US Carriers Split on AI-RAN: What AT&T, T-Mobile, and Verizon's Diverging GPU Strategies Mean for Open RAN

September 19, 2026
10 mins
Technology
ai ran at&t gpus nvidia open ran t mobile telecom verizon

US Carriers Split on AI-RAN: What AT&T, T-Mobile, and Verizon’s Diverging GPU Strategies Mean for Open RAN

If you’ve been following the telecom industry’s AI obsession, you’ve probably heard the phrase “AI-RAN” tossed around like it’s the next big thing. And it is—but what’s becoming clear this week is that the three biggest US wireless carriers have very different ideas about how to get there. A pair of reports published on September 19, 2026, from Communications Today and ETTelecom, along with a recent Nokia announcement, reveal a fascinating split in strategy that could shape the future of Open RAN and the broader mobile network.

Let’s break down what’s happening, why it matters, and what it means for the networks that keep our phones connected.

What Is AI-RAN, Anyway?

Before we dive into the carrier strategies, let’s get our bearings. RAN stands for Radio Access Network—it’s the part of a mobile network that connects your phone to the internet via cell towers and base stations. Traditionally, this equipment has been built with specialized hardware from companies like Ericsson, Nokia, and Huawei.

AI-RAN is the idea of injecting artificial intelligence directly into that radio network. Instead of just using AI in data centers or back-end systems, carriers would use it to optimize the radio signals themselves—things like improving spectral efficiency (how much data you can squeeze into a given slice of airwaves), predicting traffic patterns, or automatically fixing problems.

The big question the industry is wrestling with is: what kind of computer chips should run this AI? Some, like NVIDIA, argue that graphics processing units (GPUs)—the same chips that power AI models like ChatGPT—are the natural choice. Others, including many traditional telecom equipment makers and some carriers, think that CPUs (central processing units) or specialized hardware can do the job just fine, often more efficiently.

That’s the backdrop for this week’s news.

Three Carriers, Three Philosophies

Verizon: The Skeptic

Verizon’s Chief Technology Officer, Yago Tenorio, has made his position clear: he’s not convinced GPUs are necessary for the core work of the radio network. According to the ETTelecom report, Tenorio “voiced reservations about their necessity for core radio workloads.” In plain English, Verizon is skeptical that adding expensive, power-hungry GPUs to cell sites is the right move—at least for the fundamental signal processing that keeps calls and data flowing.

This doesn’t mean Verizon is anti-AI. It likely means the company is looking at the economics and deciding that GPUs might not be the best fit for every part of the network. After all, GPUs are great at parallel processing for AI training, but radio signal processing has its own unique demands. Verizon seems to be saying: let’s not jump on the GPU bandwagon just because it’s trendy.

T-Mobile: The Enthusiastic Tester

On the other end of the spectrum is T-Mobile. The carrier is “actively testing AI-RAN prototypes with Nokia and Nvidia for spectral efficiency gains,” according to the same ETTelecom report. T-Mobile has long positioned itself as a technology leader, and it’s putting real resources into figuring out how AI can improve its network.

The partnership with Nokia and Nvidia is particularly notable. Nokia has been making waves with its AI-RAN platform, which combines its anyRAN software with NVIDIA’s Aerial RAN Computer. This setup allows operators to run traditional mobile connectivity alongside additional computing workloads on the same infrastructure—essentially turning cell sites into mini data centers that can run AI applications.

T-Mobile’s approach is hands-on: test it, measure it, and see if it delivers real gains. The company has been working with these vendors for a while, and its involvement signals that it sees potential in the GPU-accelerated path.

AT&T: The Pragmatist

AT&T, meanwhile, is taking what might be the most nuanced approach. According to Communications Today, AT&T’s RAN technology leadership defines AI-RAN broadly—covering any use of AI in network operations, from troubleshooting to optimization—rather than committing to GPUs as the default path. The carrier is evaluating multiple GPU platforms against a “power, price and performance framework specific to RAN requirements,” with GPUs slated for deployment only where they fit network economics.

In other words, AT&T isn’t ideologically opposed to GPUs, but it’s not going to deploy them everywhere either. It will use them where they make sense—and use CPUs or other hardware where they don’t.

AT&T has already demonstrated some interesting capabilities. It has tested integrated sensing and communication on CPU architecture, and it has worked with Ericsson (following a $14 billion open RAN contract award in 2023 that shifted a large share of its network away from Nokia) to test Ericsson’s GPU-free AI-native scheduler on dedicated hardware and Intel Xeon 6 cloud RAN systems. The company says these tests showed “similar” performance to GPU-based approaches, suggesting that you don’t always need the most expensive chip to get good results.

The Nokia Factor: A Global Push

While the US carriers are diverging, Nokia is trying to build momentum globally. On September 16, 2026, Nokia announced it was expanding its AI-RAN trials to eight named operators across four regions: A1 Group, Chunghwa Telecom, du, e&, Mobily, stc, TPG Telecom, and Zain Saudi. These operators are advancing proofs of concept or live field trials using Nokia’s AI-RAN platform and NVIDIA’s Aerial RAN Computer.

