Intel, ARM and the Innovators Dilemma
Why succeeding as an ARM chip maker will be next to impossible for Intel and AMD
I recently received an interesting challenge to my assertion that Intel and AMD are at a crossroad, due to the rise of ARM processors from the likes of Apple, Amazon, Qualcomm and many others.
Its naive to think days of Intel and AMD are over. They are multi billion $ companies who can just as easily license ARM ISA and integrate that into their own CPUs. In a way, they can create something much better (for PC users at least): an x86/ARM hybrid CPU that can run both ISAs natively. As PC-side software developers slowly transition to ARM, so would they. Then they would use their decades worth of experience in developing ARM CPUs. No harm done..
As you can probably guess, I strongly disagree with this assertion. We can address this issue from a number of angles. Let me do one of the most obvious ones first.
The Difficulty of Competing with Your Own Products
There is a good reason why the big auto-makers struggle with beating Tesla at making electric cars. They face a very real dilemma: If they make an amazing electric car they could undermine their own gasoline car business. GM, Ford and others make more money selling gasoline cars than electric cars. Electric car production will initially be limited, due low battery volume production and many other factors. Thus EV production cannot be scaled up to the same volumes as gasoline car production. Hence GM and Ford cannot make as much money selling them.
This is not an issue for Tesla. They are not competing with themselves, since they don’t make any gasoline cars.
Intel and AMD face the same problem if they start making ARM CPUs. It would be akin to Ford making an EV, a product competing with their existing lineup. They face the problem that if their ARM chip is too good, it will cannibalize the sales of their x86 CPUs. That is not a desirable proposition for either of them. They both enjoy high markup on x86 chips because they enjoy a duopoly. Nobody else can enter the x86 market.
Thus Intel and AMD risk doing the same as GM. That they make an ARM product that simply isn’t good enough, because they are too afraid of making too good of a product. As with GM when they released the Bolt, it means you end up wasting precious development resources on a product that ultimately flops. Not because it was electric, but because it could not offer competitive price/performance/experience compared to the competition.
Exiting the x86 market is a hard thing to do. If Intel decides that ARM is the future, then it makes no sense to continue investment in x86 development. However abandoning the development of x86 risks AMD doubling down and stealing most of the x86 marketshare. Thus Intel is deprived of crucial income in the midst of a transition phase.
Hence AMD and Intel get caught in a prisoners dilemma situation, where nobody is able to make the decision that would be optimal for both of them.
Past Success is No Guarantee for Future Success
While Intel and AMD have had success in the x86 chip market, that does not guarantee future success in the ARM world. History is riddled with the corpses of former successful companies which failed to adjust to the new times.
We don’t even need to go far back in history. Nokia and Blackberry were certain that a new entrant in the market such as Apple posed no threat to them. They were wrong.
The name of the world’s once most successful steam locomotive company is long forgotten. It failed because it kept trying to make the best steam engines in the world when diesel was taking over. That is akin to attempting to keep making the best x86 chip while ARM is taking over. Sooner or later you will run out of steam.
SpaceX has done much the same in the space industry. The Russians used to dominate the space launch industry but have since thrown in the towel and admitted they are no longer capable of competing. Yet it is not long ago that they mocked Elon Musk and his efforts.
It is also easy to overestimate the advantages of established players. Take electric cars for instance. Superficially we think of GM, Ford and BMW as better at making cars than a smaller and younger startup such as Tesla. Yet “making a car” is not some generic skill. It involves expertise in a wide variety of fields. A big part of what companies such as GM and Ford do is to make the gasoline engine. That is a significant part of their skillset. So is everything related to dealing with such an engine: the gas tank, managing the distribution of weight in the car due to the significant mass of the car engine etc.
With an electric car manufacturing the important skillsets change significantly. Does GM and Ford actually have expertise in making electric drive trains? No, in fact they don’t, but Tesla does. In fact Tesla makes by most accounts the the best electric drive train in the industry. What else is of significant advantage? The battery package, of course, another area where Tesla has years of expertise and where they are the industry leader.
Both GM and Ford have over a hundred years of experience making gasoline engines and tanks. But does that matter? Not when building electric motors and battery packs has suddenly become most important.
Upon close inspection it seems insane to assume that GM and Ford should have some natural advantage or skill in this area simply because they previously made gasoline engines, which are based on entirely different principles.
