On Sunday Nvidia signed an agreement to purchase Arm from Softbank. There is no shortage of hot takes out there, but this is a very interesting development. There are, no doubt, significant obstacles to overcome, but Softbank wanted (needed?) to sell Arm, and there are only a handful of companies who could actually purchase the company, and Nvidia was one of them.
Strategically this is very interesting. Jensen has his target set entirely on x86 and displacing it and Intel specifically, from the data center as well as limit their opportunity at the edge and with AI. Nvidia is arguably the dominant force in AI and Arm the dominant force at the edge. And honestly, for all the talk that has been Arm in the data center going up against x86, Nvidia guiding it may be the only chance the Arm architecture has to actually succeed at scale, on all fronts, in the datacenter. This point from Jensen in his QA with analysts and reporters confirmed this strategy:
So I’m super excited to focus a lot of energy around turning Arm into a world-class data center CPU. That alone would excite all the customers and all the licensees of cloud data centers and enterprise data centers in high-performance computing centers who are clamoring for an online CPU.
Nvidia saying they would bring Nvidia tech to the Arm ecosystem was an interesting angle. To quote Jensen responding to a comment from Stephen Ellis from Reuters on what we can expect in terms of Nvidia technology being available and brought to the Arm network:
The first obvious thing for us to make available through our Arm’s vast network is our GPU and our accelerated computing architecture. Our AI computing is world-class, and the processor, the algorithms, the compiler, the applications for the world’s industries could be incredibly valuable. Those could be two obvious places to start. We soon have the ability to reach these thousands of developers who are creating billions of things which will soon be shipping trillions of chips.
I’ll start with Jensen’s wording that select Nvidia technology would be made available to the Arm network. The main question I had in the weeks prior, as we learned of Nvidia’s interest to buy Arm, was whether end customers were going to have to use any Nvidia technology. Jensen is saying its made available doesn’t make it sound like it, which means any more customized Arm solution will have options, which is in line with how Arm works. That being said, a good portion of Arm’s 22 billion chips shipped every year are not customized Arm solutions but more general off-the-shelf solutions. This is quite interesting when you think about the strength of Nvidia’s graphics technology alone and the potential for it to impact the Arm Mali GPU, things get interesting. Would Qualcomm want to use that if some of that solution is better than their own? What if Mediatek goes all-in with Arm, as they do, and thanks to Nvidia, can make a smartphone with better all-around computing solutions than Samsung or Qualcomm? Both Qualcomm and Samsung use some Arm IP and customize others. The idea that Nvidia could dramatically change the competitive landscape around off-the-shelf silicon is a fascinating option when just looking at mobile computing alone.
Obviously, this has a broader impact when we look at robotics, automation, and self-driving cars, and a host of other categories where Arm IP may be dramatically more competitive to its competing brands of Intel and Qualcomm, for example. But as we all know, a little competition is a good thing, and Nvidia fueling even more innovation in the broad IP from Arm would certainly do just that.
Going back to the strategy, it is clear this move is Nvidia wanting to control an architecture they way Intel controls X86. Licensing business models are hardly lucrative. Certainly not as lucrative as Nvidia’s existing business models, and even with Arm shipping ~22 billion chips a year, their annual revenue is still $1.5-2 billion compared to Nvidia’s $10 billion annual business in GPUs alone. And Nvidia’s annual GPU shipments are measured in millions compared to Arm’s billions annually.
As I said, if anyone can dramatically change the course of the future of Arm, it is Nvidia, but massive challenges remain for this deal. One including regulatory scrutiny.
During the press QA, two things stood out to me that I think are the key factors in getting regulatory approval. The first was Nvidia and Arm’s confidence this deal can happen. Jensen used his experience buying Mellanox as being key in understanding how to navigate the regulatory waters as it relates to a deal like this and the regulatory bodies’ desire to enable competition.
This indeed will help. Nvidia is also being very specific about how they talk about Arm, specifically pointing out how their technologies complement and do not compete in any way in any market, which is technically true.
The biggest issue to me is China. Arm in China is a bit of a hot mess, which is being managed by a joint venture in which Arm, and if the deal goes through Nvidia, will hold a minority share. But Arm IP is critical in China and a big bet of the Chinese government in order to create true silicon independence. China has a lot riding on the openness of Arm, and if that is ever to be perceived in jeopardy, they could create real problems for anyone bringing Arm-based solutions into their market.
While China likely can’t veto the deal, the deal itself still has to appease the needs of China, given their dependence on Arm at this point and the implications for Huawei specifically in some cases.
The biggest question I have around this is one more about politics. I believe some concessions will have to be made to appease China, and I am curious if those concessions will not fly if Trump is reelected. I think it is very clear at this point the spat between China and the US is like a Sumo wrestling match with both countries trying to be big and strong and push each other around. Neither country wants to appear weak or back down, as evidenced by this ordeal with TikTok.
Nvidia being a US-based company, may cause the Trump administration, if reelected, to be more stern on this deal if the necessary concessions to China are made. Again, the hot mess of the China Joint Venture of Arm, where some deals have gone rogue to appease China has to be straightened out, and both Jensen and Arm CEO Simon Segars noted getting the Chinese Arm JV straightened out was a top priority.
There will also be pushback from customers. I’ve already heard of several large customers who are not happy about this deal and even specifically telling Arm not to go through with it.
The Elephant in the Room
This may be one of the trickiest deals to get approved in our industry’s history, which is why I lean more on the side of the skeptic. Should this deal not go through, it leaves us with a still perplexing question and the elephant in the room? If Nvidia can’t buy Arm, then what is Arm’s future?
Softbank wants to unload this asset. This means Arm’s future is very much up in the air should Nvidia not be allowed to buy the company. As I mentioned, only a handful of companies even have the money to consider buying Arm. But Arm is also best if independent. Perhaps a joint venture is an option where the main customers, Nvidia, Samsung, Apple, Qualcomm, etc., make sense to own Arm. As difficult as this would be to manage something like that could be the second-best option to Arm’s true independence should the Nvidia deal fall through.
I have no hope this deal gets approved or denied anytime soon. This is going to take a long while, but one thing I’m quite confident in is any hope for this deal to pass requires Trump not to be reelected. Nvidia better hope and pray Biden wins in 2020.