NTT DOCOMO Deploys Nokia MantaRay AutoPilot on Public Cloud in a World First for 5G Network AI

For most of the history of mobile networks, the people who kept them running did something closer to gardening than engineering: constantly adjusting settings, adding rules, and tuning configurations to keep performance within acceptable bounds across thousands of base stations. NTT DOCOMO just handed that garden to an AI and told it to manage itself.

NTT DOCOMO announced on June 22 that it has deployed Nokia’s MantaRay AutoPilot system on a public cloud to optimize its commercial mobile network — the world’s first deployment of MantaRay AutoPilot on a public cloud for a live network, and Japan’s first deployment of the system at all. Effective June 19, the AI continuously monitors DOCOMO’s base stations, determines the optimal configuration changes to meet operator-defined quality targets, and issues those directives to the underlying network in cycles as short as 15 minutes — without any human intervention in the design or execution of those changes.

The distinction between this deployment and the November 2025 MantaRay SON rollout DOCOMO announced last year is not incremental. MantaRay SON automated the execution of network optimization based on rules that human engineers designed in advance; MantaRay AutoPilot eliminates the rule-design step entirely, replacing it with intent-based AI that determines the rules itself from a high-level quality objective. Engineers no longer configure what the network should do — they specify what outcome the network should achieve, and the AI figures out the rest.

What Happened

According to DOCOMO’s official press release, MantaRay AutoPilot operates through a continuous four-step cycle running on DOCOMO’s public cloud infrastructure. First, real-time performance data — communication quality, traffic volume, and related metrics — is collected from base stations managed by MantaRay SON. Second, using that data alongside the operator-defined intent (for example, “maintain communication speeds in a specific area above XX Mbps”), the AutoPilot AI analyzes network status, learns from the patterns it observes, and determines autonomously which parameters require adjustment and when those adjustments should be made. Third, the system issues optimization directives to MantaRay SON at intervals as short as 15 minutes. Fourth, MantaRay SON receives those directives and remotely reconfigures the targeted base stations to execute the optimization.

The previous configuration, which relied solely on MantaRay SON, ran optimization cycles daily, and required human engineers to manually analyze which configuration changes were needed and measure their effectiveness after implementation. AutoPilot compresses that multi-day manual cycle into a fully automated 15-minute loop, removing the human from both the analysis and the execution step.

The Mechanism: From Rule Execution to Intent Resolution

The shift from MantaRay SON to MantaRay AutoPilot represents a qualitatively different approach to network management, not simply a faster version of the same approach. The older system automated what engineers told it to do: given a set of pre-designed parameters and configuration policies, it could detect when conditions drifted outside those specifications and execute the pre-defined corrective action without human intervention. The cognitive work of deciding what the corrective action should be, however, still sat with the engineers.

AutoPilot removes that step. Rather than encoding expert knowledge into a set of rules that the system executes, operators define a desired outcome — what DOCOMO calls an “intent”, such as maintaining minimum throughput above a specified threshold in a given area — and the AI constructs the logic for achieving that outcome on its own. It monitors the network continuously, infers the relationship between parameter configurations and the quality outcomes they produce, and updates its approach as conditions change throughout the day. This matters in practice because congestion patterns across locations and times of day are sufficiently dynamic and complex that encoding them as static rules quickly becomes a maintenance problem as those patterns evolve: a rule set that performed well during weekday morning commute hours may perform poorly during an evening event at a stadium in the same cell area. AutoPilot’s continuous learning loop is designed to adapt to those shifts without requiring engineers to redesign the policies each time.

The Backstory

DOCOMO’s deployment of MantaRay AutoPilot builds on the foundation of Nokia’s MantaRay SON deployment announced in November 2025, which itself was the first implementation in Japan of Nokia’s 5G base station automation integrated with DOCOMO’s operational systems. That November deployment already marked a meaningful step toward autonomous network operations: MantaRay SON introduced closed-loop automation across DOCOMO’s multi-vendor LTE and 5G radio access network, enabling configuration changes to be detected and executed automatically without human intervention once the system was given its parameters. DOCOMO’s stated goal through both deployments is to reach TM Forum Autonomous Networks Level 4, defined by the global telecom standards body as the capability for AI to predict network conditions and manage the network entirely autonomously, without human intervention even at the policy-design stage.

The decision to run AutoPilot on a public cloud rather than dedicated on-premises hardware carries its own significance beyond the technical milestone. DOCOMO’s press release is explicit that the public cloud architecture was chosen to bypass the hardware procurement lead times that would otherwise have constrained the deployment timeline, and that it enables integration with external AI platforms — a future capability DOCOMO says it plans to exploit. In the broader context of Japan’s accelerating AI infrastructure ambitions, DOCOMO’s cloud-first approach to deploying frontier AI on critical national network infrastructure also fits a wider regional pattern of established telecoms leveraging cloud-native AI to move faster than traditional hardware cycles allow.

