Blog post

Will Nvidia be allowed to ARM up?

By Bill Ray | September 22, 2020 | 0 Comments

Nvidia is trying, quite literally, to add an ARM to its business, acquiring the company from Japanese investor Softbank. The deal would diversify Nvidia’s revenue, and perhaps lay the ground for business expansion, but that’s assuming the companies involved can make it happen.

That outcome is far from certain. The deal will take 18 months to complete, which is a very long time in the semiconductor industry – and even longer in international politics. Nvidia has offered $12bn in cash, the rest in shares, which will be fine as long as the price of those shares remains high. But that might not matter, as the US, the Chinese, and the British, will all have to approve, and getting that lot to agree on anything would be a major achievement.

The semiconductor value chain is long, and the markets global, so any country with a reasonable amount of domestic consumption gets a say in big mergers of this type. China certainly fits that criteria, and has shown itself more than willing to get involved – having killed off Qualcomm’s acquisition of NXP in 2018.

Under Nvida’s ownership ARM will be obliged to follow US law. That means restricting technology shipments to any company the US thinks is shifty (currently Huawei and a bunch of smaller vendors, but that list could grow). Chinese approval could be a useful bargaining chip with the US administration (whomever that might be in 2 months), but China will certainly want cast-iron guarantees that the flow of ARM technology won’t be disrupted. It’s hard to see the US providing those, or China believing them if they did.

The British might be assuaged by the promised “World-Class AI Research Center”. Jenssen Huang (Nvidia CEO) has offered to sign “legally binding documents” preserving UK jobs, but the unions are already pointing out that a lot of jobs fall outside that guarantee.

The Americans will also need to approve, which should be easy – America would seem to have everything to gain. However, the ongoing confusion over TikTok and Oracle shows how a simple deal can quickly turn into a quagmire with the right tweet or two, so even American approval isn’t a given.

ARM’s existing customers might not approve either. Nvidia is adamant that the change of ownership won’t impact the licensing, but that’s hard to reconcile with the existing competitive landscape. ARM doesn’t just license chip designs; it works with customers to integrate those designs into their own silicon, and in doing so it gains a significant insight into development cycles and roadmaps. It’s very hard to imagine AMD (which embeds an ARM processor in its Ryzen and Epyc chips) inviting Nvidia for a tour of its design labs, no matter how solid the fire wall. Even those taking off-the-shelf ARM designs might balk at having Nvidia poring over their shipment numbers and future orders.

At the lower end that might push customers to consider RISC-V, or other competitors. At the higher end (and in smartphones) there’s nothing to compete with ARM, for the moment, but any attempt to increase licence fees would be met with fierce resistance.

We have to ask, given the situation, why Nvida thinks this is a good idea? Softbank won’t lose – hanging on to ARM is much easier with $2bn in breakup money when (if) the deal goes south, but why is Nvidia taking such a risk?

We’re still sifting the evidence, talking to the industry, and drawing our conclusions. If you’re a Gartner customer then we’d love to talk to you about how we see things developing, otherwise you’ll just have to wait for the next blog when we’ll take a look at what Nvidia might be hoping to gain.

The Gartner Blog Network provides an opportunity for Gartner analysts to test ideas and move research forward. Because the content posted by Gartner analysts on this site does not undergo our standard editorial review, all comments or opinions expressed hereunder are those of the individual contributors and do not represent the views of Gartner, Inc. or its management.

Leave a Comment