The most expensive thing Cohere Technologies built was not a radio or a base station. It was an argument: that wireless signals make more sense when you stop looking at them in the familiar grid of time and frequency and instead map them by delay and Doppler. Delay tells you how long an echo took to arrive. Doppler tells you how fast the world around it is moving. Radar engineers speak this language naturally. In 2011, mathematician Ronny Hadani and serial telecom inventor Shlomo Rakib founded Cohere to make cellular networks speak it too.
Their invention was Orthogonal Time Frequency Space, or OTFS, a modulation scheme designed to stay predictable while phones, vehicles, satellites and reflected signals move. The physics was attractive. The buying proposition was brutal. Mobile operators were already spending fortunes rolling out 4G and then 5G, both built around OFDM waveforms. Asking them to adopt OTFS meant asking an industry famous for long standards cycles to take another architectural leap before it had landed the last one.
The first product was ahead of the market. The useful response was not to abandon the science, but to change where it entered the network.
The first thing that failedA 6G answer in a world still installing 5G
Cohere's own company overview is unusually plain about the mistake. Its proprietary OTFS system demonstrated strong field performance, but it was “ahead of its time” and impractical for operators to adopt before implementing 5G. By 2018 the company changed course. Instead of replacing the waveform, it extracted the underlying delay-Doppler channel detection, estimation and prediction work and applied it to networks already in the field.
The mind-change was about timing, not physics.
Cohere stopped requiring a new air interface. Its software would observe the current channel, predict which users could share the same radio resources, and coordinate beams through interfaces an operator or equipment vendor could actually deploy.
That became the Universal Spectrum Multiplier, or USM. The name is ambitious, but the job is concrete. When several people use one cell site, their signals can interfere. Massive MIMO radios have many antennas and can direct beams, yet the network needs a timely picture of the channel to decide which users can transmit together. Cohere says its geometric model reveals dominant reflectors, slows the problem of channel information growing stale, and lets the scheduler pair users whose beams can remain isolated.
What it actually sellsMore traffic through the spectrum you already paid for
USM is enterprise RAN software, not a consumer app and not a new handset feature. It can be integrated into a base station, run beside one on a server, live in central or distributed-unit software, or operate as an application in an Open RAN controller. Cohere's recurring promise is that phones, radios and antennas do not have to change. The buyers are licensed mobile operators, RAN equipment makers and cloud-network partners. Public pricing is not part of the pitch; contracts depend on network scale, integration and deployment mode.
These figures describe different tests and contracts, not one universal performance guarantee. Real network gains depend on traffic, radio conditions, antennas and integration.
The alternative shopping list is painfully familiar to every carrier chief financial officer: buy more spectrum at auction, add cell sites, install new radios, or wait for an incumbent vendor's next software release. USM tries to create another line item - software that makes the existing asset more productive. Cohere raised $46 million in a 2022 Series D to commercialize that idea and expand operator trials. In that sense, $46 million is the disclosed cost of turning the pivot into a market-ready program. The less visible cost was a decade spent proving the underlying model.
Adopt a new OTFS wireless system
Keep delay-Doppler channel intelligence
Add software to current 4G and 5G RAN
Evidence leaves the labVodafone gave the software a difficult Spanish city
Cohere's cleanest field result came from Vodafone in Ciudad Real, Spain. The operator tested USM in a dense urban area using 10 MHz of spectrum in the 2.1 GHz band. Buildings, reflections and mostly non-line-of-sight conditions made it a respectable place to stress channel prediction. When activated with multiple handsets, the software produced up to a 50 percent capacity improvement. Vodafone's head of Open RAN, Paco Martin, emphasized that this was the first urban network test and that the result avoided a costly equipment upgrade.
Bell Canada supplied a different test. In December 2024 and January 2025, Cohere placed USM on a server beside a third-party base station near Bell's labs in Mississauga. It coordinated single-user and multi-user MIMO scheduling in Bell's 850 MHz 5G standalone network without replacing radios or antennas. The headline was not a record multiplier. It was brownfield compatibility - the ability to enter an existing vendor network, the place where promising telecom software often gets stuck.