Much the same can be said about AMD and Intel entering the ARM market. Just because they excelled at making x86 chips does not mean that they will excel at making ARM chips. The optimization you make in the micro-architecture will not be the same. As I have discussed in my previous article, a RISC processor, such as ARM, can more easily use an architecture with lots of instruction decoders working in parallel. A CISC processor, such as x86, in contrast is at a disadvantage. It is harder for multiple decoders to pick an individual instruction from an instruction buffer, because you don’t know where one instruction ends and another one begins. That is because CISC instructions are of variable length. Intel and AMD has to circumvent this by brute force trial and error. That is not a solution that scales to more than four parallel decoders.
There are many such differences, which will necessarily lead to different optimization choices in the micro-architectures of AMD and Intel processors. It is thus somewhat premature to assume that these architectures refitted with ARM instruction decoders will be world beating.
Intel’s Leadership Problem
Let us also not forget that Intel has a leadership problem. It has been written and talked about extensively how Intel has turned into another Boeing. A strong engineer and product oriented leadership has been replaced by bean counters, more concerned with cost cutting and quick gains than effective long term strategy and strong product roadmap. It doesn’t matter how large your war chest is if your leadership is weak.
If the world moves towards ARM and Intel does not make their own highly competitive ARM microprocessor, they risk making themselves irrelevant. But there is a catch-22: If Intel attempts to make the best ARM processor possible they risk also undermining the x86 platform itself. If Intel is not careful they could end up being responsible for killing the x86 platform by making a superior ARM chip. Thus it becomes all too tempting to make a half-hearted attempt: A GM Bolt kind of attempt. We all know how that ends.
Will Intel Go Bankrupt?
All this doesn’t mean Intel will go bankrupt Nokia style, but it could become another IBM. Still a relatively successful company but no longer a trendsetter or dominant player. IBM gave up making PCs long ago, but found fortunes in other ventures. It is not impossible to imagine a similar destiny for Intel. Dominance on the desktop is simply replaced by a more varied but less lucrative business. x86 cannot be replaced over night. Intel can still keep a steady income from x86 sales. But if x86 becomes a market segment without growth, Intel will cease to be the important player it is today.
Encircled by Enemies
Finally we must not forget that Intel and AMD have to a large degree been shielded from competitors. Yes, AMD has had their fill fighting Intel, but Intel has always had plenty of resources to keep AMD at bay.
In the ARM world things will be very, very different. Intel will go up against numerous other companies with deep pockets. They will no longer be the one who can outspend everybody else. Google is making their own ARM chips for Android devices. They have deep pockets.
Nvidia is on track to acquire ARM, and you can bet they are going to utilize that position to build ARM SoCs with Nvidia GPUs inside. How will Intel respond to that? Intel has been far behind on GPUs but have recently upped their game with their Xe GPUs.
Still this is a field Nvidia dominates. The CUDA programming API is a big part of the reason. CUDA is a standard for programming GPUs which is controlled by Nvidia and widely popular in the industry. That gives Nvidia a leg up in high performance computing (HPC) in a similar way that controlling x86 instruction-set benefits Intel and AMD on the PC desktop.
Nvidia is a successful, aggressive company also loaded with cash. Previously Intel has never had to face them directly since Nvidia has had to partner with Intel, relying on them to provide the CPUs for their solutions. With a switch to ARM, Intel and Nvidia will become competitors.
And Intel will indirectly face Apple. Whatever ARM chips Intel make they will sell them to a computer market Apple is competing in. Apple has much longer experience making ARM chips than Intel. It is one of the most valuable companies in the world and with enormous piles of cash.
Intel struggled to fend off AMD, a much smaller player. How on earth should Intel manage to defend itself effectively against all these other players?
There is really only one way to do that, and that is to focus on a particular niche. Intel will not be able to dominate almost every type of CPU made like they did with x86. They cannot make best selling budget CPUs, desktop class CPUs, sever CPUs, laptop CPUs and whatever else we may label them. They may have to pick one and accept a much smaller share of the pie than they have been accustomed to.
Some may ask whether Apple could face the same kinds of problems as Intel and AMD. Will not all these ARM competitors also cause problems for Apple?
Read more: Apple and Its ARM Competitors
One may also question whether Apple actually picked the right CPU architecture. Isn’t RISC-V the future and not ARM?
Read more: RISC-V: Did Apple Make the Wrong Choice?