The only players in that market are Samsung, TSMC, and Intel. It'd be crazy to let one of the three global EUV lithography manufacturers disappear because their R&D costs have lowered net revenue temporarily.
People act like Intel has nothing to offer, when they're one of a low single-digits number of companies who are even capable of making modern microchips. Even China won't likely crack EUV for the next 5 years. Throwing all of the money on pure R&D would be madness. And the chips act is investing in a major R&D facility in upstate NY for next gen stuff anyway.
Scrapping their last gen process to focus on 1.8nm was a long term play, and all signs point out to it working out well.
Not to mention, it'll be nice to have at least this one US company making modern microchips, in case the many fabless US microchip companies ever need to switch to domestic production, due to economic or geopolitical reasons that are clearly on the short term horizon.
Intel has always had great "signs" but fell completely apart once the time came for execution.
Which is to say, I'll believe it when I see it. I say that as an Intel fanboi, too.
If Trump/DOGE comes in and slashes Intel funding one way or another, even better.
And then I Googled (to check the spelling of "Gelsinger"), and found this:
https://www.cnbc.com/2024/12/02/intel-ceo-pat-gelsinger-is-o...
https://techcrunch.com/2024/12/02/intel-ceo-pat-gelsinger-re...
The man's retiring.
Hm. Damn.
So who are David Zinsner and Michelle Johnston Holthaus? They seem to be money and product people, not foundry people or architects? I'll need to look more closely, but to my mind this doesn't bode well for the long term.
I see that the stock is up 5%, but I don't trust investors to think long-term enough. Investors are also happy when you Boeing the shit out of a company.
Top 5 in single core performance are Intel chips.
The best Intel core for single-threaded performance, Lion Cove, has an IPC (instructions per clock cycle) about 2% higher than AMD Zen 5.
Currently both cores are made by TSMC and they reach the same clock frequencies, therefore the Intel core is negligibly better in single-thread performance.
Nevertheless, Intel Lion Cove has a much greater area than AMD Zen 5, despite being made with a superior TSMC process. Because of this, Lion Cove is inefficient for multi-threaded performance (because in the same area more smaller cores could be crammed), so Intel is forced to use it in hybrid configurations with Skymont cores, in order to achieve an acceptable multithreaded performance.
The worst is however that Intel is not able to use its up-to-date cores in server CPUs, because they are too slow at design/validation. Their new server CPUs, Sierra Forest and Granite Rapids, use obsolete CPU cores that are not competitive with AMD Zen 5, instead of using the best Intel cores.
Unlike Intel, where 1 year or more of delay between using a core in consumer CPUs and using it in server CPUs is normal, at AMD they launch the corresponding server CPUs only a few months after launching consumer CPUs.
The biggest compute demands right now are all about GPU power, especially with the AI boom. And Intel... well, take a look at those DirectCompute benchmarks. You'll need to scroll past 100+ other cards before you even see them listed (https://www.videocardbenchmark.net/directCompute.html). They make great CPUs, no doubt, but they've got some serious catching up to do in the GPU space if they want to stay relevant in today's parallel processing world.
And in cases where I don't personally know, one of my heuristics for who's more likely on the right side of a question is who's contributing to my understanding, who's setting context, who's comparing alternatives, who seems to be pointing out specific details I didn't know about. Those are the positive indicators. On the not so good side are things that I would call reflexes, which is to say things that are not expressing ideas but are more like muscular reflexes that don't react to or build on the information of the more constructive comments.
While I'm not thrilled with Intel over the past few years, the context of their capabilities to manufacture chips, the specifics about next gen 1.8nm chips, and the way this is complementary to R&D which is happening simultaneously, those allow me to hold two things in my brain at the same time, namely that Intel has made poor strategic decisions lately and also that they nevertheless are stewards of profoundly important and valuable chip design and manufacturing capabilities.
Sure, but have you heard of creative accounting? That's how they manage to move money from one place to another without breaking any contracts...
https://www.intc.com/financial-info/income-statement
So whatever their problems are, throwing more money at R&D is unlikely to solve them.
Also, Intel has committed to not doing stock buybacks and has stopped the dividends completely. They may resume at some point, but it's unlikely to be soon.
If half your income was spent on rent that would be an insane 8,3k a month. and you would still have another 8.3k left.
That left-over amount is more than twice what i earn, and i live fairly comfortably, alone, in the middle of Oslo.
Also, when I interviewed with them out of college in the late 1990s, they offered stock options and if you worked there 7 years, you get a paid 3 month sabbatical. I'm not sure if they still do that; that was in the Andy Grove days.
I wanted to learn databases and application dev. They wanted me to wear a bunny suit and support their VAX/VMS systems. No thanks.
$200k USD would put you in the top 1% of earners here in Canada.
You are a tech engineer and "live comfortably" means that you "live like a tech engineer" in the area. Which means "not making financial sacrifices to work at company X".
Live comfortably for some people means single family home with yard, 4 bedrooms, 2-3 annual vacations, eating out once a week, shopping at expensive grocery stores, etc. etc.