Nokia’s MantaRay AutoPilot product has prior commercial deployments outside Japan: Nokia’s president of mobile networks, Tommi Uitto, publicly cited its deployment during the annual Hajj pilgrimage in Mecca, Saudi Arabia, where it executed roughly five million configuration changes across the three-day event — an environment with extremely concentrated and rapidly shifting demand on a short timeline. DOCOMO’s is, however, the first deployment running on a public cloud and the first in Japan, making the specific combination — AutoPilot plus public cloud infrastructure on a permanent commercial network — genuinely new. The decision to run on public cloud rather than dedicated silicon also resonates with a broader infrastructure trend: just as cloud-native AI inference chips like Microsoft Maia 200 are displacing dedicated hardware in hyperscale data centers, DOCOMO’s cloud-first deployment of AutoPilot reflects the same underlying logic — software-defined, elastically scalable AI is faster to deploy and cheaper to iterate on than purpose-built hardware.

Reactions

DOCOMO’s press release frames the deployment in terms of the network quality improvements customers will experience rather than the technical milestone itself: by running the optimization cycle every 15 minutes rather than daily, the system can respond to localized congestion events as they develop — a stadium event, an unexpected crowd at a transit hub, a sudden shift in traffic distribution — rather than waiting until the following day’s optimization window to apply corrective changes. DOCOMO says it will continue evaluating the system’s effectiveness in commercial operations with a view to advancing “AI-driven network management to deliver higher-quality communication services to customers.”

Nokia has positioned MantaRay AutoPilot as the key differentiator in its broader SMO framework, with the company stating its solution is the only one in the market that currently reaches TM Forum’s Autonomous Networks Level 4. The DOCOMO deployment is the most visible third-party commercial evidence yet of that claim being exercised in a live network environment at national scale.

The Dispute: Autonomy and Accountability in Critical Infrastructure

The deployment raises a question that neither DOCOMO nor Nokia’s materials address directly: what happens when an intent-based AI system makes optimization decisions in a live commercial network that produce unexpected outcomes? The shift from rule-based to intent-based network management removes a layer of explainability. In a rule-based system, when a configuration change causes a problem, an engineer can trace back through the policies to understand why that change was made. In an intent-based AI system that autonomously determines its own parameter logic, the path from intent to outcome passes through a model whose internal reasoning is considerably harder to inspect after the fact.

Japan’s national telecommunications regulatory framework, as well as DOCOMO’s own obligations as Japan’s largest mobile operator by subscriber count, will need to accommodate the reality that network decisions affecting over 93 million subscribers are now being made by an AI system operating without human review of individual optimization choices. DOCOMO’s plan to reach Level 4 autonomous network operations — where “AI can predict network conditions and autonomously manage the network without human intervention” — takes that accountability gap further rather than narrowing it. In the same week that Japan’s government published a revised AI Basic Plan explicitly addressing the legal challenges posed by high-capability autonomous AI systems, DOCOMO is deploying exactly that category of system on critical national infrastructure.

What Happens Next

DOCOMO says it will continue evaluating the system’s commercial effectiveness — an acknowledgment that a June 19 go-live is still early-stage operational experience rather than a concluded proof of concept. The clearest signals to watch will be whether DOCOMO expands the coverage of AutoPilot-managed network areas over the coming months, and whether it publishes measurable performance data from the deployment that can be independently evaluated against the prior SON-only baseline. DOCOMO’s stated intention to integrate AutoPilot with “various cloud-based AI platforms” is the more speculative long-term direction, but it signals the company is already planning what the next layer of AI integration looks like beyond autonomous parameter optimization.

Why It Matters

The world’s first public cloud deployment of MantaRay AutoPilot on a live commercial network is a meaningful milestone in the long-running industry project of reducing the human labor required to manage 5G infrastructure at scale. Japan’s mobile networks are among the most technically sophisticated and densely utilized in the world, which makes DOCOMO a credible test environment for whether this generation of intent-based AI can handle the complexity and variability of real national-scale network management rather than controlled pilot conditions. The 15-minute optimization cycle is not just faster than daily manual intervention — it represents a different order of responsiveness that, if it performs as described, would make DOCOMO’s network more adaptive to real-time demand patterns than any human-configured equivalent could practically be. Whether the autonomous AI can be reliably held accountable for the decisions it makes in doing so is the question that will follow this milestone into its next phase.

Sources

NTT DOCOMO press release (June 22, 2026); Nokia newsroom (November 2025); SDxCentral; Nokia MantaRay AutoPilot product page.