do so is intensifying. AI-driven network optimization promises measurable benefits: reduced operational expenditure through automation, improved spectral efficiency, lower energy consumption — a critical priority as telcos face mounting pressure to hit sustainability targets — and ultimately, a better quality of experience for end users.

The GSMA and O-RAN Alliance have both highlighted AI and ML as central pillars of next-generation network evolution, and standardization bodies are actively developing specifications to ensure interoperability of AI/ML models across multi-vendor O-RAN deployments. Rakuten’s real-world implementation provides valuable proof-of-concept data that these standards bodies and commercial operators alike will study closely.

Looking Ahead: A Blueprint for AI-Native Carriers

Rakuten Mobile’s journey from maverick greenfield operator to AI-native network pioneer is a story that the telecom industry will be telling for years. As the carrier continues to mature its AI capabilities — and as it scales its open-source Rakuten Communications Platform (RCP) for export to other operators globally — its architecture and operational playbook are becoming increasingly relevant far beyond Japan’s borders.

The broader takeaway is unambiguous: operators that invested early in open, software-defined infrastructure are now positioned to capture the full value of AI-driven networking. For those still on the sidelines, Rakuten’s example makes the cost of inaction clearer than ever. The intelligent network is no longer a distant aspiration — it is being built, right now, on a foundation of open interfaces and cloud-native design principles.