One thing that I think this article misses is a way in which Intel explains it's business internally. The author kind of presents it as "You have CPU design companies, CPU design companies use companies like TSMC to fab the chips, but Intel does it in house and this is a competitive advantage". This is not how Intel explains it's business internally historically. Historically it's the other way around, Intel is a manufacturing company that happens to design CPUs to drive demand.
Also, I think that the thing about their chip design not being a relevant strategic advantage is wrong. The reason they lost the mobile market is to a large extent because their chip design was bad.
The other thing to mention of course is that strategically breaking the foundry off from the rest of the Intel business lines Intel up to basically dump their foundry the way AMD did. Strategically, Intel's foundry pays a massive cost (a) by canabilising the profit of the product group - you now have an internal allocation of costs/profits that don't functionally exist, and (b) competitors don't like to use your foundry, because you're likely to be their biggest competitor and have in the past used foundry customers as terrible acquisition targets.
The problem with keeping the foundry business is they're probably never going to get back to competitive at leading edge and that dooms end products that rely on having the best process to be competitive. In times past Intel dominated the market by volume and that scale gave them the ability to pour a lot into process development. Today they are a small and shrinking part of the overall chip business so the scale advantage goes to TSMC etc. They may technically be able to spend themselves back to performance parity with TSMC but that will hurt their margins. Can't turn back the clock. There's still plenty of use for having mature older processes around for the majority of chips which are not high end compute but then again likely not at their historical margins unless captive customers like military can prop that up.
> The problem with keeping the foundry business is they're probably never going to get back to competitive at leading edge and that dooms end products that rely on having the best process to be competitive.
FWIW I feel like I remember hearing similar remarks about AMD when they spun off GloFo. It's easy to assume that whoever is currently in front, will stay in front. On a 1-5 year time scale, it's a safe bet to be sure. Yet, Intel's own loss of process leadership tells us that things can change in the long run.
I posed this question in a top-level comment, but I'll throw it out here too: world governments are more interested than ever in making sure their fabs are on the leading edge. Could this serve to create more parity than we've seen historically?
I suspect AMD's assessment at the time was similar so they offloaded fab and hopped on the TSMC bandwagon. It's worked out well for them. Interestingly Global Foundries decided not to keep chasing leading edge and instead invest deeply in nominal 12nm. They describe it as a "mature platform" that is "reliable and trusted for automotive, consumer, industrial and aerospace & defense applications".
As for whether this dynamic helps level the playing field between governments, I don't see how it would. Ever more expensive fabs make it harder rather than easier to stay competitive.
The aviation world has some historical parallels. China has been working to catch up on jet engines for at least 40 years and they're still not quite there yet. Not for lack of effort, engineering talent, theoretical understanding, etc, certain kinds of industrial capability are just hard and slow to build. A much smaller country would have no hope. My guess is chip production ends up being similar.
> I posed this question in a top-level comment, but I'll throw it out here too: world governments are more interested than ever in making sure their fabs are on the leading edge. Could this serve to create more parity than we've seen historically?
> The reason they lost the mobile market is to a large extent because their chip design was bad.
Or more directly because they scraped XScale since they didn't want to compete with other ARM designers. Wouldn't this be the opposite of what "a manufacturing company that happens to design CPUs to drive demand." would do?
Intels foundries might be considered strategic assets for the US government. Spinning them out works potentially create a longterm risk for the country.
It isn't that their chip design is bad, it's that the profit margins weren't what they wanted. So they passed on the whole market. Theoretically this means they can direct those resources at other more profitable business ventures. As we've seen this is obviously false.
Nice writeup overall, but I think it falls victim to the "trusting the number on the tin" trap. If you look purely at the names TSMC and Intel assigned to their process nodes in the 2010s, it'd be easy to look at the timeline and think "TSMC got smaller transistors than Intel, and so Intel fell behind. Ergo, the core architecture didn't matter."
In reality, TSMC only caught up to Intel's process in 2016. The playing field remained pretty even over the ensuing 4 years. The level playing field exposed Intel's stagnant core architecture, breaking their datacenter hegemony.
I think the author gets the big picture right - the process node sets the bar. But, I do wonder if the author glosses over architecture too easily. Sure, history tells us that it only matters when the "process playing field" is level. But, times are changing. There are fewer players than ever in the cutting-edge process arena, yet the cost of implementing those new processes marches forward at its usual exponential pace. Major world powers are more interested than ever in keeping their domestic fabs at the forefront of the field. In this environment, can anyone safely assume they'll be able to lean into a long-term process advantage?
The quality of the CMOS manufacturing process is the main factor that determines the multi-threaded performance of a CPU when it is constrained by a power limit.
The design of the CPU matters for this quality criterion only in the unlikely scenario when the design team was not competent and they have made a far from optimal design.
When AMD has transitioned to TSMC in 2019, they have instantly got a much better MT performance per watt than Intel, so the TSMC "7 nm" was already better than anything Intel had in 2019.
At that time, "7 nm" was already an old process for TSMC, having been used by Apple since 2018. Therefore, at the latest in 2018 TSMC was well in advance over Intel.
So what you said, that "TSMC only caught up to Intel's process in 2016. The playing field remained pretty even over the ensuing 4 years." is only partially correct. No later than 2018, i.e. no later than 2 years after 2016, "the playing field" was no longer even.
Perhaps TSMC was already above Intel long before 2018, but there are no older products in direct competition that would allow us to judge whose CMOS process was better.
