6 ms·

> Intel at the time wasn't interested in fabing ARM chips, and the rest is history. Fun fact, in the late 90's, Intel was the biggest manufacturer of ARM chips

by rhelz 3y ago

> Intel at the time wasn't interested in fabing ARM chips, and the rest is history. Fun fact, in the late 90's, Intel was the biggest manufacturer of ARM chips. I was there at the time, and I wrote lots of software to aid in their design. What happened to Intel was twofold: The first problem was that Robert Noyce, Andy Grove, And Gordon Moore died. Each one was a once-in-century intellect, and it took Intel a while to be able to find somebody who could remotely fill those shoes. The second problem was Intel was making so much money on x86 chips, that dedicating any fab capacity to anything else--including chips for Apple's iPhone, would for years have had such a huge opportunity cost, that the shareholders would have sued, and with cause. Because literally (not metaphorically) anything they would have manufactured instead would have had a drastically lower profit margin, and the stock would have tanked. I suspect that the only reason Intel made a foray into ARM chips in the first place was to head off anti-trust accusations, and once the political heat was off, they dropped them like they were hot. By the time the market shifted, almost a generation of mediocre management had left Intel less paranoid, and therefore, less likely to survive.

markhahn 3y ago
It's also interesting that Intel made some specific bad technical choices. (ie, not investing in EUV, assuming SAQP would work forever, ignoring the consequences of everything being power-limited...) This kind of decision is hard, because it's a technical-economic tradeoff, and the latter is more voodoo than math. And that's not even addressing whether you get surprised by things like the LLM boom...
- rhelz 3y ago
  You say: > It's also interesting that Intel made some specific bad technical choices. That's like saying Saudi Arabia made some bad policy decisions. If you are in Saudi Arabia, nothing has a higher return on investment than petroleum. If you invest in anything else, you are leaving money---A LOT of money--on the table. The result is that capital is drained out of any other business, and the country gets so dependent on just one industry that when the oil runs out, it is a major crisis. Put yourself in the place of an Intel CEO. You've just invested $4 Billion in a fab. People will by absolutely every single x86 chip that fab will make ~$40 billion over the course of the technology node's life time. Or, you could make chips for the iPhone, which maybe, perhaps would be a hit? And even if Apple's wildest projections come true, you'll make $8 billion instead of $40 billion, because Apple is not going to pay for Intel to pocket a 90% profit margin, when it could just go to TSMC. So you are a CEO, and wondering how on the next earnings call you are going to justify turning an asset worth $40 Billion into one worth $8 billion....
  - andyferris 3y ago
    Was it not possible to build two fabs and make $48 billion?
    - rhelz 3y ago
      What would you do? Build two fabs and make $48 billion, or build two fabs and make $80 billion? It might be hard to remember now, but Intel was in the position NVIDIA is today: they could sell absolutely every x86 processor they could make, at pretty much any price they cared to name. So even if Intel could have built another fab (which, btw, it couldn't, the world only graduates so many solid-state physics Ph.D.'s per year) but even if they could, they would STILL have used it to make x86 chips. Literally (not metaphorically or hyperbolically) nothing else they could manufacture would have as high a return on investment as making x86 chips did.
      - KerrAvon 3y ago
        I don’t buy this completely. AMD’s 64-bit humiliation of Intel wasn’t that far in the past, and they knew GPUs were important. They could have hedged their bets a bit more on the basis of sound business management without incurring shareholder wrath.
      - hnfong 3y ago
        Google/Alphabet and Facebook/Meta sunk billions into ventures that aren’t half as profitable as your hypothetical 8B deal, and shareholders haven’t sued yet. Of course the founders have majority voting shares, but that doesn’t prevent them from being sued. It only prevents them/their designated CEO from being fired.
        rhelz 3y ago
