Nvidia CEO: US chip independence may take 20 years to achieve

US may take twice as long as Biden expects to build its own chip supply chain.

See full article...

 

[quamquam quid loquor]

That risks the US falling behind, Huang suggested, while China potentially finds ways around export controls to acquire leading technology anyway that can "inspire" its own chip advancements.

That's a diplomatic way to put it.

 

[ruet]

Ok, get on it.

 

[killerhurtalot]

It's gonna take longer than that when US workers aren't willing to work the long hours and low wages in the said fabs.... just look at how long TSMC and Intel has delayed their new fabs here and tried to get workers from Taiwan and Israel to come over for some of the said positions...

 

[quamquam quid loquor]

I think the tech trade war is clearly effective, unlike more commercially oriented trade wars (tariffs).

The second half of this is we desperately need immigration reform to give all the STEM MAs and PhDs a fast track visa and immediate work authorization. I see ivy league graduates denied for a visa all the time and forced back to economic competitor nations.

 

[quamquam quid loquor]

It's gonna take longer than that when US workers aren't willing to work the long hours and low wages in the said fabs.... just look at how long TSMC and Intel has delayed their new fabs here and tried to get workers from Taiwan and Israel to come over for some of the said positions...

The best US workers are the absolute best in the world, but you get what you pay for. The dollar definitely goes further in Taiwan and the rest of Asia. That's why you see far superior workers in Asia for the same pay.

 

[SpeakThunder]

Curious why it's so hard to set up chip manufacturing as compared to other technologies. I know the machinery is expensive and complex, but is that the only reason? Is it difficult for Intel to just duplicate it's manufacturing process here in the US? Or do they not have access to the IP to do that?

 

[Andrewcw]

So Nvidia is thinking this in business terms. Government thinks this in government terms.
In Government terms they want to be able to produce anything. It doesn't mean all production has to be domestic. But if lets say every TSMC foundry suddenly disappeared off the planet. That is what the Government is hedging against. Not supply chain independence in Business terms where it becomes profitable to make it domestically.

 

[quamquam quid loquor]

Curious why it's so hard to set up chip manufacturing as compared to other technologies. I know the machinery is expensive and complex, but is that the only reason? Is it difficult for Intel to just duplicate it's manufacturing process here in the US? Or do they not have access to the IP to do that?

It's always people. When we moved all these industries to Asia for low-cost manufacturing, we fired all the people in the US and now they are old and their knowledge is stale. Look at how difficult it was for Tesla/SpaceX to get to scale and imagine that x100 mini industries.

Even if it's all US universities pushing the tech, the manufacturing experts are in Asia, not here. The most effective way to do this is "Paperclip" style, pour money into relocation grants for expert engineers and their families with fast-track citizenship.

 

[k1n9bur93r]

Curious why it's so hard to set up chip manufacturing as compared to other technologies. I know the machinery is expensive and complex, but is that the only reason? Is it difficult for Intel to just duplicate it's manufacturing process here in the US? Or do they not have access to the IP to do that?

Labor and supply chains! Need those in place to feed a Fab, and said Fabs need high up times to pay for themselves. We'll get there, but it's a lot of set up.

 

[xaxxon]

Let's be clear about this -- Nvidia LOVES the status quo and wants to dissuade people from disrupting it.

They have close enough to a monopoly on the hardware that they don't want anyone thinking they can do things faster to compete.

 

[xoe]

Still plenty of time to spare before the "great rejuvenation of the Chinese nation" in/by 2049/2050.

 

[Mad Klingon]

Supply chains are a big issue. You need machines and people to build the machines needed for the fab. You need more factories to supply the raw materials the fab turns into chips. You need mines and processing to source the raw materials for all of these.
Education is also a problem. A report from Sept 2020 stated "According to the U.S. Department of Education, 54% of U.S. adults 16-74 years old - about 130 million people - lack proficiency in literacy, reading below the equivalent of a sixth-grade level." Source - https://www.forbes.com/sites/michae...e-economy-22-trillion-a-year/?sh=2729cbd94c90
This was before we knew just how much bad an impact all the covid school closings and remote learning had on kid's education. Numbers are probably worse today.
When over half your potential workforce can't read the "How to use this machine" operator's manual, makes hiring rather difficult.

 

[Fatesrider]

Let's be clear about this -- Nvidia LOVES the status quo and wants to dissuade people from disrupting it.

They have close enough to a monopoly on the hardware that they don't want anyone thinking they can do things faster to compete.

I'd go with seven years rather than 20, assuming the U.S. puts in the effort and money to do it. It would need to be a coordinated effort with universities and such to produce people who have the skills to do what needs to be done, and would need to be adequately financed (since a lot of these mandates aren't actually funded - just orders from on high).

