Nvidia Has a Radical New AI-RAN Plan – a 6G Radio Unit Chip

Why It Matters

By bringing programmable GPUs into the RU, Nvidia could reshape the economics of 5G/6G infrastructure, lowering barriers to multi‑vendor massive MIMO deployments and opening a larger developer ecosystem. This shift may accelerate 6G rollout and give Nvidia a foothold in a market traditionally dominated by custom silicon.

Key Takeaways

• Nvidia plans AI‑RAN radio unit chip for 6G massive MIMO.
• GPU‑based RU could replace ASICs, handling up to 1,024 antennas.
• Power concerns mitigated by low‑power embedded GPUs under 100 W.
• Nvidia backs Marvell with $2 bn, enabling GPU‑compatible RAN processors.
• RAN spend fell to $35 bn, making programmable chips more attractive.

Pulse Analysis

Nvidia’s latest AI‑RAN announcement marks a decisive step beyond its traditional GPU stronghold in PCs and data centers. The company unveiled a radio‑unit (RU) chip designed for 6G’s massive‑MIMO architecture, leveraging its Aerial CUDA‑based framework to run low‑PHY functions that have historically required bespoke ASICs. By extending the same programmable silicon that powers its data‑center GPUs into the telecom front‑haul, Nvidia hopes to capitalize on a developer base of six million CUDA engineers and to offer a unified hardware platform that can evolve with future wireless standards.

The technical driver behind the move is the explosion of antenna elements in modern RUs. While a typical 5G RU handles four transceivers, 6G concepts envision up to 1,024, demanding roughly thirty‑two times more compute for beamforming and AI‑assisted signal processing. Nvidia argues that a low‑power GPU, capable of operating below 100 W, can meet these demands more flexibly than fixed‑function ASICs. This contrasts with Intel’s Granite Rapids, which stops at the DU and leaves the RU to custom silicon, leaving a gap that Nvidia aims to fill.

From a business perspective, the timing aligns with a contraction in global RAN spend, which fell to $35 billion this year after peaking at $45 billion in 2022. Nvidia’s $2 billion stake in Marvell and a parallel $1 billion investment in Nokia signal a broader strategy to lock in telecom partners and to embed its GPUs into the supply chain. If operators adopt programmable RUs, they could reduce reliance on costly, single‑use ASICs and accelerate multi‑vendor massive‑MIMO rollouts, giving Nvidia a potentially lucrative foothold in the emerging 6G market.