Nvidia’s purchase of ARM in 2020 kicked up a lot of controversy over whether ARM could be owned by one of its most prominent licensees. Now, the UK government has announced an investigation into the deal and the likely outcomes if Nvidia is allowed to buy ARM.
The UK’s Competition and Markets Authority has stated it will examine how Nvidia’s deal with ARM treats other current licensees of the company’s designs, and investigate whether the deal gives ARM “an incentive to withdraw, raise prices or reduce the quality” of the chips it builds. The organization also said it would examine the deal in cooperation with other authorities around the world, to ensure it does not harm consumers or raise prices.
Nvidia has previously promised that it would retain ARM’s licensing model and its pledge of neutrality regarding how it treats its customers. How serious it is about this promise has been a topic of some debate. Nvidia, historically, has put more effort into creating its own walled garden of technology. PhysX, CUDA, G-Sync, and other technologies introduced over the years, were created by Nvidia, for Nvidia. This has, at times, been controversial, but it’s also been consistent.
There’s fear that allowing an ARM customer to own the company will create an incentive for said firm to abuse its position, and there’s additional concern about such a prominent company in the UK being effectively ‘moved’ to the US. When SoftBank bought ARM, it signed several agreements to keep ARM in the UK. Nvidia has pledged to do the same, and to open a new AI computing center, and that it won’t complete the deal until 2022.
As to whether Nvidia can be trusted to keep its word regarding how it treats ARM — that’s a complex issue. While it’s true that Nvidia has never fallen over itself to give away its own developed technology, there’s a huge difference between owning the CPU designs that allow for the existence of the entire ARM ecosystem, with all its various licensees, and restricting the use of internally developed technology. If Nvidia started trying to jack up the prices of ARM cores or change the license model to abuse its own customers, it would be inviting the wrath of a huge chunk of the tech industry. Long-term, attempting to strong-arm the market could drive more companies to adopt the open-source RISC-V ISA.
The reason I don’t think Nvidia is likely to wreck ARM’s licensing model is that the above licensing model is what’s responsible for ARM’s success in the first place. There is nothing stopping Nvidia from building its own custom ARM core with crazy extensions and capabilities now, and while owning ARM itself undoubtedly dovetails with some of the GPU manufacturer’s long-term plans, it gains nothing by driving its own customers away. Nvidia can already build a custom ARM CPU if it wants one. It isn’t buying ARM to do that. Attacking ARM’s business model by jacking up licensing fees or changing agreements around core IP would only make ARM less popular, damaging the reason Nvidia bought the brand in the first place.
This particular deal was always going to face close scrutiny, so it’s not surprising to see the UK giving it a careful once-over. We’ll find out over the course of this year what regulators in various nations think of the underlying terms of the deal.
Could Nvidia’s DLSS Help the Nintendo Switch Hit 4K Resolutions?
Could Nintendo use DLSS to help the Switch hit 4K resolutions?
Nvidia’s RTX 3000 Prices Have Gone From Bad to Brutal
Nvidia GPU prices got much worse in February. Here's hoping March brings improvement.
Nvidia’s Jetson AGX Orin Packs an AI Punch in a Small Package
While physics to some extent predicted by "Moore's Law" has slowed down the rate of progress in some aspects of computing, GPUs continue to improve at an impressive rate. Nvidia's new Jetson AGX Orin is the same size as the older Xavier, but packs 2-8x the AI horsepower, depending on the application. We got our hands on one and put it through its paces.
Report: Nvidia’s Next-Gen GPU Could Pack 18,432 CUDA Cores, 64TFLOPS
Nvidia's upcoming Ampere, which may be named after computing pioneer Ada Lovelace, could pack as many as 18,432 CUDA cores, a substantial increase over current top-end cards like the RTX 3090.