Summary

  • NTT reports an outdoor 100-metre sub-terahertz link at 2 Tbit/s; its conference-paper metadata gives 2.04 Tbps. The system uses eight OAM modes, two polarizations and adaptive spatial multiplexing, reaching 22 streams at most.
  • NTT still identifies long-term real-environment operation and integration with existing wireless systems as future work. The decision test is whether the link saves enough installation time while meeting service, recovery, power and maintenance requirements.

A record now has to earn an address in the network

NTT’s September demonstration changes a useful boundary. In 2023 the company reported 1.44 Tbit/s across one metre. The new outdoor trial reaches 100 metres and a paper rate of 2.04 Tbps. That is a substantial jump in reach as well as capacity. It is also still a research result. NTT says long-term operation in real environments and integration into existing wireless systems remain part of the practicalisation process; the release names AI data-centre links, building connections, mobile backhaul and temporary event or disaster relays as intended use cases, not deployments already in service. The 2026 announcement provides the experiment detail; the 2023 release provides the earlier one-metre comparison.

The strategic proposition is more specific than “wireless can match fibre”. A short radio bridge may be valuable where a trench, building access, a leased route or a temporary cable run would take too long or cost too much. But a peak rate earns no operating value until equipment can be installed, aligned, connected to the network control plane and kept available. The economic unit is not the laboratory bit per second. It is useful traffic delivered across a path, over time, for the cost and delay avoided.

What the 22 streams actually describe

NTT used two separated frequency ranges: 136.0–148.5 GHz and 151.5–164.0 GHz. Its OAM-MIMO arrangement combines eight orbital-angular-momentum modes, adaptive in-mode spatial multiplexing of up to two streams, and two polarizations. The complete configuration reached a maximum of 22 spatial streams. Those are not 22 OAM modes. The release describes a layered design in which analog waveguide circuits multiplex and separate the OAM modes while digital processing handles the remaining in-mode MIMO work.

Three changes extend the earlier short-range result. A smaller Multi-UCA Butler-matrix waveguide circuit produces and separates the modes. Imaging reflectors enlarge the effective aperture while keeping the wavefront structure needed for OAM. An alignment procedure searches for a high SINR state while accounting for interference between modes. NTT reports that, in a specified eight-element by two-ring example, this analog/digital split reduces digital computation load by several hundred-fold versus full-digital processing. That is a bounded computation comparison, not a claim that the complete radio consumes several hundred times less power.

NTT calls the transmission rate comparable to next-generation 800G and 1.6T optical formats. The comparison is informative about scale, not proof of equivalent service. The public release does not state application goodput, latency percentiles, packet loss under sustained traffic, total system energy, commercial price or long-term availability. A radio link also has to join two endpoints, their power and cooling, monitoring, switching and fault-recovery procedures. A 2-Tbit/s headline cannot stand in for those measurements.

The installation-time case

The strongest first market may be the place where building a fibre path is feasible but slow, disruptive or constrained. Between two data halls or buildings, a radio link could bridge a gap while civil work proceeds, serve as a temporary alternate route, or add capacity without waiting for a new duct. At a mobile site, wireless backhaul can shorten the time between a new location and usable network service. For an event or emergency, a link that can be moved and commissioned quickly has option value even if it does not replace the permanent route.

That value disappears if a nominally quick radio installation requires expensive reflectors, specialist alignment, repeated visits, new spectrum approvals or a difficult interface to existing equipment. The relevant comparison is total time-to-service and cost-to-service against the real alternative at that location. It must include the radio pair, mounting, power, alignment, integration, field support and capacity that survives a fault. Fibre remains the better choice where it is already available, reliable and inexpensive; the test is which path gives the operator the outcome it needs.

A separate 2026 study in npj Wireless Technology examines beam management for mobile sub-terahertz links. It describes severe sensitivity to alignment under movement. That study is not an evaluation of NTT’s equipment, and the NTT trial is not presented as a moving-terminal demonstration. The distinction matters for 6G claims: a fixed point-to-point backhaul candidate and a continuously tracked mobile access link are different operating problems.

Evidence that would change the decision

A buyer should ask NTT and an integrator to demonstrate a representative link for long enough to expose seasonal variation, alignment drift, obstruction, maintenance visits and recovery after an interruption. The test should record installed time, useful throughput under the buyer’s traffic, latency and loss distributions, uptime, failover duration, total power and energy per delivered bit. It should identify the permitted frequency plan and show how interference, equipment replacement and remote monitoring will work. None of those values is supplied in the cited NTT release.

The first deployment should be bounded: one route with a clear reason that radio is preferable, a fallback path where continuity matters, and an agreed exit if service targets are missed. A staged trial can tell an operator whether speed relieves a real bottleneck or simply moves cost into alignment, power and support. The decision can then remain local to the data-centre or network team that bears the operating consequences.

The result deserves attention because it has moved an OAM-MIMO demonstration from a one-metre bench-scale path to an outdoor 100-metre link without relying on a single wider band. The useful next proof is not a bigger number. It is whether operators choose the radio, install it on schedule, integrate it cleanly and keep the service usable. NTT’s own roadmap places those proofs ahead.

Sources