Nvidia ramps up its AI-RAN play

• Nvidia is on a mission to underpin major digital infrastructures and development processes with its processors and software
• It’s already at the centre of the AI datacentre universe
• Now it has launched AI Aerial, its hardware/software combo solution for AI-enabled radio access network (RAN) architectures, though the supporting tech is already part of its 6G pitch
• Its proposition is supported by a roster of big name partners, including Dell, HPE, SuperMicro, Wind River, Red Hat, Samsung, Fujitsu and close collaborator SoftBank
• Nvidia is also taking AI Aerial into an ecosystem partnership with T-Mobile US, Ericsson and Nokia
• The move will only add to Intel’s current woes

Nvidia has cemented its position as the one of the key technology suppliers and, just as importantly, influencers in the mobile network infrastructure sector with the launch of AI Aerial, a combination of software and hardware designed to revamp the operations and capabilities of radio access network (RAN) architectures.  

That Nvidia, already the fulcrum of next-generation datacentre and enterprise IT developments thanks to the AI-driven demand for its graphics processing units (GPUs) and associated software, has had its eye on the mobile infrastructure sector is no secret – it made it very clear last summer that it intended to insert itself into the RAN technology ecosystem and become the focal point of AI developments in mobile network architectures – see Nvidia smells blood in the RAN.

That initial foray into the mobile networks sector hinged on tests and trials with Japanese mobile operator SoftBank that focused on the development of a distributed accelerated computing cloud platform that could run multiple mobile networking and generative AI (GenAI) workloads on a shared IT platform – see Nvidia and SoftBank team on GenAI, 5G/6G platform.

Nvidia and SoftBank followed up their initial collaboration with the formation in February this year of the AI-RAN Alliance, an industry body focused on the AI-enabled revamp of mobile networks that also attracted Amazon Web Services (AWS), Arm, Ericsson, Microsoft, Nokia, Samsung Electronics, and T-Mobile US as founding members – see AI-RAN Alliance launches at #MWC24.

But suggesting a new way to build and run mobile networks is one thing – delivering the technical foundations to help make it happen is another.

Now Nvidia has taken the next step with AI Aerial, which the vendor describes as “a suite of accelerated computing software and hardware for designing, simulating, training and deploying AI radio access network technology (AI-RAN) for wireless networks in the AI era” that “offers high-performance and energy-efficient software-defined RAN, improved network experience and new revenue opportunities with edge AI applications to host internal and third-party generative AI applications.”

That ticks a lot of boxes for network operators, of course, though such capabilities (as has been noted before) do not come cheap, so while the proposition is enticing, the capex associated with such deployments would likely bring tears to the eyes of network operator CFOs despite the return on investment (RoI) models making for interesting reading.

So what exactly is Nvidia bringing to the table with AI Aerial? Well, its RAN toolkit is pretty much what it had already proposed earlier in the year for the mobile sector’s 6G R&D efforts, but it seems Nvidia is keen to embed itself as quickly as possible in the 5G era and it has amassed a broader and bigger-hitting set of partners.  

From its own portfolio it is offering: 

• CUDA-Accelerated RAN – this includes software libraries to enable partners to develop and deploy virtualised RAN workloads on Nvidia processors. CUDA (Compute Unified Device Architecture) is Nvidia’s containerised, cloud-native AI software stack.
• Nvidia Aerial AI Radio Frameworks – these include software libraries for the popular PyTorch and TensorFlow machine learning frameworks that can be used to “develop and train models for improving spectral efficiency and adding new capabilities to 5G and 6G radio signal processing.” This also includes the Nvidia Sionna Neural Radio Framework, a “simulator that provides development and training of neural network-based 5G and 6G radio algorithms”.
• Omniverse Digital Twin – a system-level network digital twin development platform that “enables physically accurate simulations of wireless systems – from a single base station to a comprehensive network with a large number of base stations covering an entire city. It incorporates… CUDA-Accelerated RAN and user-equipment simulators, along with realistic terrain and object properties of the physical world.”

In terms of AI Aerial partners, Nvidia says Softbank (naturally) and Fujitsu are key collaborators, while other vendors ready to integrate their wares with Nvidia’s tech in a RAN deployment include: Cloud platform developers Red Hat, Wind River, Canonical and Aarna Networks (note that VMware, now part of chip giant Broadcom’s empire, is not in the mix); virtualised routing vendor Arrcus (which recently announced Nvidia as an investor); and server vendors Dell Technologies, Hewlett Packard Enterprise (HPE) and SuperMicro.

In addition, edge cloud platform provider Vapor.io is ready to host AI Aerial workloads and systems integrator World Wide Technology is ready to help bring tests, trials and projects to life. 

And there’s more… test and measurement firms Ansys and Keysight Technologies are using the Omniverse Digital Twin in their simulation systems, while other technology and academic institution partners – namely Deepsig, ETH-Zurich, Northeastern University and Samsung – are collaborating on Nvidia’s ongoing 6G research programme and helping to develop the Aerial AI Radio Frameworks.

That’s quite the ecosystem, especially when placed alongside the might of the AI-RAN Alliance’s other members and the news that Nvidia has teamed up with ambitious network operator T-Mobile US, Ericsson and Nokia to invest in an AI-RAN Innovation Centre based at T-Mobile’s headquarters in Bellevue, Washington – see T-Mobile US, Nvidia take the AI-RAN initiative.

This all looks like a post-summer push for much greater influence in the telecom sector by Nvidia, especially at a time when more major mobile operators are starting to consider their next RAN infrastructure moves.  

It’s interesting also that Nvidia references software-defined RAN and virtualised RAN workloads in its announcement but doesn’t touch on Open RAN. However, the head of Nvidia’s telecom business, Ronnie Vasishta, noted last year that the AI-RAN, shared workload model was well suited for virtual and Open RAN deployments because it improved the economics and assets optimisation of such architectures. And in the announcement about the launch of the AI-RAN Innovation Centre, the partners in that particular initiative note that “AI-RAN is a game-changing technology because it will enhance the current Open RAN architecture with the addition of the accelerated computing… AI-RAN will make the promises of Open RAN more viable, while also going beyond.”   

This will add fuel to the virtual RAN and Open RAN fires in the coming months, it seems.

And it spells further bad news for Intel, which is currently on the back foot and looking for ways to scale back its investments and reset its operating model – see Intel carves out foundry unit in strategic overhaul.

Intel has, to date, been the dominant chip supplier and ecosystem partner in the virtual and Open RAN system sector, which is where the mobile infrastructure sector is heading (albeit not at breakneck speed). Intel has already lost the initial AI infrastructure battle with Nvidia and now, it seems, it has something to be very worried about in the mobile networking domain.

But, as mentioned, every plan, especially these days in the hard-pressed telecom sector, needs to make economic sense, and that’s where Nvidia might hit a brick wall with its AI-RAN propositions.   

For more insights into AI developments in the RAN, check out TelecomTV’s latest free-to-download DSP Leaders report, ​​AI’s Impact on the Radio Access Network.

- Ray Le Maistre, Editorial Director, TelecomTV