Nvidia Plans GB10‑Style SoCs for Windows PCs, Sources Say

Nvidia’s GB10‑style SoC, unveiled last October in a joint effort with MediaTek, already powers a niche class of AI‑focused Linux workstations from Dell, Asus, MSI, Gigabyte and others, according to a Wall Street Journal report that cites insiders familiar with the project. The chip combines a MediaTek‑designed “big‑little” CPU tile—ten Arm X925 cores paired with ten A725 cores for a total of 20 execution units—with an Nvidia GPU tile whose specifications mirror the company’s RTX 5070 desktop GPU. In internal benchmarks the CPU portion delivered performance within 10‑15 percent of AMD’s top‑spec Strix Halo mobile processors across a range of workloads, while the GPU tile is rated for up to a petaFLOP of AI throughput at FP4 with sparsity, although real‑world applications fall far short of that theoretical ceiling (Wall Street Journal).

Power and thermal envelopes remain a stumbling block for mainstream adoption. The GB10 draws roughly 140 W under load—about three times the typical draw of contemporary thin‑and‑lite laptop SoCs and more in line with a gaming notebook’s power budget. That figure reflects the chip’s 128 GB of LPDDR5X memory and a 200 Gbps ConnectX‑7 Ethernet controller, features that are overkill for most consumer scenarios but essential for the high‑throughput AI workloads the platform targets (Theregister). Consequently, early GB10‑based machines have been priced in the $3,000‑$4,000 range and shipped exclusively with a lightly customized Ubuntu 24.04 Linux distribution, rather than Windows, limiting their appeal to a broader market (Theregister).

Despite those constraints, Dell and Lenovo are already planning Windows‑compatible notebooks that will incorporate the GB10 architecture later this year, the Wall Street Journal reports. Those OEMs appear to be betting on the growing demand for on‑device AI acceleration, especially in creative‑content, data‑science and enterprise‑edge use cases where latency and data‑privacy concerns make cloud‑only solutions less attractive. By integrating Nvidia’s GPU capabilities directly onto the same die as the Arm CPU, the SoC could deliver a tighter memory bandwidth path and lower inter‑chip latency than traditional discrete‑GPU configurations, a claim that aligns with Nvidia’s broader strategy of “superchip” integration outlined in its recent CES keynote (Wired).

The move also signals Nvidia’s intent to re‑enter the integrated‑graphics market that it once dominated in the mid‑2000s, when lightweight GPUs were soldered onto notebook motherboards. Since then, Intel’s integrated Iris Xe and AMD’s Radeon Graphics have become the default for most consumer laptops, leaving Nvidia largely absent from that segment. The GB10 effort, however, is the first time the company has attempted to bring its GPU expertise to an Arm‑based SoC that could run Windows, potentially reshaping the “Wintel” landscape that has been Intel‑centric for decades (Theregister). If successful, Nvidia could leverage its AI‑centric hardware advantage to offer a differentiated value proposition—high‑performance AI inference and training on a single chip—against Intel’s upcoming Meteor Lake and Sapphire Rapids roadmap.

Analysts note that the GB10’s power draw and premium pricing place it squarely in the high‑end notebook tier, where competition from Nvidia’s own RTX 40‑series mobile GPUs and AMD’s Ryzen 7000‑series APUs is fierce. Yet the integration of a dedicated AI engine, combined with the chip’s 200 Gbps Ethernet link, may appeal to enterprise customers seeking on‑premise AI acceleration without the cost and complexity of separate GPU cards. The upcoming Windows‑based prototypes will be the first real test of whether Nvidia can translate its data‑center dominance into a viable consumer and enterprise laptop offering, or whether the GB10 will remain a specialized tool for a limited set of power users (Wall Street Journal).