Why is Nvidia buying ARM?
Nvidia is about to buy ARM for $40 Billion, how will this affect rest of the industry, and why are they doing it?
Nvidea has given a press release of their intentions, so let’s run through the highlights and give some comments.
I have speculated on these prospect and what they mean in earlier articles but now we got more details from Nvidea itself about what they intend to do. To put this all into context we need to understand the role of data centers in the computing industry.
A data center looks roughly like what you see in the picture above. Data centers are big buildings with lots of computer hardware. They are specially made to be able to supply power, cooling and security to all this expensive hardware.
A data center can house any kind of hardware in principle. In a typical cloud data center it contains lots of separate computers which operate mostly independent of each other. But these computers could be interconnect in specialized ways to function as super computers. So when we talk about data centers is a rather broad term.
However they are very important to Nvidea, because this is where they have seen most of their growth over the last years. Customers increasingly need more computational power to do data science, machine learning and scientific computing.
It was early discovered that graphics cards where also good for a lot of generic number crunching tasks, not just for displaying pretty 3D graphics. Realizing the huge market this represented Nvidea started making cards such as the Nvidia Tesla, which is not really a graphics card anymore, even though it uses a GPU, or rather a GPGPU (General-Purpose Graphics Processing Unit). Actually all modern GPUs are GPGPUs.
Nvidea started making a bucketload of money filling up data centers with these kinds of cards. So all looked good in the merry kingdom of Nvidea, but then a new threat reared its head on the horizon.
The Threat of Systems on a Chip (SoC)
If you have been following Apple news or mobile phone and tablet development, you know that the big trend today is towards systems on a chip (SoC). This means putting CPU, GPU, memory and everything all one one microchip.
This type of integration has been ongoing for a while. The new Playstation 5 will also essentially be an SoC (not entirely since memory will be external). AMD which provides chips for both Playstation and XBOX calls these chips APU (accelerated processing unit) which is basically a CPU and GPU on one chip.
This kind of tight integration gives opportunities for higher performance and lower cost.
Nvidea has a big problem. Since AMD bought ATI they have the ability to make both CPU and GPUs and integrate those. Nvidea in contrast has no CPU of their own. As AMD and Intel move further down the APU path, Nvidea is seeing themselves getting pushed out of the PC market and the console market. They have managed to keep an edge by having the most powerful GPUs and selling them to demanding gamers and data centers. But the advantage in cost and performance in full integration is clear. Thus if this trend is allowed to keep developing AMD and intel would end up eating Nvidea’s lunch.
Thus Nvidea has really needed a CPU of their own to make their own SoC which they can use to build the next generation data centers or maybe even gaming consoles.
The Nvidea Counter Attack
ARM offers Nvidea a way out of this dilemma. ARM is increasingly having a presence in data centers. Amazon is offering the cheapest cloud services today with their own custom ARM designs (Graviton2). The worlds fastest super computer is current ARM based (Sep 2020). And of course Apple is switching their computers over to ARM.
From the press release we have this statement:
Uniting NVIDIA’s AI computing capabilities with the vast ecosystem of Arm’s CPU, we can advance computing from the cloud, smartphones, PCs, self-driving cars and robotics, to edge IoT, and expand AI computing to every corner of the globe.
What this is saying is that Nvidea is intending to merge their GPUs with ARM CPUs to create a powerful package they can use to supply data centers with complete solutions, and not risk that they get cut out of the equation by companies controlling a CPU design.
Thus in the future if you want an ARM CPU with a GPU on it, this will be an Nvidea GPU, thus advancing Nvideas dominance with e.g. CUDA. CUDA is an Nvidea specific API for programming their GPUs. It is really dominating in scientific computing.
It will also allow Nvidea to wrestle their way into smart phones and the other areas mention. ARM is the gateway drug. It is much harder to do this by controlling a CPU alone. A company cannot merely pick a GPU and then just bundle a CPU later. It works the other way around. You pick CPU first and this decides all the other options you have.
What is the Nvidea/ARM future?