Nokia also said its earlier work with T-Mobile, SoftBank, Indosat Ooredoo Hutchison, and NTT DOCOMO is continuing. The company reported that its AI-RAN platform has demonstrated improvements of more than 20% in spectral efficiency—a significant gain that could translate into better network performance without adding new spectrum.

Nokia’s platform combines its anyRAN software with NVIDIA’s hardware, allowing operators to run AI workloads alongside traditional RAN functions. The company expects further improvements through a software roadmap extending into 2027 and 2028.

This global expansion is important because it shows that AI-RAN isn’t just a US phenomenon. Operators in Europe, Asia-Pacific, and the Middle East are all exploring the technology. As Remy Pascal, Practice Leader for Mobile Infrastructure at Omdia, put it: “The breadth of operator engagement announced by Nokia shows that AI-RAN is evolving from a research topic into a strategic priority for a growing number of service providers.”

Why the Split Matters for Open RAN

Open RAN is the movement to make radio networks more open and interoperable, using standardized interfaces so that operators can mix and match equipment from different vendors. It’s a big deal because it promises more competition, lower costs, and faster innovation.

AI-RAN and Open RAN are closely related. Many of the AI-RAN platforms being developed are built on Open RAN principles—using disaggregated hardware and software, and running on cloud-native infrastructure. The choices that carriers make about AI-RAN will therefore have a direct impact on the Open RAN ecosystem.

If Verizon’s skepticism prevails, it could slow the adoption of GPU-based AI-RAN and encourage more CPU-based or hybrid approaches. That would be good news for companies like Ericsson and Intel, which are pushing GPU-free AI-native schedulers and cloud RAN systems. It would also validate the idea that Open RAN can deliver AI benefits without necessarily requiring the most expensive hardware.

If T-Mobile’s testing leads to widespread deployment of GPU-accelerated AI-RAN, it could accelerate the trend toward using NVIDIA’s Aerial platform and similar technologies. That would strengthen the NVIDIA-Nokia partnership and potentially create a new standard for AI-RAN implementations.

AT&T’s pragmatic middle ground might be the most realistic path for many operators. By evaluating multiple platforms and deploying GPUs only where they fit the economics, AT&T is essentially saying that there’s no one-size-fits-all solution. This approach could lead to a more diverse and resilient Open RAN ecosystem, with different vendors and chip architectures coexisting.

The Economics of AI-RAN

At the heart of this debate is a simple question: does AI-RAN make financial sense? GPUs are expensive, both in terms of upfront cost and power consumption. If they don’t deliver enough benefit to justify those costs, carriers will be reluctant to deploy them widely.

Nokia’s reported 20%+ spectral efficiency gains are promising, but they come with caveats. The company hasn’t disclosed detailed results for each operator, nor has it specified when trials might lead to commercial deployments. And spectral efficiency gains don’t automatically translate into revenue—they depend on how operators use the extra capacity.

AT&T’s framework of evaluating GPUs against power, price, and performance is a recognition of this reality. The company is essentially saying: show me the business case, and I’ll deploy GPUs where they make sense. That’s a pragmatic approach that other operators are likely to follow.

What to Watch Next

As we move into late 2026 and beyond, here are a few things to keep an eye on:

T-Mobile’s trial results: If T-Mobile’s tests with Nokia and NVIDIA show substantial gains, it could sway other carriers toward GPU-based AI-RAN. If the gains are marginal, it could reinforce Verizon’s skepticism.
AT&T’s deployment decisions: AT&T has said it will deploy GPUs where they fit network economics. Watch for announcements about specific use cases and locations. These will provide real-world data on where GPUs add value.
Nokia’s commercial progress: Nokia has been positioning its AI-RAN platform as a commercial product since July 2026. The company needs to convert trials into actual sales to prove that there’s a market. Its expanded partnership with Microsoft around AI agents in telecom networks, announced in mid-September, is another piece of this puzzle.
Ericsson’s response: Ericsson is Nokia’s biggest competitor in RAN. If AI-RAN becomes a major battleground, expect Ericsson to ramp up its own offerings. The company has already been working with AT&T on GPU-free AI-native schedulers, and it may look to differentiate itself by offering more efficient, CPU-based solutions.
The Open RAN ecosystem: How will Open RAN vendors adapt to the AI-RAN wave? Companies like Mavenir, Rakuten Symphony, and others are already working on AI-native Open RAN solutions. Their ability to integrate with both GPU and CPU-based platforms will be crucial.

The Bigger Picture

The divergence among US carriers is a sign of a maturing market. In the early days of any new technology, there’s often a rush to adopt the latest and greatest. But as the hype fades, companies start to make more rational decisions based on their specific needs and economics.

AI-RAN is still in its early stages, but the fact that carriers are taking different approaches is actually a healthy sign. It means they’re thinking critically about what works for them, rather than following a single vendor’s narrative. That’s good for competition and innovation.

For Open RAN, the lesson is that there’s no single path to AI-native networks. Some operators will embrace GPUs, others will stick with CPUs, and many will use a mix of both. The key is to build flexible, interoperable systems that can adapt to whatever hardware makes the most sense.

As Jensen Huang, NVIDIA’s CEO, said when Nokia launched its commercial AI-RAN platform in July 2026, RAN is “the next AI infrastructure.” He may be right—but the road to that future is going to look different for different operators. And that’s okay.