NVIDIA executive says Arm is gaining momentum in AI data centers, pointing to GPU and TPU demand

NVIDIA Chief Architect Rene Haas said in an interview that the growing use of Arm-based processors in AI-focused data centers is helping validate the industry’s shift away from traditional x86 platforms. Haas argued that as more AI workloads move into large-scale training and inference systems, demand for specialized compute accelerators will rise across architectures, and that “every AI workload is going to run through Arm.”

Haas’s comments highlighted a view that the race for AI capacity is no longer confined to a single server design. Instead, he characterized the datacenter buildout as accelerating adoption of Arm, while still pulling through high-performance accelerators from companies like NVIDIA. The implication is that, even as server CPU architecture changes, the compute bottleneck in modern AI systems continues to drive spending on GPUs and other accelerator hardware.

In the same framing, Haas cited competitive indicates from Google’s TPU (Tensor Processing Unit) ecosystem. TPUs are custom AI accelerators designed by Google for machine learning workloads. By pointing to TPUs alongside NVIDIA’s GPUs, Haas suggested that the underlying economic and performance needs of AI workloads are pushing major cloud players toward purpose-built accelerator options, even when server architectures evolve.

The interview also tied the architectural transition to NVIDIA’s revenue growth, saying that the shift away from x86 and toward Arm in AI infrastructure is “driving” the company’s results. NVIDIA does not typically comment on revenue drivers in granular terms outside of formal filings and earnings materials, and the CNBC interview as presented here did not provide additional detail such as specific customer contracts, shipment volumes, or percentage impacts by architecture.

For readers, the key technical distinction is that Arm is an instruction set architecture used by many modern processors, while NVIDIA GPUs are accelerator chips designed to accelerate parallel computations common in deep learning. In practical terms, data center operators can mix Arm-based CPU platforms for general compute and networking with GPU accelerators for the heavy lifting of model training and inference.

Within the broader technology sector, Haas’s remarks land in a context where cloud and enterprise buyers increasingly standardize on platforms optimized for AI throughput and energy efficiency. Arm-based servers have attracted attention from hyperscalers and infrastructure buyers seeking power and performance advantages, particularly as AI fleets scale and energy costs become a larger part of operating expenses.

Still, important details are missing from what was disclosed in the interview coverage. The post-and-article material provided here does not include the exact quotes beyond the high-level claims, nor does it specify which NVIDIA product families or data center systems are seeing the strongest Arm-related tailwinds. It also does not detail whether NVIDIA expects Arm adoption to affect gross margins, supply availability, or customer mix in a measurable way.

What to watch next is whether NVIDIA’s upcoming investor communications, such as earnings calls, investor presentations, or formal remarks, quantify any linkage between Arm server adoption and GPU revenue. For market participants, the sharper question is not whether Arm gains share, but how much of the incremental AI datacenter spending flows to NVIDIA’s accelerator platforms versus alternative accelerator strategies.