Summary
- NTT DOCOMO selected Ericsson RAN Compute for baseband processing across 4G, 5G and 5G-Advanced.
- Ericsson presents the platform as suitable for centralised RAN and a foundation for AI-native networks.
- The vendor claims up to twice the capacity of its previous generation.
- Ericsson also claims energy use of less than half, without publishing the comparison conditions in the report.
- DOCOMO says additional processing capacity is needed to meet rising traffic while improving communication quality.
- Order value, model mix, deployment count, geography, schedule and measured network outcomes remain undisclosed.
Baseband capacity now constrains radio capacity
An antenna can only serve traffic that its baseband layer can schedule, encode and coordinate. As carriers add spectrum layers, Massive MIMO and more complex 5G functions, processing becomes part of the capacity ceiling.
DOCOMO’s selection therefore concerns the operating core of the radio access network, not a back-office server refresh. The platform must support old and new generations at the same time while maintaining timing and quality requirements.
Masafumi Masuda, who leads DOCOMO’s RAN design department, linked the upgrade to traffic demand and communication quality. That is a requirement statement, not proof that either outcome has already improved.
The “twice” and “half” claims need a test denominator
Ericsson says the platform offers up to two times the capacity of its previous generation and uses less than half the energy. “Up to” marks a boundary: the result depends on workload, configuration, software, temperature and which products are compared.
Capacity may mean supported cells, processing throughput or a configuration limit. Energy may refer to a unit at a specified load rather than the power of an entire centralised site.
Until methodology and field data are available, the numbers should remain attributed to Ericsson. They are procurement claims, not independent network measurements.
Centralisation can move energy rather than simply remove it
RAN Compute can pool baseband resources in a centralised architecture. Pooling can reduce idle capacity and make maintenance more consistent, but it also requires transport links, resilient facilities and power for the shared compute location.
A complete energy comparison would include radios, baseband, cooling and fronthaul or midhaul equipment. Measuring only the replacement shelf could overstate the system benefit.
The operating question is whether processing consolidates without creating transport bottlenecks or larger failure domains. DOCOMO has not published that architecture or its redundancy design for this selection.
One platform across generations reduces one kind of fragmentation
Japanese mobile networks will carry 4G and multiple 5G stages for years. A common compute family can simplify spares, software release management and technician training.
It can also deepen vendor dependence. Hardware support, software licensing, upgrade cadence and interface compatibility determine whether a common platform stays flexible over its service life.
The report does not state whether the order replaces existing equipment, supplements it or applies only to selected clusters. That distinction will set the real migration cost.
AI-native is an option, not a deployed workload
Ericsson describes RAN Compute as foundational for AI-native networks. Extra processing and centralised control can support optimisation models and automation nearer the radio layer.
The phrase does not establish that DOCOMO is already running AI applications on the selected equipment. It also does not disclose compute reserved for radio functions, model inference or third-party workloads.
Useful evidence would identify a production use case, its permissions, fallback path, latency and effect on network quality. Platform readiness and operating AI are different milestones.
Field results must close the procurement case
The December 2025 Massive MIMO rollout shows an existing Ericsson-DOCOMO deployment relationship. The new baseband selection extends that relationship into processing.
The public report leaves out value, quantities and schedule. Those omissions prevent an estimate of coverage, capital intensity or the speed of replacement.
The next useful disclosure is a before-and-after sample: throughput under comparable load, energy per carried terabyte, failure rates, processing headroom and operating cost. Until then, the selection is strategically clear but operationally unmeasured.