This, of course, assumes that there's the political will to do this. Potentially, I can see it happening, unlike a lot of other things, since it caters to the racists who want to isolate America and to the folks who care about domestic security rather than those who just pay lip service to it.

It was never a good idea to overseas critical industries in the first place, especially when those in power knew China was insincere and not playing by the rules from the get-go. Reversing that QUICKLY would take effort. But it COULD be done much faster than Huang intimates if the political will can be sustained.

 

[quamquam quid loquor]

I'd go with seven years rather than 20, assuming the U.S. puts in the effort and money to do it. It would need to be a coordinated effort with universities and such to produce people who have the skills to do what needs to be done, and would need to be adequately financed (since a lot of these mandates aren't actually funded - just orders from on high).

This, of course, assumes that there's the political will to do this. Potentially, I can see it happening, unlike a lot of other things, since it caters to the racists who want to isolate America and to the folks who care about domestic security rather than those who just pay lip service to it.

It was never a good idea to overseas critical industries in the first place, especially when those in power knew China was insincere and not playing by the rules from the get-go. Reversing that QUICKLY would take effort. But it COULD be done much faster than Huang intimates if the political will can be sustained.

Racists are a huge reason for why we are being challenged and behind in some areas. I always bring up Qian Xuesen, cofounder of JPL, who racists denied citizenship to and kicked out of this country. Qian Xuesen of course became the father of the Chinese space program, where he has near deity status.

We already suck up all the smart people around the world to our world-class universities. Just give them guaranteed greencards and a fast-tracked citizenship!

 

[SirOmega]

It's always people. When we moved all these industries to Asia for low-cost manufacturing, we fired all the people in the US and now they are old and their knowledge is stale. Look at how difficult it was for Tesla/SpaceX to get to scale and imagine that x100 mini industries.

Even if it's all US universities pushing the tech, the manufacturing experts are in Asia, not here. The most effective way to do this is "Paperclip" style, pour money into relocation grants for expert engineers and their families with fast-track citizenship.

Intel was the worldwide leader for a long time. And the more I reflect on my management job, it wouldn't surprise me that the faltering of the last 15 years was due to the retirement of the technical people in the organization. Not just the folks at the top (Moore, Grove), but engineers of all kinds that grew up with the company from the 80s and 90s to early 2000s. Who had decades of knowledge about design, integration, fabrication, etc. All left and then there wasn't sufficient knowledge transfer to the next generation. So they spent ~10 years (?) figuring out 10nm.

But I'm part of the problem. Here I sit with my CompE degree... not doing CompE. I could have got a CS degree and not much would be different (I thoroughly enjoyed my hardware engineering classes). So there needs to be a better pipeline of CompE students going into the field to really bring it back to the US.

Another poster said it up - the chip production supply chain has never been 100% US. Intel would send the chips overseas for packaging (they had two main chip packaging plants.. Malaysia and one I cant remember). If they want to de-China the process, then the focus is on fabrication technology and taking that lead back. And dare I say, open that leading process up as a foundry service. Apple would probably love to have a choice other than TSMC (though I don't expect Intel to take them as a customer).

 

[H2O Rip]

I suspect this is meant to throw some cold water on both Amazon's and Msft's announcements in AI chips recently, at least in a way that might give Nvidia's investors jitters. There is a grain of truth in there, but as others have noted it's mostly because people doubt the political will being ironclad around it.

 

[Eldorito]

It's gonna take longer than that when US workers aren't willing to work the long hours and low wages in the said fabs.... just look at how long TSMC and Intel has delayed their new fabs here and tried to get workers from Taiwan and Israel to come over for some of the said positions...

That's probably a lot less to do with wages and hours and a lot more to do with experience. You can't teach people experience and there's not a whole lot of cutting edge fabs to hire people from in the US.

Lower wages is why they went to Asia, but you can't just haul someone to the US and expect them to continue to take a low wage. Plus Taiwan/Israel gave plenty of incentives like the US is doing now to actually build the fabs. When there are jobs that builds the education system to support and the training pipeline.

 

[stormcrash]

Intel was the worldwide leader for a long time. And the more I reflect on my management job, it wouldn't surprise me that the faltering of the last 15 years was due to the retirement of the technical people in the organization. Not just the folks at the top (Moore, Grove), but engineers of all kinds that grew up with the company from the 80s and 90s to early 2000s. Who had decades of knowledge about design, integration, fabrication, etc. All left and then there wasn't sufficient knowledge transfer to the next generation. So they spent ~10 years (?) figuring out 10nm.

But I'm part of the problem. Here I sit with my CompE degree... not doing CompE. I could have got a CS degree and not much would be different (I thoroughly enjoyed my hardware engineering classes). So there needs to be a better pipeline of CompE students going into the field to really bring it back to the US.

