Maximize Spectral Efficiency with AI-Native RAN and NVIDIA AI Aerial

Spectrum is one of the most valuable assets in wireless communications. Over the last 30 years, telecom operators in the US have spent more than $240B to acquire wireless spectrum. A goal of a radio access network (RAN) system is to extract the maximum spectral efficiency (bits/second/Hertz) possible, which translates into more capacity, stronger network resilience with fewer dropped packets, and better economics per site.

As operators look for ways to extract more value from existing spectrum, Massive MIMO  (Multiple-input, multiple-output) has become an important method for increasing capacity and improving how efficiently the network serves users. 

Massive MIMO promised a revolutionary leap in spectral efficiency. However, in field deployments, the industry is operating below what the technology can theoretically achieve, leaving immense capacity unused. The root causes are fundamentally system-level problems. The network struggles to accurately track user locations, signals overlap and cause interference, and the system fails to pair users efficiently for simultaneous data transmission. 

The industry has examined these challenges through a “compute-constrained” lens, treating compute as a scarce resource, forcing compromises to fit complex algorithms within the strict limitations of CPU power and performance budgets.

NVIDIA AI Aerial is changing that paradigm. With parallel computing, compute is no longer the bottleneck for operating within existing power budgets. By embracing an AI-native, highly parallelized architecture, we don’t have to ask, “How can I squeeze more into the same compute?” Instead, we ask, “How can we reinvent the algorithm across the full stack to maximize the spectrum?”

This “algorithms-first” approach enables the network to finally run the complex tracking and pairing models required to close the massive MIMO performance gap.

This blog describes the benefits of GPU acceleration in the RAN, and how AI-native RAN helps close the massive MIMO performance gap. It shows how NVIDIA AI Aerial enables a new class of Layer 1 and Layer 2 algorithms designed to unlock greater spectral efficiency in real-world deployments. 

Why GPU acceleration benefits spectral efficiency

Modern RAN pipelines consist of mathematically dense algorithmic tasks, with highly specific compute characteristics. A scheduler that evaluates a wider combinatorial space, a channel estimator that learns from richer observations, or a beamformer that adapts jointly across users can improve throughput and quality of service for user applications. Table 1 is a breakdown of the highest-impact RAN workloads, their compute characteristics, and why GPU acceleration is required to execute them effectively:

RAN workloadKey compute characteristicsWhy GPUs matter

Multi-User MIMO (MU-MIMO) User Equipment (UE) pairing	Combinatorial search for frequency and spatial resource allocation across large user pools	Handles combinatorial scale and real-time AI inference • Parallelizes large-scale UE pairing and candidate MU group evaluation • Enables real-time AI inference with larger models and batched processing across cells
Beamforming and precoding	Large matrix operations and per-user/layer optimization	Supports massive spatial multiplexing • Maps tensor-heavy linear algebra naturally to GPU architectures • Maintains multiplexing performance as antenna and layer counts continue to grow
Deep reinforcement learning link adaptation	AI inference over state history, ACK/NACK, CQI, and complex channel evolution	Enables scalable policy execution • Supports larger, higher-performing AI models and larger batch sizes while meeting strict slot deadlines
Channel estimation	Dense signal processing over SRS/DMRS with high antenna counts	Reduces pilot overhead • Supports advanced channel estimators that leverage more observations than conventional methods
Scheduling	Complex sorting, PRB allocation, and cross-cell fairness objectives	Overcomes CPU core limits • Excels when scheduling expands beyond individual cells to dense, cross-cell optimization
Neural receiver	Tensor-heavy equalization and detection using high-dimensional IQ data	Enables AI at the waveform • Supports waveform-adjacent AI models that are impractical to run on CPUs alone

Table 1.  Value of GPU by RAN workload

The following sections review two workloads—beamforming and link adaptation—and explore how GPUs unlock new spectral efficiency gains.

Beamforming

Beamforming quality determines how effectively a network converts channel knowledge into usable throughput. Classical methods are effective but are shaped by computational compromise, relying on simplified models to meet limited budgets. As systems move to higher antenna counts and denser MU-MIMO operation, these simplifications result in lost signal-to-interference-plus-noise ratio (SINR) and lower realized throughput.

ML-based beamforming weight generation solves this by using richer channel information for weights, though it raises the compute burden sharply. NVIDIA analysis shows that in a 64T64R MU-MIMO scenario with 16 users and 2 layers per user, where users are randomly assigned SNRs in the [0-20] dB range, AI beamforming requires significantly more FLOPs than traditional regularized Zero Forcing (rZF), but this increase in compute produces a spectral efficiency gain of up to 1.62x higher throughput at 32 layers compared to the traditional zero-forcing beamforming algorithm.

Even at a lower number of layers, ML-beamforming delivers 1.28x throughput improvement. 

Method	Complexity per cell	Spectral efficiency gains
rZF Beamforming	272M FLOPs		1.0x (baseline)
AI Beamforming	2.58B FLOPs		1.28x at 16 layers 1.62x at 32 layers

Table 2: Complexity compared to throughput in 64T64R MU-MIMO

GPU compute enables higher-quality beam weight generation at the scale and latency required for RAN deployment. This connects directly to field validation. SoftBank and NVIDIA recently reported stable outdoor 16-layer massive MU-MIMO operation on a GPU-based AI-RAN platform. The trial delivered roughly 3x the spectral efficiency of a conventional 4-layer baseline. GPU value is not abstract acceleration; it is the practical ability to sustain higher-order spatial multiplexing in a real system.

