Given Intel's operational track record for the last say 7 years, a manufacturing group that's been tightly coupled to its internal design group for eons focused on a very narrow product line, its past half-hearted and unsuccessful attempts at being a foundry, competing against its potential customers, years of organizational and operational mal and under investment, and a competitor like TSMC, it doesn't seem like this has a good probability of success.
Anything can happen, but basically, Gelsinger is selling a pivot that goes against Intel's intrinsic nature for a really long time. Which customers are going to sign up to be the test customers for Intel for anything of meaningful scale? I guess Intel could discount prices immensely like Samsung, but is that worth the risk to a major player?
If I had to pick a company to be a national US champion to revitalize the US semiconductor industry, is Intel really the right pick vs. say Texas Instruments?
Regarding Texas Instruments, what's the latest and most advanced fabrication facility they have? The latest information I could find suggest 45nm.
The options in the US don't really look so bleak to go back to 45nm. Intel, Samsung and TSMC have all announced plans for new fabrication facilities in the US. $20B for two plants in Arizona for Intel, $17 billion for new plant and upgrades to S2 in Austin for Samsung, and $12B for a new plant in Arizona for TSMC. I think this is the largest semiconductor investment in the US of all time. Of course it is all planned and plans may fall through.
I was thinking more along the lines of selecting a company that plays better with the rest of the industry and has a better operational history if you wanted to bring back more of a general semiconductor manufacturing ecosystem back to the US. Focus more on building out capabilities for the rest of the industries that don't need bleeding edge node development while Intel figures out its bleeding edge.
Intel strikes me as historically one of the more ethically challenged companies that heavily relied on monopolistic power. And then when it got into trouble and that monopoly's cracks were widening at a good pace, it started waving the flag after spending $20B+ on stock buybacks.
Intel has the need for leading edge logic processes and the money to try it. No other US company does. Micron is focused on memory and will not go into logic processes.
There are a few contract fabs in the US but none of them are anywhere near the volume Intel will have to offer as 14nm lines become available for other customers.
Anyway, I don't understand your proposal. Intel is doing what it can. Other US semiconductor companies are doing what they can. Unless you want the government to pressure Intel to sell some of their 14nm fabs to companies who have an interest in contract manufacturing I suspect this is just the way things will go.
>Focus more on building out capabilities for the rest of the industries that don't need bleeding edge node development
i think that would be a huge strategic error. Most of those "the rest of the industries" are in turn going to get access to and reap the benefits of the bleeding edge as soon as they can, so building out non-bleeding edge specifically for them is a losing game.
The announcement itself seems to be a kitchen sink. They are going to do everything in a new and cool way! Kind of obvious that Pat can't break the death grip of the manufacturing people - after all it is that vertical integration that has brought all those record revenues in the recent years. Of course it has also caused Intel to fall that far behind.
Interesting that both Intel and TSMC chose AZ. Isn't chip manufacturing very water intensive? Other states like say Michigan have more water than they know what to do with, between their giant aquifers, great lakes, and 1 meter of annual rainfall.
> Arizona has high interest because it has the five key factors for semiconductor fabs – available land, infrastructure (power, water, etc.), skilled talent, no natural disasters, and favorable tax incentives.
Not sure if the author is simply reciting received wisdom of dubious accuracy. Perhaps some of those things are far more important than others--e.g. pre-existing talent and tax incentives.
Or maybe our assumptions are wrong. Last month I made a similar point as yours regarding water and nuclear power plants in Nevada, to which someone replied noting the Palo Verde Nuclear Generating Station in Arizona, the largest power plant in the United States, which sources its water from sewage. See https://news.ycombinator.com/item?id=26176034 and https://en.wikipedia.org/wiki/Palo_Verde_Nuclear_Generating_... In addition to providing an example of water utilization, I imagine the cheap power produced by Palo Verde probably also figures into siting preferences.
Fun Fact: While Phoenix is indeed in a desert, it is a high desert (unlike the Mojave desert in Southern California), and there are canals that bring water from the Northern part of the Arizona, which has more mountains (in terms of square miles) than Switzerland. Also lots of great places to go skiing.
Arizona has so much water that historically it spent 2/3 of its water on growing water intensive crops like alfalfa and cotton. The nice thing about growing crops in the desert is year round sunlight allows you to get more harvests in. As rain damages cotton, the ideal place to grow it would be something like a desert with canals bringing in water from mountains nearby.