        Theres a few differences between Intel and Google here. The first two points are from Peter Thiel: 1. Google really didn't know what to do with its excess money, which was why it was just stockpiling it and not using it to invest in its own business. Intel, by contrast, always knew (at least in the Noyce/Grove era) what to do with its money. 2. Google's "side projects" were false-flags and smokescreens to make it look like it wasn't a monopoly. (Thiel says it much better himself in his article and speech titled "competition is for losers"-highly recommended). Intel did a similar thing around the turn of the century, when it started making ARM chips--not because they were the most profitable thing they could be doing but because they didn't want to look like a monopoly. And, of course, you really hit the nail on the thumb here: > Of course the founders have majority voting shares Yeah. The executives of google and facebook just don't face the same accountability from the shareholders that the execs of any honestly-governed public company does. Its a two-edged sword: if google or facebook needs to spend a lot of money to pivot, they have the flexibility to do that in a way which other companies don't. However, the flip side implication is that, e.g. Zuckerberg can just spend $10 billion on a huge project, without being at all held accountable if it fails.
HarHarVeryFunny 3y ago
It seems that what happened to Intel was a lot worse than just missing Andy Grove's paranoia... - They missed the mobile market - They missed the gaming market - They missed the AI/ML market - They got multiple generations behind TSMC and Samsung in the fab business - They lost a whole bunch of market share to AMD, including datacenter The NY Times article is paywalled, so I can't read it, but I have to wonder what kind of fabs they are meant to be building for our (taxpayers) $8.5B ? I wouldn't have any faith in them building a SOTA fab at this point.
- chiefalchemist 3y ago
  When you connect those dots this feels more like a pseudo bailout in disguise.
  - rhelz 3y ago
    Oh, its not a pseudo bailout. It's a bailout full stop. Doesn't mean its a bad idea.
    - inemesitaffia 3y ago
      8 billion can't build half a fab though.
  - coliveira 3y ago
    This is exactly what it is. With the added issue that, in all likelihood, $8B is not even enough to start making a change. The US gov would have a better time giving these $8B to startup companies in the chip space. Intel will just squander this money and give some of it to shareholders.
- rhelz 3y ago
  You are forgetting two things: 1. All those happened after Andy Grove's time. 2. And when Andy Grove was at the helm, Intel did not miss opportunities like that. E.g. when Intel's memory chip business started losing market share to the Japanese (who were at the time a low-wage country), they were able to transition to CPUs in time.
araes 3y ago
> would for years have had such a huge opportunity cost, that the shareholders would have sued, and with cause. I've been considering the Intel issue (had family there) and the other post in the thread mention the decade of misses. This actually seems like one of the most reasonable rationales that I don't hear mentioned very often. You might want to do something, yet your hands our tied, because the investors will sue for poor leadership.
- jacobr1 3y ago
  It is the classic innovator's dilemma. There are ton of incentives to keep the focus on today's cash cow in any business.
starspangled 3y ago
I don't think it was that Intel didn't have enough capacity to manufacture more. Intel was a manufacturing-first company that beat out "high end" and specialized competitors by economies of scale. Over the long term, the strategy was to build as much high performance silicon as possible. The problem looks like they went down the typical dinosaurification route that their vanquished high end competitors the likes of DEC, Sun, HP, SGI, IBM, etc., which was to switch from manufacturing as much as possible, to protecting high margin x86 parts. Their x86 PC and server organization out-muscled their manufacturing organization (either that or finance did), and so the ended up being very reluctant to adopt GPGPUs or ARM mobile because those were a threat to x86. By the time they realized they'd missed the boat in those markets, they put out these ridiculous x86 mobile cores and x86 GPGPUs that were costly failures. This pattern of protecting profit margins on high end products while fixed costs every generation increase and performance increases so low end grows up and erodes high end, causing further retreat and protectionism, is exactly what happened with those aforementioned big iron companies. When the bean counters look at the state of things without an understanding of the technology trends, that's the logical thing to do, but they paint themselves into this corner. Intel might have just been saved by the bell with Dennard scaling breaking down around the time they were losing their manufacturing advantage. That meant that instead of competitors on a generation or two process lead wiping the floor with them as they did to the minicomputers, it was a much smaller disadvantage and they've been able to somewhat hold on. That's my non-insider take on it anyway.

