Arm & NVIDIA – The unthinkable.

Another unthinkable event to add to 2020’s long list.

• For nearly 20 years I have strongly believed that Arm could not be acquired by another semiconductor company as Arm’s acceptance as an industry-standard has hinged on its independence which will now be at greater risk than it was as a public company or part of SoftBank.
• The open-source alternative (RISC-V) to Arm could now see a big boost in interest from NVIDIA’s rivals and from China as Arm may be seen as US technology even more than it already was.
• NVIDIA and Arm proved me spectacularly wrong (see here) with the announcement of a deal to acquire Arm from SoftBank for $40bn.
• This is not a straight-forward transaction and has all the hallmarks of SoftBank feeling the need to sell Arm while at the same time being unable to take another write down.
• The $40bn is made up of: 1) $12bn in cash, 2) $21.5bn in NVIDIA shares, 3) $1.5bn being issued to Arm employees and 4) a $5bn cash payment contingent to Arm’s performance in the coming years.
• If one ignores the last two, the acquisition price is $33.5bn which given that Arm is in the middle of a huge investment phase, represents a good deal for SoftBank.
• The whole deal is being presented as the “creation of the world’s premier computing company for the age of AI” but how this works or what it really means is very unclear.
• NVIDIA and Arm are not really AI companies.
• NVIDIA is a picks and shovels company where its ability to produce the best massively parallel compute chips which are ideal for both graphics processing and neural network training has made it by far the leading supplier to the AI industry.
• Arm is a provider of IP and processor designs that are used in billions of devices and upon which a massive and thriving ecosystem has been built.
• Arm’s involvement in AI is has been predominantly focused on processor designs and IP that run cloud-trained neural networks most efficiently on devices at the edge of the network.
• There could be some overlap between the two by more closely marrying the training of algorithms with their execution at the edge but outside of this, there is very little fit at all.
• NVIDIA is committing to maintaining Arm’s HQ in Cambridge, UK and to further expand the site by making it a centre for “world-class AI research and education”.
• NVIDIA is also committing to maintaining the Arm licensing system exactly the way that it is and I am certain that the big licensees like Apple, Qualcomm, MediaTek, Samsung and so on will have been consulted and their sign-off given before going ahead.
• This is crucial because any hint of Arm losing independence (and favouring NVIDIA products) will increase the interest in alternatives such as RISC-V.
• I suspect that NVIDIA will cease making processors that compete directly with Arm licensees using Arm designs which will not be very difficult as its activities in this area have been very quiet for some time.
• This type of independence is difficult, but not impossible to achieve as the example of Samsung which is both a major supplier and competitor to Apple clearly shows.
• Samsung very capably struck this delicate balance for many years and despite constant rumours of favouritism for the internal customer, it has never lost a major client for this reason.
• Hence, I think that it will be possible for NVIDIA to maintain Arm’s independence, but it will have to tread very carefully.
• Overall, I think that the fit between Arm and NVIDIA is not as obvious as the press release claims, and so there still remains the possibility of a return to the market at some point if Arm sees further growth under the NVIDIA umbrella.
• The AI story is also fraught with risk as RFM research has found that the limitations of deep learning combined with excessive hype and expectations of its capabilities are likely to mean that there is another AI winter just around the corner.
• There are still many practical applications for AI in the commercial world but of the three that NVIDIA mentions (healthcare, robotics and self-driving vehicles) only healthcare is likely to deliver on its promises.
• The fact that almost all robots move around on wheels combined with the spectacular failure of autonomous driving to meet its self-imposed deadlines show just how hard these problems are to solve (see here) and deep learning is probably not the answer.
• Deep learning has been around in a practical sense for nearly 10 years and has failed to make a meaningful impact in either of these fields in my opinion.
• Hence, Arm and NVIDIA will need to focus on the practical applications of this technology if NVIDIA shareholders are to earn a return on the $40bn invested.
• This is going to take some time.