But semiconductor plants offer much better revenue per gallon of water than alfalfa or cotton, so it makes a lot of sense that Arizona is trying to migrate its water allocation to higher value-add industries, and of course the state would rather be known for semiconductors than for cotton.
Seems Intel senses the way the semi market has shifted.
In the early days of computers it made a lot of sense to co-develop IP and fabrication techniques, especially for CPUs. CPU IP was a way to move up the value chain and capture more $, and CPU IP had huge lock-in.
Nowadays better tooling makes it easier to develop chips, better software reduces architecture lock-in, and it's all about mass customization (of IP blocks) for thousands of specific use cases as the market has opened up and chips are in everything. The value chain has shifted and its more important to differentiate on fabrication than it is on IP.
In the short run x86 is a cash cow, but in the long run it steals focus away from pursuing the more optimal pure play foundry approach. Intel's main risk is being the worst of both worlds - less attractive IP and less attractive fabrication.
Since chips are so important there's also geopolitical considerations at play. If you want to build a semiconductor ecosystem in a region (like the US), you need contract foundries. If you just have Intel puttering along making their own stuff and ignoring others, you get a bunch of local chip designers having to outsource manufacturing, which creates supply chain risk. I imagine as Intel shifts it will get favorable treatment by US Gov't in the way TSMC does from Taiwanese Gov't.
But you can get an X86 plus ??? with the Intel fabs, this is huge. Intel will be your partner for your post x86 journey. This is clearly a "crossing the innovators chasm" territory.
Foundry service seems like a good idea. 14nm may be a bit long in the tooth for modern phones, laptops, servers but it would be a cutting edge process for most embedded products. Intel built more 14nm capacity than usual because of the 10nm problems. As they finally move to 10nm and 7nm EUV processes, it makes sense to try to sell their 14nm fab space/time to anyone who wants it.
Or maybe they could pivot and work on foundry equipment that small businesses could setup in their garage?
I think it should be a good time to start democratising the chip production and so far there is probably no company who can provide tools in this segment affordable for small and medium businesses.
First of all there are serious economies of scale to chip processing, especially at these low scale, that I think it's unlikely a SME could compete in this space.
Secondly chipmaking is not always the most environmentally friendly process, at least on a local level (Intel[2], AMD[1] have superfund sites at former plants, they aren't the only one), so I'm not sure people should want someone running one "from their garage"
Economies of scale work against that model. Unless for hyper-specialized chips, and even, there is no way low-volume companies can compete against the industry behemoths.
I suspect because none of that has much to do with how Intel operates. They do not even make the machines they use in these processes. So it would be one hell of a pivot.
Companies like Nikon and ASML, who make lithography machines and so on, are not really interesting in the garage fab space but that has nothing to do with Intel.
Hopefully Gelsinger will be to Intel what Nadella was to Microsoft. Both are engineers following finance people who while they still made a lot of money, lost the tech initiative.
I would love to see a 3 way competition between Intel, AMD, and Apple for consumer chips.
I was speaking here mainly for a consumer perspective buying complete systems. For me, my personal computers such as my laptop and/or desktop will likely contain a CPU made by Apple(if I buy a Mac) or Intel/AMD(if I buy a PC). Right now for laptops, Apple is blowing away the competition, especially Intel. I would love for Intel to get back in the game so I have more choices.
Even if they remain vertically-integrated, if they can sell an Apple datacenter blade or produce bespoke architectures for major cloud providers, the market is large -- Apple-scale.
Cloud providers want dependable suppliers with low costs, openness with design, integration with other suppliers and long term roadmaps and support. Outside dependability, Apple is not good on any of the other aspects. Apple run a high margin business, and their processes reflect that.
Apple are a high margin business. I highly doubt they want to harm their gross margin, but its possible. Apple however have not been that successful co-designing.... well, anything. They are highly opinionated that they are correct and this has typically lead to them blazing new trails, but has also lead to mis-steps. For bespoke systems, that would rely on building what the customer asks for. Apple's generally isn't great with working with outside suppliers, generally trying to bring the work inside. Finally this would require apple to tie their hands (in the same way microsoft has) with compatibility to provide the long term support that the vendors require.
Apple would never become the suppliers the cloud providers want. The only way they would be chosen is if the product is significantly better then what otherwise exists. I can easily see Apple leapfrogging incumbent server vendors for performance or such for a short term, but I expect the others will catch up very quickly. Cloud hardware is already well integrated by companies motivated to eek out every little bit of performance (see facebook and the OCP). Despite being significantly more different from an engineering perspective, I actually think manufacturing cars is more similar from a business perspective for Apple.