Now put all these $200k or dual income $400k people in the same area and it becomes a race to the bottom of "who is willing to live closest to paycheck to paycheck".
Microchip manufacturing is supposedly one of the most difficult things to do in the world, but why should the smartest people in the world do that if they have a maximum potential of a few hundred thousand dollars per year when writing software can get them far more?
Letting your organization rot for 15 years while the owners reap the rewards is fine, just don’t be surprised when you’re irrelevant in the future. Or maybe do that because you’ll get a bailout from the US federal taxpayers.
They need the money if they want to do foundry business competing against TSMC. And I would argue even $8B is too little. Latest Leading edge Giga Foundry in Taiwan is roughly $20B. You will probably need at least $30B in US.
Very unfortunate Intel is still hanging on to its GPU unit which is burning money. I guess part of the AI hype meant Pat Gelsinger cant let that part go.
I think where Intel really shows promise for GPUs and accelerators is in the price-point area.
The Gaudi 3 accelerators that entered general availability two months ago manage to outdo Nvidia's hopper chips when comparing precision-to-precision, and retail for half as much. The overall success of that line really depends on whether work to overcome the lock in of CUDA will pan out sufficiently. Personally, I think it will, as AMD, Google, Amazon, and others are all making (and successfully rolling out) chips which depend on breaking the industry out of the walled garden of CUDA via direct integration with the most common higher level tools used by ML engineers.
Given the availability bottleneck for Blackwell, there is a growing market for alternatives.
And, again, half the price of Hopper chips, for better performance, for a product that's already on the market, is a pretty attractive proposition that will certainly see more orgs coming through the door. Might need to put one of these $14k 128GB intel hopper-peers in my workstation...
CUDA is definitely the biggest hurdle as you pointed out though. NVidia's bet there paid off big time.
The 1st gen Intel Arc cards had a bit of a rough launch and still have issues with things like Unreal Engine games, but something like A580 for 150-175 USD and A750 or A770 for 175-250 USD, or even less, feels like good value! While Nvidia is doing whatever they want price wise, the amount of money in my bank account seems to like Intel's line of products more (currently using an A580 which I got for around 200 USD near launch to replace my old RX 580, getting a 3060 would have been and still is like 300 USD where I'm from). While XeSS is pretty good, since they don't have CUDA (it's just entrenched at this point) and framegen, they might as well compete on the price.
Which makes their launch of the Core Ultra CPUs that much more puzzling. They easily blow my current Zen 2 based Ryzen CPU out of the water... but I'll probably go for a Ryzen 7 5700X as my next CPU, because at least there the pricing makes sense. Intel's new CPUs being more or less a sidegrade/slight downgrade for gaming (even if improvement for productivity in some cases) wouldn't have been such a big deal if the pricing was actually good, instead of asking for 300 - 600 USD in addition to needing a new motherboard.
Money is fungible though. If I gave you $1000 and stipulated that you can only spend it on your mortgage, and you do indeed do that, you still have an extra $1000 you normally wouldn't have at the end of the month. Now let's say you decide to buy a TV with your extra $1000. Did the $1000 I give you go towards paying your mortgage or buying a TV, which is something I explicitly said you couldn't do?
Now scale that scenario up to Intel's case. Methinks they'll build a fab AND do a stock buyback. Under normal circumstances, they'd have to pick one of those things to do.
the 8bn is more of a pointy stick at this stage to try and get your 56 year old failing chipmaker to do something besides profiteering (buybacks layoffs and mergers.) Its a wasted effort. short of nationalizing Intel, the US is going to lose a crown jewel of its technological advantage solely due to mismanagement.
The only players in the space globally are TSMC, Intel, and Samsung.
The US has many fabs that are on par with Chinese fabs.
Had they been allowed to buy from ASML and other leading manufacturers of manufacturing equipment, they would have had better chances to compete with TSMC than Intel.
Even with the current sanctions, now when Gelsinger has been forcefully retired, which raises serious doubts about the viability of the Intel 18A CMOS process, which was the bet of Gelsinger for restoring the competitiveness of USA, this increases the chances for China to reach a point when they will have better semiconductor fabs than USA, which would not be good.
The hope that nobody will be able to make photolithography equipment competitive with ASML is likely to be an illusion identical with the belief of Intel from 10 years ago that nobody can make CPUs competitive with theirs.
Anyway, no one is even close to doing what ASML is doing.
Edit: From a rapid Google search seems line under 90nm there is only ASML. Canon and Nikon compete on older nodes.
Semiconductors Engineering is modern equivalent of Rocket Science
They say they are on track to delivery 18A - leading node in next 7 months
The forced retirement can be explained only in two ways, either the 18A is projected to be a failure and USA has lost any chances to become competitive with TSMC, or the retirement has been imposed by political forces that are completely incompetent and which do not understand anything about the semiconductor industry, and also in this case it seems likely that USA will fail to become great again in semiconductor IC manufacturing.
People aint gonna give you 2 3 quarters to enjoy stuff
Once 18A is delivered the 5N4Y is completed