ARM Wrestling

CNBC reported late last week that Qualcomm has objected to Nvidia’s proposed acquisition of Semis IP licensor Arm. According to this somewhat thinly sourced report, the US FTC regulators have moved their review of the acquisition to a second, more intensive phase, and sought out comment from Arm licensees including Qualcomm. We have no way of confirming this report, but honestly, we are not surprised, we think the deal is likely to have a very hard time closing.

First, some background. Arm creates software that performs the core mathematics at the heart of processors. Semis companies license Arm’s Intellectual Property (IP) and burn the Arm libraries into their chips. Arm cores do the heavy duty math around basic computation that different companies can apply to their particular needs. So Qualcomm uses Arm to power its modems and applications processors. Nvidia uses Arm for its GPUs. Arm provides IP that is vital to all chip companies, saving those licensees from having to re-invent the wheel. Or to modify the analogy. Arm makes car engines, then the chip companies wrap the engine with their own frames and designs. It is a highly efficient model that has powered the rise of modern electronics.

Arm itself has a long history. Based in the UK, the company was involved in computers from very early days. It really came into its own with the rise of mobile phones, especially smart phones. To greatly oversimplify, Arm beat out alternatives (like Intel’s x86 architecture) by being more power efficient, a crucial trait for portable devices like phones. Arm did so well with phones that it extended its reach into pretty much every electronic device out there including PCs, servers, home networking, IoT and even power adapters. Therein lay the problem.

Arm has a natural monopoly. There are tremendous network effects built into their model. Once a licensee designs Arm into their chips, all the software that runs on that phone has to be written for Arm. For higher level apps, this is a smaller problem, but for the low level software and firmware that consumers may never see, this is major lock-in for Arm. Porting software to some new processor is incredibly costly. It is possible, there are alternatives, like the open source RISC V, but the Arm installed base is so immense, it would take a decade to move away.

Arm licensees are highly dependent on Arm, and definitely have strong opinions about who owns Arm. That being said, like any monopolist, Arm over the years became somewhat sclerotic. Working with them is not easy, there were frustrations with their negotiating tactics, roadmaps and progress (or lack thereof) on those roadmaps. And as much as Arm talks about moving into new markets, most licensees have legitimate complaints about actual resources Arm dedicated to supporting those new markets. So it was something of a relief when Arm was acquired by Japanese conglomerate Softbank in 2016. The hope was that Softbank would inject resources and more importantly new energy into Arm.

Those hopes did not entirely materialize, but they also did not run their full course. Softbank ran into its own set of problems (e.g. WeWork among others) and by 2018 were looking to cash in on Arm. That culminated in 2020 with the announcement that Nvidia intends to acquire Arm.

The problem is that Nvidia is not a neutral party like Softbank. Nvidia competes with many of Arm’s licensees in a number of areas. And while we could argue how much this matters in GPUs (AMD) or networking products (Nvidia acquired Mellanox last year), the biggest conflict is likely to be in the emerging market for automotive semis. Every major chip company has ambitions for this market and are going to be very uncomfortable licensing Arm IP from Nvidia who also has eyes on this market.

Everyone we have spoken with has expressed some version of “Nvidia is going to have a hard time with this deal”. Arm has so many licensees, that especially among the largest ones there is an enormous area of conflicts with Nvidia. Factor in US regulators who seem to be on a track for increased deal scrutiny in all industries, but especially tech. Add the European regulators who are explicitly concerned about US technology dominance. Then finally, as with all things semis, insert China’s regulators into the mix. All of this makes for a recipe of great difficulty for Nvidia to navigate. Nvidia likely has credible arguments to counter and assuage concerns, but it certainly looks like we are in for a protracted wrestling match.

Photo by Larry Costales on Unsplash

One response to “ARM Wrestling

  1. Pingback: Heterogeneous Compute | Digits to Dollars·

Leave a Reply