Apple is a consumer electronics company and they keep reiterating that. Their whole strategy is to be an island unto themselves where they can maintain high gross margins and squeeze their suppliers.
They're not interested in competing on other people's terms in commoditized markets. They sell luxury goods.
I'm not sure about your 'half' figures, but that's how Apple positions itself. An iPhone is a status symbol as much as a phone. Macs are too, but to a much lesser degree. Also their entire line up is not luxurious, but some items are nakedly luxury items like the some of the watches.
One of the reasons for Apple's success is that it's managed to pull off mass-market luxury better than anyone else.
Arguably, luxury items doesn't have to be exclusive, just overpriced.
It would be interesting to see how that would work and what sort of unique features Apple would offer in a proper enterprise product, but I'm struggling to imagine Apple returning to that market in earnest. Even Xserve was not really about being a server in its own right so much as it was for providing infrastructure to support their desktops.
Apple sells luxury goods. They don't seem interested in the datacenter market. Plus it seems like all the cloud companies are designing their own ARM chips, I doubt they want apple's.
>Gelsinger will be to Intel what Nadella was to Microsoft
Nadella is good only because his predecessor was crap. ( Sorry Steve Ballmer ).
I dont think they are in the same league. Gelsinger will be a far better CEO as this company roadmap has shown. Engineers alone dont make great CEO, that was the word Andy Grove told Gelsinger. You need a product person, with vision, strategy, engineering and business senses. And somehow they are a very rare breed.
Samsung, Global Foundries, UMC, and SMIC are other large players in the foundry space. It would be very difficult for TSMC to eliminate these competitors.
Intel has proven for the last 4 or so years that a mature 14nm node can be competitive with <=10nm nodes.
Progress in chip manufacturing doesn't happen exclusively at the node size of the cutting edge. As the manufacturing technology matures, I'm sure we'll see advances across all the major foundries, even if they aren't all competing to be the first to a new node. And the fact that all of these foundries remain in business is a strong indicator that perhaps the most expensive, most cutting edge manufacturing process is not optimal for every single chip manufactured.
Some of these foundries explicitly do not compete on cutting edge node-sizes. Global Foundries, for instance, have said they will focus on specialty circuitry, but will not try to compete on things like <10nm node sizes. I think it is fair to say they are not really competitors as their business is sufficiently different.
Whether they compete on cutting edge nodes is irrelevant to whether they compete. My comment was explicitly in response to the idea of a TSMC foundry monopoly.
It's certainly valuable to understand the specific nature of their competition, but such conversation tautologically defines TSMC as not-monopoly.
Ninja-edit: And even if you want to restrict the definition of competition in chip manufacturing to only those foundries with cutting edge nodes, then Samsung's existence is sufficient argument against the claim of monopoly on that front.
"Another trust issue is competing with customers. Today’s systems companies are the fastest growing fabless customers (Apple et al) and who is one of the biggest systems companies in the world? Samsung. Which is why the majority of Samsung Foundry customers are chip only companies. Word to the wise for Intel Foundry Services do not compete with customers."
Intel foundry services must be a pure play foundry like TSMC in order to win customer's trust.
Smart move on Intel's part, there has been a big push for US manufacturing capacity and this will feed that. It is interesting to me that it took a crisis to force Intel to start offering IDM services. Something that I think they should have done years ago to absorb extra capacity.
IBM, US Fabs, $20B Foundry Investment, IFS ( Intel Foundry Services ) , Open / Industry EDA Tooling. This is all about US Fabs. And this is what Intel should have done in 2015. But the timing right now is simply perfect. I still get very emotional seeing Pat Gelsinger with Intel [1], but this whole strategy is about as good as it gets.
I mean one could write a thousand word essay lining these up from a geo-political point of view.
I am wondering if Apple would test IFS with their Modem. It make sense to have 2nd Foundry source, and Apple doesn't want to fund Samsung Foundry, NAND and DRAM. The best thing about all of these is the new IFS services, along with two additional Fabs meant lots more leading edge wafer capacity.
Then there is x86. All the essential x86 patents has expired. x86-64 was launched in 2003? I kind of wish they made a licensable "cleaned up" x86 and call it AE86 :)
Microsoft has 6x the cash on hand compared to Intel, 7x the market cap, and roughly 3.5x less debt-as-a-percentage-of-market-cap. $100 billion is nearly 40% of Intel's total market cap; $10 billion is less than half of a percent of Microsoft's total market cap.
