NVIDIA invests $2B in Marvell

NVIDIA is making a bold play to dominate both the hardware and software foundations of AI data centers, spending $2 billion on Marvell Technology to lock its NVLink Fusion inside the heart of tomorrow’s racks. This move means even when customers choose non-NVIDIA accelerators, NVIDIA still collects a toll—its interconnects, switches, and software must be in the mix. The catch? Competitors like AMD, Intel, and Broadcom are pushing UALink, an open standard that could weaken NVIDIA’s grip if it actually ships at scale. But right now, NVLink Fusion is live and racking up wins, while UALink is still vaporware.

There’s another twist: NVIDIA’s not just selling chips. With Mission Control 3.0 and a possible acquisition of SchedMD (maker of the ubiquitous Slurm scheduler), it’s tightening its hold on how AI factories are run, from power management to software orchestration. This deepens lock-in and triggers alarm bells for customers who want to avoid dependence on a single vendor—especially as upcoming hardware and procurement cycles hinge on open, neutral platforms.

Benchmarks show NVIDIA still outpaces rivals in sheer performance, and flashy new models like Gemma 4 promise to push more AI to the edge. But with supply chain risks looming and questions swirling about who’s really driving demand, the real battle is about who owns the rails of AI infrastructure—and whether NVIDIA’s tollbooth model can outlast emerging open alternatives. Featuring insights from Bloomberg, Reuters, Barron's, and Yahoo Finance.

Powered by Apisod.com