Another poster said it up - the chip production supply chain has never been 100% US. Intel would send the chips overseas for packaging (they had two main chip packaging plants.. Malaysia and one I cant remember). If they want to de-China the process, then the focus is on fabrication technology and taking that lead back. And dare I say, open that leading process up as a foundry service. Apple would probably love to have a choice other than TSMC (though I don't expect Intel to take them as a customer).

And luckily for Intel, just as many were ready to write them off as dead on the leading edge fab front, they seem to be getting their act back together just as TSMC is running into some headaches. Living at the bleeding edge you're always only one misstep away from potential disaster in terms of keeping the lead (or keeping up with the leader) and fortunes can easily reverse so if you can keep up and keep the investment going you opponent's rainy day may become your golden ticket.

Also I seem to remember something about Intel having a packaging plant in south america somewhere but can't remember where and if it still operates or was only back in the early 2000s

 

[Irritable Immunologist]

Intel has numerous (in fact overwhelmingly) US fabs, and until about ten years ago was perhaps the world leader in semiconductor fabrication. Packaging is a bit different, but not unsolvable.

Consequently Jensen's statements here appear quite self-serving.
Quite out of character, I know.

Targeted incentivization of domestic Materials Sciences/Engineering programs, selective Visa programs, and disincentivizing outsourced production would appear to go some way toward altering the status quo.

 

[lurknomore]

On the science pipeline front, as long as hundreds of thousands of young Americans continue to wonder whether the high cost of college is worth it, the US will keep losing its lead.
I've worked in high tech for a long time, and the vast majority of my coworkers were not born in this country. Many of them eventually go back "home" and bring that knowledge with them. Or we hire Indians to work cheap and we send them our tech, and they bolt after a year or two.

Wanna keep winning the tech race ? Free college is a great place to start.

 

[Rival Vector Plot]

Don’t give up the cause or momentum. Either way, in the end, we’re gonna got need semiconductor independence. .

 

[Rival Vector Plot]

On the science pipeline front, as long as hundreds of thousands of young Americans continue to wonder whether the high cost of college is worth it, the US will keep losing its lead.
I've worked in high tech for a long time, and the vast majority of my coworkers were not born in this country. Many of them eventually go back "home" and bring that knowledge with them. Or we hire Indians to work cheap and we send them our tech, and they bolt after a year or two.

Wanna keep winning the tech race ? Free college is a great place to start.

A nation investing in the education of its people is an investment in the future.

 

[Emmanuel Deloget]

The best US workers are the absolute best in the world, ...

No offense intended, but a citation (wikipedia style) would'nt hurt here.

 

[original_mds]

On the science pipeline front, as long as hundreds of thousands of young Americans continue to wonder whether the high cost of college is worth it, the US will keep losing its lead.
I've worked in high tech for a long time, and the vast majority of my coworkers were not born in this country. Many of them eventually go back "home" and bring that knowledge with them. Or we hire Indians to work cheap and we send them our tech, and they bolt after a year or two.

Wanna keep winning the tech race ? Free college is a great place to start.

Some of these fields already are full of dedicated scholarships and free programs for qualified students. Free college is much more of a boon to liberal arts degrees, which won't help much with this problem.

 

[Secondfloor]

I believe we’ll see AI models produce unanticipated, unconventional but ultimately more productive advancements that will challenge the long range of this estimate.

Newsflash: Nvidia already sells products for AI assisted chip design. Stop thinking about them as a gaming company. They’ve been a software company for over a decade. Just go look at their job postings.

 

[piotrek54321]

Let's be clear about this -- Nvidia LOVES the status quo and wants to dissuade people from disrupting it.

They have close enough to a monopoly on the hardware that they don't want anyone thinking they can do things faster to compete.

Exactly. Now, 20 years is a lot of time to establish a competitor that would actually force Nvidia to get on with it and innovate or lose market share.

 

[quamquam quid loquor]

No offense intended, but a citation (wikipedia style) would'nt hurt here.

I remember reading a paper on innovation by country, which I’ll try to find again.

One link is tech companies - proxy for innovation and thus the best workers, shows US companies in the lead. Another proxy is tech company revenue - shows US companies in the lead.

https://en.m.wikipedia.org/wiki/Technology_company
https://en.m.wikipedia.org/wiki/List_of_largest_technology_companies_by_revenue
If you look at the most advanced technologies as a subset of the tech sector, you see it’s driven by many US companies and US employees. The ethnicities are like the United Nations, but for the most part the employees work in the US.

 

[Sumwun]

Intel has numerous (in fact overwhelmingly) US fabs, and until about ten years ago was perhaps the world leader in semiconductor fabrication. Packaging is a bit different, but not unsolvable.