DRL link adaptation

Link adaptation is a complex MAC-layer control function. The scheduler must repeatedly pick a modulation and coding scheme (MCS) to maximize throughput while keeping block error rates (BLER) near targets for QOS and maximizing spectral efficiency under changing mobility and interference conditions.

Traditional link adaptation schemes, such as outer-loop link adaptation (OLLA) implemented in vector engines in CPUs, are lightweight, reacting to feedback through hand-crafted predefined logic. Deep reinforcement learning (DRL) link adaptation changes this by learning the MCS-selection policy directly from observed radio behavior. 

A DRL agent uses channel quality indicator (CQI) history, ACK/NACK feedback, and short-term channel evolution to adapt to site-specific conditions in real time. Early NVIDIA engineering results show a 1.3x throughput gain over baseline OLLA at the cell edge when combined with channel-orthogonality-based user pairing. Further improvements are expected when DRL-LA is combined with more advanced MU-MIMO pairing algorithms. 

The technical turning point is latency scaling. A small, distilled model can run on a CPU equipped with vector engines. The highest spectral efficiency gains require a larger model handling richer MU-MIMO conditions at large batch sizes, while maintaining stable BLER below target.

As shown in Figure 3, the GPU-based architecture stays below the ~30 μs typical reference budget across the tested user range, even for a 396K-parameter model. In contrast, a single-core CPU implementation exceeds the latency budget starting with the first scheduled user. CPU inference can support only a low-complexity model, while GPU inference enables higher-capacity models that deliver greater performance gains. Link adaptation becomes a true spectral efficiency lever only when policy quality and inference latency improve together.

Research validation

Recent academic studies also find the same result. ML-based channel estimation and equalization, and link adaptation outperform classical baselines under difficult radio conditions. Below are the summaries from three separate research papers:

Table 3. Ecosystem findings on AI-driven RAN algorithms

Unlocking spectral efficiency with NVIDIA AI Aerial

NVIDIA AI Aerial does more than accelerate RAN workloads; it enables new network capabilities. The following five capabilities—including improved spectral efficiency—highlight how the platform advances next-generation wireless networks.

1.  Algorithm-first efficiency

Instead of relying on simplified heuristics to fit narrow CPU budgets, AI Aerial enables operators to run mathematically dense, AI-native Layer 1 and Layer 2 models. Aerial shifts the design target from “what fits on the CPU?” to “what maximizes radio performance?” by evaluating massive combinatorial spaces in real time.

2. Model growth without redesign

As AI moves closer to the physical waveform, models for channel estimation, equalization, and adaptation will grow. Rigid ASICs freeze algorithms in silicon, but the programmable tensor compute headroom of a GPU enables AI to live at the heart of the RAN. AI Aerial absorbs the growth in software rather than forcing a new hardware partition every time the algorithm changes.

3.  Scale and coordination

Next-generation networks require dense, overlapping coverage, creating a massive computational burden for cross-cell interference tracking. Traditional CPU schedulers struggle with the latency scaling required for this level of coordination. AI Aerial handles this through massive memory bandwidth and shared data spaces between cuPHY and cuMAC, executing complex multi-cell math instantly to support higher-order MU-MIMO implementations 

4.  Integrated sensing and communication

Integrated Sensing and Communications (ISAC) essentially turns a radio network into a ubiquitous radar system. This introduces a workload that requires processing standard communications while simultaneously executing deep-learning classification models. AI Aerial provides a parallel architecture that blends high-throughput communications with sensing.

5. Monetize AI infrastructure

Telecom networks are typically provisioned for peak traffic, leaving infrastructure severely underutilized during off-peak hours. AI Aerial changes the return on assets by supporting the dynamic allocation of 5G/6G and AI workloads on the same GPU. Operators can reallocate spare GPU compute to host monetizable edge inference applications, transforming dedicated telecom equipment into revenue-generating infrastructure for the broader AI economy.

The architecture of the AI era

The shift to AI-native RAN is the path to closing the massive MIMO performance gap, giving the network the intelligence and parallel compute to turn theoretical spectral efficiency into real-world gains.  

NVIDIA AI Aerial was built for this transition. As a software-defined, accelerated computing platform for AI-native RAN, it makes advanced Layer 1 and Layer 2 intelligence practical at scale. The result is a more capable RAN that helps operators unlock more value from every Hertz and opens the door to new AI-driven revenue opportunities. 

NVIDIA is working with industry leaders like Nokia to deliver AI-RAN platforms integrated with Nokia’s anyRAN software, enabling operators to deploy AI-native 5G-Advanced networks ready for 6G evolution. With platforms like NVIDIA ARC-Pro enhancing performance, efficiency, and programmability, operators can accelerate their path to fully intelligent, software-defined radio networks. 

Learn more: 

Learn more about commercial Aerial RAN Computer (ARC)-Pro with NVIDIA RTX Pro 4500 Blackwell Server Edition GPU. Explore the NVIDIA AI Aerial platform to build, simulate, and deploy AI-native networks, from research to commercial deployment. See how Nokia is enabling AI-native 5G-Advanced and 6G networks.