It is clear from this press release that Nvidea has big plans to dominate data centers in the future. Until recently ARM has been all about small devices such as microcontrollers and mobile phones.
But now we can expect to see a massive push and big investment into making ARM a big player in high performance computing:
NVIDIA will build on Arm’s R&D presence in the U.K., establishing a new global center of excellence in AI research at Arm’s Cambridge campus. NVIDIA will invest in a state-of-the-art, Arm-powered AI supercomputer, training facilities for developers and a startup incubator, which will attract world-class research talent and create a platform for innovation and industry partnerships in fields such as healthcare, robotics and self-driving cars.
Machine Learning is a big new thing. We see that in self driving cars, image processing, digital assistance and many other areas. This requires massive processing power, and large data centers and more super computers will have to be built to serve this market.
Nvidea now has a chance to kick Intel off its throne. This kind of computing will rely on having both fast CPUs and GPUs. Intel has fast CPUs but their GPUs are no match for Nvidea. Nvidea will be in a race to beef up ARM while Intel will try to strengthen their GPU lineup.
Outcome for Britain and Europe
As a European I do find this development interesting. It Nvidea keep their word, it means greatly expanded R&D and super computing activity in Britain. This is part of the new economy and the future. Building up talent and capability in Europe will positive for the future.
At the same time I cannot help but being worried about what this means for the future. Large American multinationals buying up European tech companies have not always had a good outcome.
While this can be debated, it is my opinion that an ex-Microsoft executive taking over Nokia and the later Microsoft buyout was a complete disaster for Nokia. Sure it was a struggling company but the Microsoft takeover accelerated the decline. This is not the first time Microsoft does that. Microsoft basically killed another leading European tech company: Skype. Skype owned the market for voice over IP. Everybody used it and the stability was amazing compared to what we had been used to.
After Microsoft took over we saw a gradual tumbling down and who really use Skype anymore these days?
Right across where used to work in Oslo, we had Tandberg which had been a very successful Norwegian company which basically invented video conferencing. It was a company where nobody ever quit. It had an awesome company culture. People loved going to work there.
When American Cisco bought them out, things changed. Over time a lot of people quit a company that never had people quitting. Former Tandberg employees started almost a little cottage industry of startups in Norway.
You see some similar concerns in Israel, often called startup nation. They create many innovative tech companies. Except a lot of these get bought by large American multinational. Being a branch of a larger company has its limits. Growth may begin to happen elsewhere in the world. A lot of the profits may end up outside of the country.
So while I am exited about the potential for new exiting things happening in Cambridge, England, I think one should not be too naive about this. Nvidea can invest more money but there is also a danger that they kill the Golden Goose. Will e.g. Nvidea respect ARM corporate culture well enough, or will there be an exodus of ARM employees going elsewhere?
I like to think the best, as Nvidea really needs this to work, but mergers are seldom the magic bullet management imagines them to be.
China and Why the Deal May Fail
In the press release there is an important piece of information:
The proposed transaction is subject to customary closing conditions, including the receipt of regulatory approvals for the U.K., China, the European Union and the United States. Completion of the transaction is expected to take place in approximately 18 months.
If China is required to okay this deal, there is a high chance it will not go through. China is already in an adversarial relation with the US, especially since Trump took power. The US is exploiting its control of key microprocessor technology such as the x86 ISA, to choke China.
China desperately want to avoid depending on the US for strategically important technology such as microprocessors. China wants to dominate AI research and super computers. They need microprocessors for that. Not to mention all the electronic products they make which need microprocessors inside. For this reason China has preferred to push for ARM and RISC-V rather than x86.
I cannot imagine that China will look fondly at a US company, Nvidea getting control over critical technology like ARM CPU designs. Then again China may decide that battle is lost anyway and that ARM being based in Britain would always be at risk from US influence. China may simply double down on RISC-V.
To what degree China can stall this deal I don’t know. They may only be able to sabotage it and not entirely stop it. However don’t be surprised if this deal hits a bump in the road due to China.