Intel should have focused on AI rather than chip making, TSMC founder says

[BruceA]

It is in TSMC’s best interest to see Intel focus on AI and sell of fabrication, duh!!!!!

Then there is one, I don’t count Samsung as a legitimate threat or competitor. Intel could still be a technology competitor.

If Intel discards the fab and gives up and cedes it to TSMC. Then there is one and they got Apple, Qualcomm, MediaTek for Mobile and Nvidia, AMD, Intel, Amazon, Google, Microsoft in their hand, of course Intel should discard chip making and TSMC captures another 40B or revenue in a few years.

Trying to compete with Nvidia on AI front side assault is like Power, SPARC, PaRISC and all the other dreamers trying to compete against x86…

 

[Xebec]

"The companies selling the pick axes will also get rich. Intel can play both sides of this equation. In fact, Intel is the only one who can play both sides."

Most of Intel's potential leading-edge node customers are also its competitors. Does it make sense for Intel to bet everything on offering cheaper and better foundry services to them? This could enable Intel's product competitors to undermine its product division - the division that generates the majority of Intel revenue and profit (if any). Does this strategy make sense?

I think of it this way --

Intel is competing with these companies either way. If they sell an Intel AI chip produced at an Intel Fab, they get 100% of the value of a product.

If they lose a sale to a competitor who fabs at TSMC, they get 0% of the value of the product.

If they help Nvidia produce an AI chip at Intel foundry, then Intel is still getting some value (call it 20%) of the product.

Now there is a balance and reputation game to play.. they also need ironclad internal (transparent) processes for deconflicting priorities. But if they can execute at both fab at products - they can get a larger portion of the pie than if they're just doing one or the other.

 

[Daniel Nenni]

"Intel needs to fill fabs".

Spending billions of dollars to build multiple new fabs without knowing who the customers will be to fill the capacity - is it naive or reckless?

Can Intel focus on a smaller number of fabs and locations?

"The companies selling the pick axes will also get rich. Intel can play both sides of this equation. In fact, Intel is the only one who can play both sides."

Most of Intel's potential leading-edge node customers are also its competitors. Does it make sense for Intel to bet everything on offering cheaper and better foundry services to them? This could enable Intel's product competitors to undermine its product division - the division that generates the majority of Intel revenue and profit (if any). Does this strategy make sense?

Reckless. Fill the fabs you have THEN build more. Intel presented some good papers at IEDM. We will publish stuff next week.

 

[tedc]

The (business) problem here is that when you're down (i.e., uncompetitive products, loss of leading edge tech), the prudent decision isn't to go broader (foundry) and directly take on wildly successful incumbents (Nvidia), because you've just made the hill you need to climb that much steeper.

Intel is a unique company, and their path to success shouldn't just be emulating existing players. Pat's folly was believing Intel could simultaneously compete against TSMC, Nvidia, AMD, etc., companies with drastically different business models.

 

[Artificer60]

Hold on - this doesn't seem self-consistent. If Intel competition forces TSMC to cut its prices, then that surely is a large dent in TSMC's business.

But that would require Intel having tech leadership AND cost leadership AND the appetite to initiate a price war against TSMC. If the first two were met, why wouldn't Intel price above TSMC (why price a superior product lower) ? After all, Intel has many of the same fixed costs to recover off far smaller volumes than TSMC.

Apologies, seems I didn't express myself clearly. I'm don't think Intel is likely to force cost cuts, but I do think their presence could keep TSMC from raising prices as much as they might otherwise. Each technode costs more than the last, so everyone is going to be faced with ever increasing prices for the next technode. I just think Intel's presence gives customer's a stick to keep TSMC honest. I do not see Intel driving TSMC's prices down directly. I don't see customers flocking to Intel anytime soon, but the threat will have some (small?) impact on TSMC's pricing.