Basically, Microsoft is just much much bigger and has much much more cash on hand and much much more ability to raise debt compared to Intel. All of these factors mean that relatively speaking, $10 billion to Microsoft is a sum that would barely even make a blip on their annual results; $100 billion is life-or-death for Intel. I don't think it's clear that Intel would even be able to scrounge up $100 billion if they had to.
This makes me wonder why Microsoft doesn't invest more in semiconductor tech. Microsoft and Intel's interests are very much aligned, and they have a common competitor (Apple) who is ahead of them both.
I've been theorizing that Microsoft could get involved in the move to build a high end fab stateside, mostly cause they are already well integrated in the military industrial complex.
Alternatively, Microsoft could acquire the x86 business from Intel and then invest in Intel as a dedicated foundry.
Although, that raises the question of whether MS would be better off staying married to x86 long-term or moving in the direction of Apple Silicon / Rosetta. I'm inclined to think that they would be best served by sticking with x86 long-term for the sake of optimal backwards-compatibility, so long as there isn't some inherent reason for the x86 instruction set to be dramatically less efficient than ARM. I don't see why they couldn't ultimately catch up to or at least become competitive with Apple by pushing development and adoption of x86 SOCs alongside standards like Evo, which their taking over the business could potentially help accelerate.
I think right now any US company that can step up to the plate and fund fabs will get a green light. We're in pretty dire straits right now and this semiconductor shortage is going to last for a while.
You can stretch it all you want but while iPhones may have powerful CPUs that could qualify them as "computers" they're not general purpose computers, so while I could use them to write an email and book a train ticket, I can't do any school or office work on them and iPads mostly remain Netflix/youtube/redit browsing devices for the couch after the honeymoon period is over so can you really put them in the same category as "computers"?
My smart fridge also has a powerful ARM CPU and a touchscreen and a modern OS. Is it also a computer? Can I use it for productive work? Doe every wearable and IoT device qualify as a "computer"?
So sticking back to general purpose computers that are not locked in a vendor ecosystem and let you install and run whatever you want from whatever sources, Apple is less than 10%, so yeah, with ~80% marketshare Microsoft is basically a monopoly in that space.
I think there are two parts to this answer: the fabless component (meaning designing chips), and the fab component (meaning manufacturing chips). Taking Apple as an example: Apple is fully in the fabless part with Apple Silicon, but Apple has no in-house fab component (they outsource the chip manufacturing to TSMC).
For the chip design part, Microsoft to some degree is already doing this. Microsoft has worked with AMD and Qualcomm to design custom chips for things like the XBox and Surface devices, and word on the street is that Microsoft is moving into designing their own ARM chips top-to-bottom similar to Apple [1]. I think moving into custom chip design is no brainer for all of the tech players with a market cap > $1 trillion, and as far as I'm aware, most of them are already doing it.
The second part is the much trickier part. Actually getting into the fab business at a cutting edge process node is extremely difficult, because the fab business is both extremely capital intensive AND takes an enormous amount of time to build up. If Microsoft or any other tech player wanted to muscle their way into the cutting edge fab business, the requirement isn't a one-time $20 billion investment; it's $20 billion a year every year for at least a decade just to catch up. The catching up part is insanely difficult too; for starters, the EUV photolithography machines you need to even begin building a fab are made by only like two companies in the world, and both companies are backordered for years because of TSMC, Samsung, Intel, and others. Then, once you have caught up, you have to continue pouring money into the business forever in order to keep up. TSMC and Samsung and (up until recently) Intel are able to do this because of the sheer volume of manufacturing they do; smaller fabs that don't have the volume just aren't able to keep up because they don't have the revenue. The fab business is a business where economies of scale are really serious and really large, and I don't think Microsoft alone is at a scale large enough to for these economies of scale to start to take effect. Even established chip players like NVIDIA and AMD don't even consider themselves large enough for these economies of scale to apply today.
So the question for Microsoft (and really any fabless chip company) really basically boils down to: do they commit to $20 billion a year every year indefinitely until the end of time to build and run their own in-house cutting edge fab business, or do they just pay TSMC each year to manufacture for them, and thereby offload the risk and cost of the fab business? The overwhelming answer in the industry is: just pay TSMC. That's what Apple, Google, Amazon, NVIDIA, AMD, Qualcomm, etc. all do.