The partnership structure is part of the product. Vodafone and Bell are both investors. Intel, VMware, Capgemini Engineering and Telecom Infra Project participated in the earlier multi-vendor Open RAN demonstration. Mavenir announced a Multi-G integration. Every relationship attacks the same problem from a different side: Cohere needs channel data and a path into scheduling; operators need credible integration; equipment and platform vendors control many of the doors.
The original bet returnsTwo products running on two telecom clocks
Cohere never buried OTFS. In October 2025 it launched Pulsone, a separate business built around an evolved Zak-OTFS waveform. USM is meant for the installed base. Pulsone points toward integrated sensing and communications, non-terrestrial networks and 6G. The configurable design is intended to support current OFDM operation first and evolve by changing parameters as networks and devices become ready. It is the original ambition with a gentler migration path.
Integrated sensing is where the story gets strange in a useful way. A radio network already emits energy and receives reflections. If it can measure delay as range and Doppler as velocity, the same infrastructure can communicate with phones and help locate moving objects. In 2025, Cohere worked with Duke University and Virginia Tech on a real-time neural receiver running on NVIDIA Jetson hardware. In March 2026, Pulsone won a GSMA GLOMO award for mobile technology. In July, Cohere received a $28 million U.S. government contract for a multi-waveform RAN prototype designed to detect, classify and track aerial threats while resembling ordinary cellular traffic.
The company that once could not persuade carriers to replace a waveform now plans to let several waveforms share one software-defined network.
A third product layer, ECHO, arrived in May 2026. Enhanced Channel Insight with Holographic Observability turns USM's channel geometry into streams of data for AI-RAN, localization, network planning and proposed digital twins. Cohere says a mobile network can generate hundreds of millions of referenceable datasets a day. The sensible way to read that claim is as a new option, not yet a settled business: capacity optimization produces valuable observations, and those observations may support future applications beyond capacity.
The part worth copyingPreserve the advantage. Redesign the installation.
Deep-tech founders can copy Cohere's separation of invention from adoption. The company did not dumb down the mathematics. It changed the interface between the mathematics and the customer. The first version required the industry to cross a standards chasm. The second used existing handsets and radios, offered several deployment locations, and collected proof in progressively messier environments: simulation, lab, urban field and brownfield network.
There is also a useful discipline in the two-clock portfolio. USM asks what value the research can create inside today's constraints. Pulsone asks what becomes possible when standards, hardware and buyers catch up. One line can earn credibility and integration experience while the other keeps the longer technical option alive. Vodafone's reported $15 million convertible investment in Cohere, disclosed in early 2026 after years of trials, suggests that this patience has strategic value even before wide commercial rollout.
Where the multiplier stops multiplying
USM needs enough simultaneous traffic to pair users, multi-antenna conditions that support separated beams, timely channel information, and access to scheduling or open control interfaces. Sparse rural demand, hostile propagation, limited antenna capability or a closed vendor stack can shrink the benefit. If an operator or OEM will not expose the required hooks, good mathematics stays outside the base station.
That is why Cohere sits in an awkward but interesting part of the market. It competes not only with specialist RAN applications, but with Ericsson, Nokia, Samsung, Huawei and Mavenir software; with new radio equipment; with extra cell sites; and with spectrum itself. Incumbents own installed relationships and can fold optimization into broader contracts. Cohere's defense is waveform independence, a focused patent portfolio, deployment flexibility and the evidence that its channel model can work across FDD and TDD, old and new architectures.
Fourteen years after founding, Cohere is no overnight success story. It is a case study in how long telecom takes to digest a new idea. The first bet failed at the point of adoption. The company changed its mind about the route, not the destination. Now its mathematics has three jobs: squeeze more traffic through today's spectrum, turn radio channels into data for AI, and prepare networks to sense as well as communicate. That is a narrower claim than reinventing wireless all at once. It may also be the reason the invention finally has somewhere to go.