If Nvidia Bought Arm Holdings, Would That Be The End Of Arm...

Jon Cartu Claims: If Nvidia Purchased Arm Holdings, Would That Be The Finish Of Arm…

The news that Nvidia

NVDA
is in talks to buy Arm Holdings from SoftBank raises an interesting question. Does this change Arm’s model of being a neutral provider of chip technology to everybody from mobile phone makers, to Amazon for its Echo and Dot products, to Apple

AAPL
for its future Mac computers? In order to explore this question, it helps to have a little understanding of Arm’s business model, and how it is very different from that of a company like Intel

INTC
.

All computers have something that is called an instruction set architecture. This describes how the computer works by providing programmers with a set of things the computer can do, like add two numbers, or store them somewhere. A software engineer then can write programs, which are a sequence of instructions that tell the computer step by step how to accomplish what he or she wants it to do. Historically instruction set architectures mattered a lot, because programs written for a personal computer with an Intel microprocessor in it, for example, could not run on an Apple Macintosh computer with a Motorola or PowerPC microprocessor in it. The Intel chip handled certain operations differently, it stored data differently, the two were not  compatible.

Intel rode the tremendous growth of the IBM

IBM
PC, which used the company’s proprietary x86 instruction set architecture. It benefited enormously because the x86 PC combined with Microsoft

MSFT
Windows became the de facto standard for business computing, and the company invested heavily in designing and producing faster and more powerful chips every year, which it could sell at high prices. The chips often sold for hundreds of dollars each, and they commanded high gross margins. Intel (and Microsoft) made far more profit selling their parts of the PC than the PC makers themselves.

But as computers got more powerful, instruction set architecture started to matter less. This was because of several technological innovations. First, software developers raised the level of abstraction for programmers; in other words they came up with ways to create virtual layers that always looked the same, regardless of what the instruction set architecture down below was. This was made possible by faster processors that could translate from that standard layer to whatever was down below. The best example of such an abstraction is the web browser, which doesn’t really care what kind of computer you are talking to on the other end – how it stores data, how it does its math – it’s all like a layer of insulation that keeps you from having to worry about it. 

Having said that, programs that run natively on a particular instruction set architecture will run more efficiently (and faster) than ones that use a translation layer. But with computing power so cheap and plentiful today, it doesn’t matter so much if you waste some translating that layer abstraction ­– that’s how Apple will migrate from Intel’s x86 to Arm for future Macs (in this case the translation layer would be called an emulator).

So now it starts to get interesting. Arm uses a different instruction set architecture. The company actually got its start as a team that left Apple in the 1990s, and it focused on an instruction set architecture designed for low power consumption. That’s one reason it became popular with handheld devices like phones. But the other important difference is that Arm does not sell chips, it licenses its intellectual property (IP), like the design of its central processing units (CPUs). Usually it licenses IP to companies like Qualcomm

QCOM
or Nvidia, who will then take the CPU design IP and embed it in a larger design of its own, for example the Qualcomm SnapDragon processor family. Instead of paying hundreds of dollars for a single chip, the license fee to Arm might be less than a dollar per chip. So a company that uses an Arm processor in its chip doesn’t have to pay Intel for its manufacturing profit. Qualcomm can go to someone like TSMC with its chip designs, have TSMC manufacture then, and then sell the chips to someone like Apple or Xiaomi for use in their phones. Then it gets to earn the profit selling the chip.

Nvidia uses Arm CPUs – that means they license the Arm IP – in many of its products. This is what makes other companies nervous. Arm has long played “Switzerland,” a neutral supplier to all comers, one that would not also compete with its customers. 

One the one hand, one could make the argument that instruction set architecture is a commodity, and it is all the other design IP that a company surrounds the Arm core with that forms the real basis of competition. This is the argument of people who build on top of a Linux operating system – hey that operating system piece is a commodity so it doesn’t matter. On the other hand, the worry might be that Nvidia will influence Arm’s future R&D directions in a way to advantage itself over its competitors, or provide early information on future designs in a way that conveys advantage. I guess if you pay $32 billion, you might feel entitled to do just that.

How will other companies like AMD, Apple, Qualcomm, or NXP Semiconductor

NXPI
, just to name a few, feel about licensing IP from a unit of Nvidia? That’s a question Nvidia should be asking as it contemplates its move.

Jonathan Cartu

About the author

Leave a Reply

Your email address will not be published. Required fields are marked *