Fujitsu's 1Finity Builds Full-Stack AI-Native vRAN on Nvidia GPUs

Fujitsu's networking arm 1Finity has launched a full-stack, AI-native virtual radio access network (vRAN) solution built on Nvidia GPUs, aiming to let mobile network operators sell AI and RAN services to enterprises from a single platform.

The stack, announced by the recently spun-out unit, combines Nvidia's AI Aerial platform with O-RAN Alliance compliant radio units from the Japanese vendor. It runs open vRAN software on top of accelerated computing infrastructure supplied as AI-centric servers from Quanta Cloud Technology (QCT) and Supermicro.

1Finity is one of several names in the mix. A no-code platform from Aible sits inside the stack, while Bridge by Armada provides multitenant orchestration, security and billing for wide-scale compute deployments.

What is the pitch to operators?

The company positions the offering as a route for MNOs to run a comprehensive portfolio of AI and RAN-based solutions, shorten go-to-market timelines and deliver varied services directly to enterprise customers.

That framing comes with a market reality check. The stack also extends Nvidia's AI Aerial footprint in Japan, where the platform has already won validation from SoftBank and NTT DoCoMo, the country's two largest operators.

The track record is concrete. DoCoMo's 2023 deployment used Fujitsu 5G-focused virtual compute unit (vCU) and virtual distributed unit (vDU) software running on Wind River's distributed cloud platform. Before last year's spin-out of 1Finity, Fujitsu also supported vRAN research and development for vendors including SoftBank and network test specialist Viavi Solutions.

In other words, the Fujitsu lineage in Japanese vRAN predates the new branding, and the Nvidia partnership formalises work that operators have already trialled in production settings.

How do executives frame it?

"By enabling operators to deliver complete AI solutions – from infrastructure to applications – 1Finity is redefining how telecom networks create and capture value in the AI era," said Patrik Eriksson, VP and head of the mobile system business unit at 1Finity, in a statement.

Nvidia framed the shift in broader infrastructural terms. "The telecommunications industry is undergoing a foundational shift as networks evolve into AI infrastructure," said Soma Velayutham, VP of AI and telecom at Nvidia.

"By building its full-stack enterprise solution on the Nvidia AI Aerial platform, 1Finity is empowering MNOs with high-performance, software-defined AI-native infrastructure that seamlessly processes AI and RAN workloads," Velayutham added. "This is essential for operators looking to maximize compute utilization, deliver advanced capabilities, and unlock new revenue streams in the AI era."

Commercial substance versus platform marketing

The engineering claim is straightforward: one software-defined infrastructure layer that processes both AI and RAN workloads, improving compute utilisation for operators that would otherwise run separate silicon for each.

The commercial claim is larger. Both vendors argue the combination opens new enterprise revenue streams, from infrastructure to applications, and shortens deployment timelines. No customer contracts for the new full-stack package were named at launch, so the revenue argument currently rests on the earlier Japanese validations rather than signed deals for the integrated offering.

The standards context matters here too. The radio units are O-RAN Alliance compliant, which keeps the offering compatible with the open interfaces operators and regulators in several markets have pushed for as a check on single-vendor lock-in.

What happens next?

With SoftBank and DoCoMo already familiar with Fujitsu's vCU and vDU software from prior deployments, Japan is the natural first market for the combined stack. The pace of adoption elsewhere will likely depend on whether operators in other regions follow Nvidia's AI Aerial validation path and commit to running AI workloads on the same infrastructure that carries their RAN traffic.