This is an article about the downfall of Intel, a company that was once as dominating in hardware as Microsoft was in software and Google in search. It is interesting to see how such changes happen.
What Stuff Matter For Chips
If you do not know anything about a subject, it is difficult to even understand what factors to consider to evaluate the importance of something. So, if you know nothing about economy it is difficult to understand what constitutes a good economy. Indeed it is difficult to understand even if it makes sense to look upon economies in such sense.
This section wants just to explain in very simple terms what matters when evaluating chips so you can follow the rest of the article.
When it comes to CPUs a few things matters, i.e. a few things are relevant:
- the instruction set architecture (ISA), i.e., the set of instructions that a chip supports (e.g., x86, ARM).
- the microarchitecture, i.e., the specific design that implements a set of instructions (e.g., NetBurst, Zen).
- the fabrication process used to create the chip, also referred as the node (e.g., 180 nm)
A specific chip, like the Ryzen 5900X, supports an ISA (x86), following a microarchitecture (Zen 4) and it is built using a fabrication process (7nm). Chips with the same microarchitecture differ in terms of power in the abstract sense. So, one chip can be more powerful than another in the sense that can process more information, do more work than another one.
An instruction set is relevant mainly for software that uses the chip because it represent what basic blocks the software uses. An instruction is not “more powerful” than another, just like a language is not “more powerful” than another. However, it is important because you need to pick one to run a program.
For chips other than CPUs, the instruction set architecture is not relevant.
The fabrication process is important because it affects costs, energy consumption and the level of maximum complexity that a chip can have. A better, more advanced fabrication process allows to create chips that consume less power and can realistically be more complex. A better fabrication process does not make chips cheaper in absolute terms but it makes them cheaper relative to the level of complexity that can support. Just like you could make a car engine that had 1000 hp in the 19261 but it is easier to do it now.
When Exacty Intel Was Dominating?
There was a time when Intel was the domineering company of semiconductors. You could say was the only company that really mattered. That is because Intel had a dominating positions in all three relevant factors for chips. It has now lost the crown on all three.
We were at the end of the last millenium. Intel was the main contributor to the x86 ISA architecture, their Pentiums microarchitectures were unrivalled and they had the most advanced fabrication process. No other architecture was still relevant, so for all use cases the best products were Intel products.
They had one real competitor: AMD, they started (literally and legally) selling copies of Intel chips but they were now making their own chips. The issue was that they were always lagging behind Intel, they were the cheap alternative.
The Rise (and fall) of AMD
Things started to change at the beginning of the new millenium. AMD was now able to design respectable microarchitectures and even rose to design the standard extension to the ISA architecture: x86-64. They were peer of Intel in developing the dominating language of chips.
So, AMD was able to compete on quality of microarchitectures and ISA design, so why Intel was still ahead? The Intel advantage was the fabrication process: making factories (knows as foundries) was becoming very costly and AMD could not compete in that aspect.
This technological superiority and history of reliability basically guaranteed an high-level of marketshare for Intel. AMD had some hits, but it was not able to become a stable and reliable producer of high-quality chips. So, AMD was often relegated to fight for the low-market, even when it had comparable microarchitectures. That is because large computer makers could not afford to design for chips that might dramatically fall in quality from one generation to another.
In addition to that, Intel was also better run as a company, while AMD leadership was less focused. So, at then end of the first decade AMD was on the verge of bankrupcy.
Nvidia Leads the Way
Unfortunately for Intel, while AMD was faltering Nvidia was still going strong. Nvidia design Graphic Processing Units (GPUs), graphic chips. They do not compete with Intel directly, but they are notable for being the most notable success story of fabless companies. They had the same problem that plagued AMD: they could not afford to build foundries and actually make the chips, so they did not.
They design the chips but contract existing foundries to make for them. These foundries could make inferior fabrication nodes than Intel, but that was not a problem for Nvidia, because they did not compete with Intel. They just needed to create the best graphic chips to succeeed and that is what they need. Now, GPUs were still a little market compared to CPUs but they were the best example that you can dominate an advanced chip industry without actually making any hardware.
In fact, AMD was also forced to stop making their own chips and going fabless because they could not afford foundries. They still had to compete with Intel, though.
Smartphones Become Popular
The fundamental strength and weakness of Intel was that they were leaders at everything, so when competing with smaller rivals they could offset temporary weaknesses in one thing (e.g., a bad microarchitecture) with their strength in other things (e.g., fabrication process).
The popularity of smartphones change this reality. Smartphone were not technically a new invention, since they existed for decades, but they become popular with the Apple Iphone. The first issue was that smartphone allowed the rise of another ISA architecture: ARM. It was created by a company that was not just fabless but also did not even design microarchitectures2.
The main issue for smartphone is power consumption and the x86 architecture did not take that aspect into consideration. It was powerful and complicated by design because on desktop and server computers this approach worked best. So, the ARM architecture provided a less power hungry, cheaper option. It was designed by a small company that allow complete control over what to do with the design, so ambitious companies could design their own chips that were compatible with existing hardware, but were better in some way.
Smartphones Become the Biggest Market
The rise of smartphones did impact the CPU market in some direct way, since now many people used smartphone as their primary computing devices. This reduced the value of the traditional CPU market. ARM designs also began to be used in laptop and desktop by Apple, a company that likes the vertical integration afforded by ARM. This certainly limited Intel growth and stature, which at this point was not anymore the indespensible chip company. However, the deadly issue was another: Intel had lost the fabrication process crown.
As mentioned, Nvidia was a fabless company, relying on third-party to create their chips, but their market was smaller, so Intel still had margins to mantain a lead in fabrication. The smartphone market instead became bigger than the PC and server market. And all the major designers of chips for smartphone were fabless companies. This meant that the companies making CPU chips for smartphone had enough money to create better fabrication process than Intel’s. And that is what TSMC did since circa 2016.
So, now Intel did not have an advantage in their fabrication process, but a drawback. They did have to create their chips with an inferior fabrication process. And when AMD recovered and created Zen, they now had an equal quality microarchitecture produced by a better fabrication process. So, now Intel was in trouble, and still is.
The Future of Intel
AMD is now able to reliably compete with Intel in designing microarchitecture for x86 chips. The x86 architecture is not anymore the only game in town and it is actually relegated to the less important PC market. And Intel has lost the incredible advantage of always having the best fabrication process.
None of these things are likely to change in the near future. There are simply strong market forces and competitors that make impossible for Intel to return to the only company that matters in chips. What it could change is that Intel could regain parity of fabrication process with TSMC. So, that they are no more at a disadvantage with AMD.
Mind you, even with this objective disadvantage, Intel still is the largest producer of CPUs for the PC market, thanks to its long history of parternship with computer manufacturers. So, Intel is not in any danger of failing like AMD did some years ago.
Intel is also trying to enter the GPU market, which is not just used for games anymore, but it is fundamental hardware for Artificial Intelligence (AI). So, Intel can still remain an important player in the chip industry, but now they will have to fight without an overwhelming advantage against everybody else.