> The design of the CPU matters for this quality criterion only in the unlikely scenario when the design team was not competent and they have made a far from optimal design.
I think this is an overstatement in a couple of ways:
> unlikely scenario
We're still less than a decade removed from the final descendants of Bulldozer. We can't directly compare it to Zen because of the process differences, but Bulldozer sucked in several ways that weren't related to its 28nm process.
> made a far from optimal design
AMD and Intel continue to release improved x86 architectures. Alder Lake introduced heterogeneous cores to x86. Zen 5's dual decode path is less visible, but an innovation in its own right. If there is a long-term local max in x86 design, we haven't found it yet.
That's not to say flagrantly bad design is impossible (again, Bulldozer) - it's only to say that there are indeed two avenues to making a CPU better. Even if one avenue (process) sets the hypothetical ceiling for the other (architecture), we haven't found that ceiling very often for any particular process node. Incremental improvements are a routine part of any uarch, even when the process remains the same.
Note to author, you don't need 6+ clumsy paragraphs explaining your Ohm's Law analogy. You can just explain you're using an analogy in a single sentence.
As an aside, I feel the analogy in the introduction undersells Ohm’s law. This is not a commonly understood (or explained) point below the college physics level, but it makes the already overstretched statement that “[t]he best ideal models are simple” almost painful: as far as linear laws go, Ohm’s law is a miracle. Sure, there are linear approximations to almost everything, but with a few exceptions they fall apart the moment you look them funny, especially the bulk ones (like Ohm’s law instead of like the diffusion equation). In short, you’re much more likely to encounter linear drag than Ohm’s law.
The best ideal models are simple. Unfortunately for ideal models, more often than not, data says no[1].
I think Intel stock is a good buy now to hold for 3-5 years now. I think the shakeup under way and engineering leadership will most likely bring it back. Not financial advise lol.
For it to be a good buy at $20 there needs to be a path to exit somewhere noticeably over that.
In particular, you need to be somewhat optimistic that whatever happens to Intel at the end of the road will go in favour of the shareholders. Whether it's split into foundry and design, or acquired by someone who wants the fabs, or stolen by the government on national security grounds. I don't know whether any of those exits result in shareholders gaining money. It seems unlikely that any involve shareholders gaining 50% or so.
Or you could bet on one of their competitors who seem to know what they're doing. Superficially I like global foundry as a US centric manufacturer of semiconductors.
Up until recently, I would disagree. My intent was to simply hold Intel stock forever and collect the dividends along the way. Then they eliminated the dividends.
> "Last fall, a Wall Street analyst called AMD “un-investable,” following the announcement of its terrible Q3 2012 numbers and the sudden loss of its CFO."
Intel bylaws state that the CEO has to retire at age 65. Pat will reach that age in March 2026. I haven't seen anyone being groomed to take the lead in 18 months, so I am concerned that the leadership transition will be jarring.
It would be good to have some additional competition and choices in this space. If they are strategic and have long term vision they might do it but would need to understand that they're not dealing with the same as the CPU market, the audience is very different and diverse in their goals and motivations.
Are that many chips actually affected (AFAIK it's only consumer desktop CPUs which is a niche market)? They also seemingly changed their warranty policy, so unless the melting CPU caused additional damage what would they be sued over?
Given the confusing messaging from Intel on this, my personal guidance is to assume all Intel CPUs are affected from the prior few generations. Based on the lack of coherent messaging from Intel internal QA is either non-existent or simply ignored, so it seems likely Intel doesn't really know what product has been affected.
Level 1 Tech Wendell implied that it’s been seen in server cpus as well but seems less likely as Max TDP is more controlled so max current damage seems less likely.
Will the need to build newer more advanced (and expensive) fabs continue forever? Will it ever get to the point where there's 'enough' transistors to not need to build more chips, but rather keep pumping out existing models?
My question hinges on moore's law eventually rounding off, but eventually 'more speed' doesn't equal business outcomes for companies using these chips right? Specifically for CPU-style chips, not GPUs*
The duopoly used to be the only game in town; now ARM is everywhere. In this situation, where Intel and AMD get to bloody each other over x86_64 price and performance, in a shrinking market, while carrying more historical/operational baggage than most of their non-x86 competitors, doesn’t seem like a bright future for either of them.
There’s a throwaway line about not being able to empirically test the business strategy model as applied to Intel, but that’s kind of a key point. Ohm’s Law can be empirically validated, so we know we’re not vastly oversimplifying in modeling because it stands up under empirical test. Even if you thought this kind of analogization between the physical and human worlds was a good idea (and I very much don’t think that), the lack of sanity checking against empirical reality on the business strategy side stretches the analogy to a breaking point.
Look, the post is fine as far as it goes. If you want to tell a simple story, that’s fine. The problems start when you try to understand what’s actually happening in the real, chaotic world (like why Intel had a bad second quarter or why it laid off a bunch of people) with reference only to your too simple model.
Also, I think that the thing about their chip design not being a relevant strategic advantage is wrong. The reason they lost the mobile market is to a large extent because their chip design was bad.
The other thing to mention of course is that strategically breaking the foundry off from the rest of the Intel business lines Intel up to basically dump their foundry the way AMD did. Strategically, Intel's foundry pays a massive cost (a) by canabilising the profit of the product group - you now have an internal allocation of costs/profits that don't functionally exist, and (b) competitors don't like to use your foundry, because you're likely to be their biggest competitor and have in the past used foundry customers as terrible acquisition targets.