Intel only has one fab in the USA able to produce sub-10nm chips (Fab 27 in Ohio). Two more plants are under construction and due to come online in 2024 and 2025. Intel needs these sites for their own chips. Large, complex chips need 5nm (or smaller) capability and they take years to build and a lot of highly trained people to operate them. Samsung and TSMC are also building fabs in the USA but they are both running into problems mostly in finding suitable contractors and staff, which is why they are importing expertise from Asia.

We are also dependent on ASML, at present the only company in the world with the ability to create EUV photolithographic systems that can handle sub-7nm chips. ASML is building a factory in the USA but currently these systems are manufactured in the Netherlands.

Actually Huang said it would take one to two decades before the USA achieves semiconductor independence, not 20 years. Either way, the USA is way, way behind.

Edit: Syntax is hard.

 

[lurknomore]

I remember reading a paper on innovation by country, which I’ll try to find again.

One link is tech companies - proxy for innovation and thus the best workers, shows US companies in the lead. Another proxy is tech company revenue - shows US companies in the lead.

https://en.m.wikipedia.org/wiki/Technology_company
https://en.m.wikipedia.org/wiki/List_of_largest_technology_companies_by_revenue
If you look at the most advanced technologies as a subset of the tech sector, you see it’s driven by many US companies and US employees. The ethnicities are like the United Nations, but for the most part the employees work in the US.

That's thoroughly biased by the availability of VC capital and regulatory environment, and not a true reflection of worker quality.

 

[original_mds]

That's thoroughly biased by the availability of VC capital and regulatory environment, and not a true reflection of worker quality.

And ITAR. Much easier to work with the existing innovation tech base if you don't have to deal with export controls.

 

[ardentsonata]

Supply chains are a big issue. You need machines and people to build the machines needed for the fab. You need more factories to supply the raw materials the fab turns into chips. You need mines and processing to source the raw materials for all of these.
Education is also a problem. A report from Sept 2020 stated "According to the U.S. Department of Education, 54% of U.S. adults 16-74 years old - about 130 million people - lack proficiency in literacy, reading below the equivalent of a sixth-grade level." Source - https://www.forbes.com/sites/michae...e-economy-22-trillion-a-year/?sh=2729cbd94c90
This was before we knew just how much bad an impact all the covid school closings and remote learning had on kid's education. Numbers are probably worse today.
When over half your potential workforce can't read the "How to use this machine" operator's manual, makes hiring rather difficult.

It's much worse than you make it seem because building the machines needed for the fabs requires such highly specialized knowledge literally only one company in the world, ASML, does it.

So even if you wanted to reshore some of that here, that ship has long since sailed on building that expertise in this country.

 

[earlyberd]

Huang also profits heavily from the world's dependence on Nvidia for AI chips, so it's not in his best interests to imply that U.S. manufacturing can or should be able to keep up. He doesn't actually want any real competition from Apple, Qualcomm, AMD, or Intel.

But if the Chinese can figure out how to strip the GPUs out of GeForce cards and shoehorn them into AI compute farms of their own, bypassing trade restrictions, then what the hell stops anyone else from reverse-engineering Nvidia's chips and making a cheap alternative? You can ask LLMs and machine vision models to help you design newer, cheaper chips in a fraction of the time now, just by giving them good quality photos and technical design documents from what already exists out there.

The barrier to entry has just dropped by a good 6 feet across the board, and this guy who's supposedly been on top of AI development for the past decade can't even see how little he actually has to offer, now that all the world's researchers can get free knowledge and leverage AI to find ways to undercut him in ways never before possible.

Stop deriding U.S. workers like they're stupid. Pay more in taxes so we can pay for education and jobs programs, or we'll find ways to replace all the CEOs, CFOs, and HR with robots that actually care about sustaining society for the greater good.

 

[jtwrenn]

Just more reason to push it as hard as possible. Everything will take longer than we expect at this scale. We needed to start this 20 years ago.

 

[Cuy]

I think he seriously underestimates what motivated people can accomplish and how quickly an economy the size of the current global economy can adapt. While China specifically is not as relevant to semiconductor supply chains as other electronics, remember China's accession to the WTO was only back in 2001 and by the 2010s pretty much everything was made in China, literally the entirety of China supply chains and China's role as the world's factory only took 20 years. I'll bet this could happen in five years if there were a sufficiently motivational event to make it happen, although that kind of motivation would likely only come from something that causes a hell of a lot of short term pain.

 

[Reaperman2]

I think he seriously underestimates what motivated people can accomplish and how quickly an economy the size of the current global economy can adapt.

I assume the "motivated people" you're talking about are greased politicians. Because the only way fabs are being built today in the US is with significant government incentives that result from significant kickbacks -- sorry, "donations" -- to the pols making it happen.