Even if they are to be successful in catching up, they will at best be where intel was before Pat - playing catchup in the chip industry and any time they slip, trying to maximize profit of the already spent r&d budget instead of innovating for the next step (7nm for intel in this case). That opens the risk of being left behind (if they have inferior chips, they’ll have to use them over their superior competitors) in cloud and their product lines (surface pro, book).
Their only potential competitive advantage is if they close down Windows to MS-only chips, which I don’t see happening.
> So the question for Microsoft (and really any fabless chip company) really basically boils down to: do they commit to $20 billion a year every year indefinitely until the end of time to build and run their own in-house cutting edge fab business, or do they just pay TSMC each year to manufacture for them
There's another option for Microsoft: invest $20B once in Intel, in return for some exclusive rights.
That's a loosing play for Intel. The potential market Intel wants to tap with foundry services is way WAY larger than $20 billion, and on the flipside Microsoft doesn't have a need for $20 billion worth of exclusive fab capacity. That's way too low of a price for Intel to give up the wider market and way too high of a price for Microsoft to get what they otherwise could already get by just contracting Intel or TSMC or anyone else normally.
I think the entire point is that the fundamental thesis driving Intel's new strategy (and this is the same thesis behind TSMC, Samsung, etc) is that the cost of developing cutting edge process nodes has reached a point where it is beyond the ability of any single company to afford based on their own product volume. Therefore, if Intel wants to be able to keep up, it needs to expand volume by opening up foundry services to other companies, a-la TSMC. If this thesis is true (and over the past few years it definitely seems to be), then it explains why no individual tech player would want to get into building their own fabs, it also explains why TSMC's model is successful and why Intel is going after that same model now, and it also explains why no foundry would really want to sell exclusivity to any given single player.
Probably not. In retrospect Intel probably could have avoided their 10 nm problems by developing multiple 10 nm variants in parallel but it's not clear that there is even enough hireable talent to do that.
There is enough hireable talent. Samsung is producing 8nm at volume, and TSMC 7nm, with Apple reserving all of the upcoming 5nm.
Clearly the talent is there to support multiple cutting edge nodes developed in parallel. Whether Intel could afford to hire multiple teams like that is a different question.
As an embedded hobbyist, I wondered this the other day. Why Intel wouldn’t open up old fabs to embedded developers. It could be a huge boon to that industry which is clearly growing rapidly, and much is developed overseas at larger sizes.
I hope the new IDM 2.0 strategy doesn't mean the end of the Platinum/Gold/Silver/Bronze branding. My local gas station car wash has for decades used the same branding nomenclature for their various tiers of car wash rigor, and it has been very successful for them. I can't imagine that central processing units should be any different from a car wash.
I get that Arizona has lax regulations but it seems like they’re gonna have a tough time convincing top tier talent to move to Arizona of all places. Maybe it’s just me.
Most of the top tier semiconductor talent is not in Silicon Valley, and some of it is already in Arizona. Intel has had fabs in Arizona since forever, and they have a fairly big footprint there.
This is underwhelming. By the time they manage to get 7nm fab it will already be obsolete.
Why not something more ambitious or admit they cannot compete with Apple and AMD any longer and focus on chipsets or other peripherals?
They could free x86 specification and let other companies with more energy pick a fight with ARM, no?
No one builds a leading-edge fab for one process node. The new factories will be able to support more advanced processes, unless Intel is wildly incompetent.
i don't think "nm" matter anymore. the profitable yield level is more important. Intel's Rocket Lake is still on 14nm. it doesn't look like Intel's 10nm yield is profitable.
ASML's financial reports suggests EUV machine production rate has not been impacted:
>In EUV... we successfully increased our manufacturing output in 2020 to 35 systems, however due to fab readiness of our customers we shipped 32 of these systems
It seems they built EUV machines faster than their customers were able to install them.
Furthermore ASML insists the upgraded NXE:3600D machines will start shipping in mid 2021 as planned.
Anything can happen, but basically, Gelsinger is selling a pivot that goes against Intel's intrinsic nature for a really long time. Which customers are going to sign up to be the test customers for Intel for anything of meaningful scale? I guess Intel could discount prices immensely like Samsung, but is that worth the risk to a major player?
If I had to pick a company to be a national US champion to revitalize the US semiconductor industry, is Intel really the right pick vs. say Texas Instruments?