by rhelz 3y ago

markhahn 3y ago

It's also interesting that Intel made some specific bad technical choices. (ie, not investing in EUV, assuming SAQP would work forever, ignoring the consequences of everything being power-limited...) This kind of decision is hard, because it's a technical-economic tradeoff, and the latter is more voodoo than math. And that's not even addressing whether you get surprised by things like the LLM boom...

rhelz 3y ago
You say: > It's also interesting that Intel made some specific bad technical choices. That's like saying Saudi Arabia made some bad policy decisions. If you are in Saudi Arabia, nothing has a higher return on investment than petroleum. If you invest in anything else, you are leaving money---A LOT of money--on the table. The result is that capital is drained out of any other business, and the country gets so dependent on just one industry that when the oil runs out, it is a major crisis. Put yourself in the place of an Intel CEO. You've just invested $4 Billion in a fab. People will by absolutely every single x86 chip that fab will make ~$40 billion over the course of the technology node's life time. Or, you could make chips for the iPhone, which maybe, perhaps would be a hit? And even if Apple's wildest projections come true, you'll make $8 billion instead of $40 billion, because Apple is not going to pay for Intel to pocket a 90% profit margin, when it could just go to TSMC. So you are a CEO, and wondering how on the next earnings call you are going to justify turning an asset worth $40 Billion into one worth $8 billion....
- andyferris 3y ago
  Was it not possible to build two fabs and make $48 billion?
  - rhelz 3y ago
    What would you do? Build two fabs and make $48 billion, or build two fabs and make $80 billion? It might be hard to remember now, but Intel was in the position NVIDIA is today: they could sell absolutely every x86 processor they could make, at pretty much any price they cared to name. So even if Intel could have built another fab (which, btw, it couldn't, the world only graduates so many solid-state physics Ph.D.'s per year) but even if they could, they would STILL have used it to make x86 chips. Literally (not metaphorically or hyperbolically) nothing else they could manufacture would have as high a return on investment as making x86 chips did.
    - KerrAvon 3y ago
      I don’t buy this completely. AMD’s 64-bit humiliation of Intel wasn’t that far in the past, and they knew GPUs were important. They could have hedged their bets a bit more on the basis of sound business management without incurring shareholder wrath.
    - hnfong 3y ago
      Google/Alphabet and Facebook/Meta sunk billions into ventures that aren’t half as profitable as your hypothetical 8B deal, and shareholders haven’t sued yet. Of course the founders have majority voting shares, but that doesn’t prevent them from being sued. It only prevents them/their designated CEO from being fired.
      - rhelz 3y ago
        Theres a few differences between Intel and Google here. The first two points are from Peter Thiel: 1. Google really didn't know what to do with its excess money, which was why it was just stockpiling it and not using it to invest in its own business. Intel, by contrast, always knew (at least in the Noyce/Grove era) what to do with its money. 2. Google's "side projects" were false-flags and smokescreens to make it look like it wasn't a monopoly. (Thiel says it much better himself in his article and speech titled "competition is for losers"-highly recommended). Intel did a similar thing around the turn of the century, when it started making ARM chips--not because they were the most profitable thing they could be doing but because they didn't want to look like a monopoly. And, of course, you really hit the nail on the thumb here: > Of course the founders have majority voting shares Yeah. The executives of google and facebook just don't face the same accountability from the shareholders that the execs of any honestly-governed public company does. Its a two-edged sword: if google or facebook needs to spend a lot of money to pivot, they have the flexibility to do that in a way which other companies don't. However, the flip side implication is that, e.g. Zuckerberg can just spend $10 billion on a huge project, without being at all held accountable if it fails.

HarHarVeryFunny 3y ago

It seems that what happened to Intel was a lot worse than just missing Andy Grove's paranoia... - They missed the mobile market - They missed the gaming market - They missed the AI/ML market - They got multiple generations behind TSMC and Samsung in the fab business - They lost a whole bunch of market share to AMD, including datacenter The NY Times article is paywalled, so I can't read it, but I have to wonder what kind of fabs they are meant to be building for our (taxpayers) $8.5B ? I wouldn't have any faith in them building a SOTA fab at this point.

chiefalchemist 3y ago
When you connect those dots this feels more like a pseudo bailout in disguise.
- rhelz 3y ago
  Oh, its not a pseudo bailout. It's a bailout full stop. Doesn't mean its a bad idea.
  - inemesitaffia 3y ago
    8 billion can't build half a fab though.
- coliveira 3y ago
  This is exactly what it is. With the added issue that, in all likelihood, $8B is not even enough to start making a change. The US gov would have a better time giving these $8B to startup companies in the chip space. Intel will just squander this money and give some of it to shareholders.
rhelz 3y ago
You are forgetting two things: 1. All those happened after Andy Grove's time. 2. And when Andy Grove was at the helm, Intel did not miss opportunities like that. E.g. when Intel's memory chip business started losing market share to the Japanese (who were at the time a low-wage country), they were able to transition to CPUs in time.

araes 3y ago

> would for years have had such a huge opportunity cost, that the shareholders would have sued, and with cause. I've been considering the Intel issue (had family there) and the other post in the thread mention the decade of misses. This actually seems like one of the most reasonable rationales that I don't hear mentioned very often. You might want to do something, yet your hands our tied, because the investors will sue for poor leadership.

jacobr1 3y ago
It is the classic innovator's dilemma. There are ton of incentives to keep the focus on today's cash cow in any business.

starspangled 3y ago

I don't think it was that Intel didn't have enough capacity to manufacture more. Intel was a manufacturing-first company that beat out "high end" and specialized competitors by economies of scale. Over the long term, the strategy was to build as much high performance silicon as possible. The problem looks like they went down the typical dinosaurification route that their vanquished high end competitors the likes of DEC, Sun, HP, SGI, IBM, etc., which was to switch from manufacturing as much as possible, to protecting high margin x86 parts. Their x86 PC and server organization out-muscled their manufacturing organization (either that or finance did), and so the ended up being very reluctant to adopt GPGPUs or ARM mobile because those were a threat to x86. By the time they realized they'd missed the boat in those markets, they put out these ridiculous x86 mobile cores and x86 GPGPUs that were costly failures. This pattern of protecting profit margins on high end products while fixed costs every generation increase and performance increases so low end grows up and erodes high end, causing further retreat and protectionism, is exactly what happened with those aforementioned big iron companies. When the bean counters look at the state of things without an understanding of the technology trends, that's the logical thing to do, but they paint themselves into this corner. Intel might have just been saved by the bell with Dennard scaling breaking down around the time they were losing their manufacturing advantage. That meant that instead of competitors on a generation or two process lead wiping the floor with them as they did to the minicomputers, it was a much smaller disadvantage and they've been able to somewhat hold on. That's my non-insider take on it anyway.