TSMC CEO says it will be "a long time" before they can meet AI chip demand, with price hikes likely

midian182

Posts: 11,747   +178
Staff member
What just happened? Is the AI crisis showing any signs of abating? Not according to C.C. Wei, the CEO of TSMC. Speaking at the company's annual shareholders' meeting, Wei said the world's largest contract chipmaker won't be able to meet customer demand for years. He also revealed that TSMC would "like" to raise prices for its customers.

Wei said demand for AI chips remains extremely strong and customers are still upbeat about the industry's outlook. "It will be a long time before we can meet customer demand," he said.

That's not exactly a surprising revelation, of course. TSMC sits at the center of the AI hardware boom, producing advanced chips for many of the companies spending billions of dollars on accelerators, servers, and data center infrastructure.

Nvidia, Apple, AMD, Broadcom, and others depend heavily on the Taiwanese firm, which means every surge in demand eventually lands on TSMC's doorstep.

Wei said the company is working hard to avoid becoming a bottleneck in the global semiconductor supply chain, but the problem is no longer just about wafer capacity. According to reports, the squeeze now extends across equipment suppliers, upstream vendors, power, chipmaking capacity, and advanced packaging. So even if TSMC wants to build more, the rest of the industry has to keep up as well.

The company is certainly spending money like it expects AI demand to last. TSMC's US investment plans now total $165 billion, covering new Arizona fabs, advanced packaging facilities, and an R&D center. However, Wei warned that it would take a very long time to meet US customers' needs using American production, and the company's earlier goal of having 30% of its 2nm-and-below capacity in the US is looking difficult due to environmental permit delays and labor shortages.

The pricing comment is likely to attract almost as much attention as the shortage warning. Wei said TSMC would "like" to raise prices, citing rising component costs, but added that TSMC would not introduce hikes as abrupt as some memory makers.

The wider market is already feeling the consequences of AI's endless appetite for hardware. DRAM, NAND, and HBM shortages have pushed up prices and made everything from graphics cards to laptops more expensive.

The obvious caveat is that all of this assumes the AI spending boom continues. If this really is a bubble that's on track to burst, demand could crater at an unprecedented speed. Until that happens, Wei believes the AI chip crunch still has a long way to run.

Permalink to story:

 
They've been having multiple price hikes a year for a few years now. AI demand has been slowing
 
The implication is that "We" will take our time to boost production because you never know, and "You" will face a price increase because we want it without being forced to.
 
It maybe a bubble but there still is more demand than capacity. It is common to see Anthropic's servers overloaded. And Microsoft had to disable new user registrations to Github Copilot.
 
The implication is that "We" will take our time to boost production because you never know, and "You" will face a price increase because we want it without being forced to.
That machines that they use come from ASML and cost half a billion each. Even if they could "just buy more", their capacity is limited to un a couple hundred EUV machines a year. Expanding capacity on these new nodes isn't as simple as just buying more EUV machines. Further, every fab in the world is trying to get their hands on EUV from ASML. Orders are placed years in advance
 
That machines that they use come from ASML and cost half a billion each. Even if they could "just buy more", their capacity is limited to un a couple hundred EUV machines a year. Expanding capacity on these new nodes isn't as simple as just buying more EUV machines. Further, every fab in the world is trying to get their hands on EUV from ASML. Orders are placed years in advance
Which is why we need China to finish reverse engineering the tech and start cranking machines out.
 
We've been hearing about the supposed AI bubble for years now. I'm no longer convinced it's even real at this point. Even if AI fails (it won't), some other inane task will come along to demand enormous compute resources all the same. Before AI it was Crypto.
 
There is one fly in the ointment that people are overlooking. CHINA. Everyone knows they have been trying to "retake" Taiwan and bring them into the fold, like they did with Hong Kong after the UK left in the 90's. Right now, China knows it would not be a good idea, given the alliance Taiwan has with Japan, the USA and other countries, but if more problems continue in the middle east, distractions in other areas of the world might make it a little easier for China to cut off Taiwan from the rest of the world. Then they would probably attempt to do what they said they would do in Hong Kong. We'll just provide "leadership" and "security" but will have a hands off approach to Taiwan. (Ask Hong Kong how well THAT worked).
Taiwan can to a degree defend itself, but not with the firepower that China has.
China knows that if they go in guns a blazing, they will destroy a lot of the "chip" making abilities. Plus, it has been mentioned that Taiwan has a doomsday button to either destroy or disable a lot of the engineering, production of chips. It's a good thing they are starting to branch out with fabs outside of Taiwan. Non pun intended, but putting all your chips in one basket considering how close you are to China isn't a good thing.
 
Which is as good as openly admitting that the DRAM manufacturing "price-fixing" cartel is up-and-running once more....
Not even remotely correct. The laws of economics always operate, no matter how much anyone whines and stamps their feet in frustration.

Which is why we need China to finish reverse engineering the tech and start cranking machines out.
It's truly astonishing how certain people are willing to give a dictatorial regime control over the world economy, simply to save a few dollars on their next video-gaming device.
 
We've been hearing about the supposed AI bubble for years now. I'm no longer convinced it's even real at this point. Even if AI fails (it won't), some other inane task will come along to demand enormous compute resources all the same. Before AI it was Crypto.

You can't make the comparison between AI and Crypto. Most people never used Crypto for anything, but they use AI LLMs everyday (ChatGPT, Claude and Gemini mostly) to create and research things for any type of office work. It's so mainstream that its absurd to compare it to crypto.
 
Last edited:
You can't make the comparison between AI and Crypto. Most people never used Crypto for anything, but they use AI LLMs everyday (ChatGPT, Claude and Gemini mostly) to create and research things for any type of office work. It's so mainstream that its absurd to compare it to crypto.


Which is exactly my point.
 
Please, please burst like a failing earthen dam and drown those whom have pillaged and taken advantage of the consumer!!
 
The bubble copium, still 4 years later, is just hilarious to see. The reality is, AI is becoming more popular than personal computing itself, indispensable in the average office now, especially developer's. The anti AI crowd has provided plenty of material for the future conspiracy theory studies and has documented itself in the irreversable interweb.

The only thing that will ever reduce hardware demand is a breakthrough in optimisation algorithms.

 
The bubble copium, still 4 years later, is just hilarious to see. The reality is, AI is becoming more popular than personal computing itself, indispensable in the average office now, especially developer's. The anti AI crowd has provided plenty of material for the future conspiracy theory studies and has documented itself in the irreversable interweb.

The only thing that will ever reduce hardware demand is a breakthrough in optimisation algorithms.

Exactly - during the last 4 moths we've been automating so many tasks it is unreal. I own a small Digital Agency and the amount of worked pumped out through AI assistance is amazing. Quality has gone through the roof too - we don't work less, we wokr more because we are excited with what we can do anymore. The problem with AI is how people use it. A smart hard wroking person + AI = great results, A dumb lazy person + AI = shitty results.
 
Back