If Intel achieves tech leadership, I would expect them to demand a premium for their services if only to offset the inherent cost disadvantage of manufacturing in the US. So I don't think they are looking to gain cost leadership. I think they will set prices comparable or maybe slightly below TSMC to attract customers while TSMC's supperior customer service and ecosystem will ensure they hold the bulk of the market.

This was always the argument for why it was beneficial to have a competitive AMD, to keep Intel driving technology and keep prices down. In a two competitor environment (TSMC and Intel) I would expect the same dynamic and argument to apply.

 

[Artificer60]

The (business) problem here is that when you're down (i.e., uncompetitive products, loss of leading edge tech), the prudent decision isn't to go broader (foundry) and directly take on wildly successful incumbents (Nvidia), because you've just made the hill you need to climb that much steeper.

Intel is a unique company, and their path to success shouldn't just be emulating existing players. Pat's folly was believing Intel could simultaneously compete against TSMC, Nvidia, AMD, etc., companies with drastically different business models.

To my mind he was just trying to do too much too fast. When you are practically printing your own money, you can do whatever you want. Today's Intel has to lose that mindset and quickly.

 

[tooLongInEDA]

The (business) problem here is that when you're down (i.e., uncompetitive products, loss of leading edge tech), the prudent decision isn't to go broader (foundry) and directly take on wildly successful incumbents (Nvidia), because you've just made the hill you need to climb that much steeper.

Intel is a unique company, and their path to success shouldn't just be emulating existing players. Pat's folly was believing Intel could simultaneously compete against TSMC, Nvidia, AMD, etc., companies with drastically different business models.

This reminds me of the German strategy in WWI in choosing to fight a war on two fronts (and ending up in the same situation in WWII). Except in Intel's case, there are probably more than two fronts now.

 

[Barnsley]

Quantum Computing seems to be the next big thing. How is Intel doing on Quantum? TSMC has not mentioned it yet but I know they are working with Google and others on it. Skate to where the puck is going, not to where it has already been.

What tech node was this manufactured at?

 

[IanD]

I agree that TSMC doesn't have to worry about Intel stealing a large amount of their volume. But I do think that Intel poses a potential threat.

Intel has decades of proof points that they can be a technology leader. If they stay on track with 18A and 14A they may be the tech leader again. Intel seems to believe this currently.

Even with tech leadership I don't see Intel putting a large dent in TSMCs business (and they don't need to). What I do see as a possibility is Intel taking the tech lead and giving TSMC's customers a large bargaining chip that impacts TSMC's current de facto monopoly position. Losing the tech lead would limit TSMC's ability to freely set prices wherever they want them. Even if the customers don't move, TSMC couldn't ignore the fact they don't have tech leadership and would have to set price accordingly.

The problem for Intel is that nowadays "tech leadership" doesn't mean what you seem to think it does... ;-)

It doesn't mean having the sexiest most advanced technology on a Powerpoint slide, or even in a pre-production fab -- it means being able to mass-produce it in megafabs with good yield and at a competitive price, with good PDK and customer support, and preferably with a strong IP ecosystem.

For example, IBM have always been bullish about their fabulous advanced technology, including SOI, FinFET, GAA... -- but what they come up with in the labs has always been terrible to manufacture, as the companies who've tried to do this will know all too well.

Intel have historically aimed high with their technology and often hit (copper, FinFET) but also sometimes disastrously missed (10nm, EUV) -- and even when they hit, since their only real customer was the inhouse CPU group they didn't need to bother making the technology easy to use (terrible PDKs and support) because the design group just had to suck it up, and they didn't need to get really high yields or low wafer cost because their CPU product margins were so high -- I know for a fact that their yielded die cost has been up to 2x compared to TSMC in the past. And their IP ecosystem for customers to use is nowhere, unless you happen to be building CPUs like Intel... :-(

Changing this "raw technology is king, the customer will have to use it" mindset is Intel's biggest problem, just like it was for IBM. And if they don't -- which needs a total rethink in approach -- they have no chance of competing with TSMC, whose approach is "the customer is king, technology serves the customer"... ;-)