Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/tsmc-to-make-advanced-2-nm-chips-in-us-sooner-to-meet-ai-demand.23842/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2030770
            [XFI] => 1060170
        )

    [wordpress] => /var/www/html
)

TSMC to make advanced 2-nm chips in US sooner to meet AI demand

Daniel Nenni

Admin
Staff member
World's top chipmaker plans to secure land in Arizona beyond $165bn investment

20251016 TSMC

A Taiwan Semiconductor Manufacturing Co. building in Phoenix, Arizona, is seen on Oct. 3. (Photo by Rebecca Noble)

TAIPEI -- Taiwan Semiconductor Manufacturing Co. on Thursday said it plans to speed up its time frame for producing advanced 2-nanometer chips in the U.S., while also signaling further expansion beyond the $165 billion it has already committed to investing in America, according to company Chairman and CEO C.C. Wei.

"We are preparing to upgrade our technologies faster to N2 [2-nm] and more advanced processes in Arizona, given the strong AI-related demand from our customers," Wei told investors and reporters in an earnings conference. Previously, TSMC had said its third plant in the U.S. state will produce 2-nm chips before the end of the decade.

Wei also said TSMC is close to securing a second large plot of land close to its Arizona site to provide more flexibility in response to the strong, multiyear AI demand.

"Our plan will enable TSMC to scale up to an independent Gigafab cluster in Arizona to support the needs of our leading-edge customers in smartphone, AI and HPC [high-performance computing] applications," he said. Gigafab refers to chip facility clusters that can at least churn out more than 100,000 12-inch wafers a month.

Wei's comments came shortly after U.S. chipmaker Intel put its most cutting-edge 18A chip into mass production in Arizona.

Wei said the ramp-up schedule for the second plant in Japan and the first plant in Dresden, Germany, will be based on customer demand and market conditions.

TSMC reported a record profit for the July-September period as the AI computing wave continues to boost chip and data center spending.

Net profit for the quarter reached 452.3 billion New Taiwan Dollars ($15.12 billion), surging 39.1% from a year ago on revenue of NT$989.91 billion, a rise of 30% on the year and also a record high.

TSMC expected its revenue to be between $32.2 billion to $33.4 billion for the last quarter of 2025, growing 22% at the midpoint from a year ago, above industry estimates. The world's top chipmaker said it now expects about 35% growth for revenue for all 2025.

The contract chipmaker raised its 2025 capital expenditure plan to between $40 billion and $42 billion, compared with its previous guidance of $38 billion to $42 billion, due to structural AI-related demand remaining to be "very strong."

Wei downplayed the impact of export controls and Chinese restrictions on AI chips, saying that he has confidence in TSMC's customers. "If the China market is not available, I still think AI's growth will be very dramatic, and as I said, [they are] very positive in performance.

Shares in TSMC, the world's top chip manufacturer, have risen more than 50% this year, driven higher by investor optimism over demand for the company's products as key clients including Nvidia, AMD and Broadcom have signed multiyear partnerships with major AI service providers such as OpenAI and Oracle to expand AI data center capacity.

In addition to the AI boom, TSMC's record earnings have also been supported by new product launches from Apple, another major client, with its latest iPhones, iPads and MacBooks contributing to increased chip orders.

Applied Materials, the top U.S. chip equipment maker, said it does not see any slowdown in the AI boom and related demand, while top European chip tool maker ASML, another key TSMC supplier, on Wednesday said it does not expect its revenue for 2026 to be lower than 2025 despite a slowdown in its China business.

TSMC's revenue from North America continues to grow, hitting 76% in the July-September quarter, up from 71% from a year earlier, thanks to growing demand from all of the leading AI chip developers in the U.S. In contrast, revenue from China for the same period fell to 8%, down from 11% a year ago.

"Some negative impact from tariffs or a weakened economy has not materialized, suggesting TSMC's near-term performance remains well-supported and risk from external macroeconomic factors is manageable," said Arthur Lai, semiconductor analyst with Australian banking group Macquarie. TSMC's outlook is also being boosted as more AI chip clients migrate to even more advanced production nodes, which could help lift the company's average selling price, Lai added.

 
How is ASML going to meet this high demand with HN-NA EUV? ASML will ship a total 60-70 production level EXE:5200 systems in the next four years. TSMC will need more than that for full HNA-EUV implementation at a given process node? Intel and Samsung will also need them for logic and then comes the memory makers.
 
I wonder what "faster to N-2" means. I believe the original schedule was N-3 in 2028 and N-2 in 2030.

I know they said their customers want US manufactured AI chips, and I don't doubt that is true. But I also have to wonder if Intel launching 18A and proclaiming it as "the most advanced semiconductor process ever developed and manufactured in the United States" has something to do with it. Intel's claim is carefully nuanced, but true and I'm not sure that TSMC likes being shown as second best in any way, shape or form.
 
How is ASML going to meet this high demand with HN-NA EUV? ASML will ship a total 60-70 production level EXE:5200 systems in the next four years. TSMC will need more than that for full HNA-EUV implementation at a given process node? Intel and Samsung will also need them for logic and then comes the memory makers.
it's still too early for Hi-NA to contribute HVM (High volume manufacture). In Q3 earning call on Wednesday, ASML CEO just said won't expect Hi-NA to start HVM in customers until 2028-2029. Given this, significant shipment (two digits per year) will not start until end 2027 or even 2028... long enough for ASML supply chain to prepare.
 
How is ASML going to meet this high demand with HN-NA EUV? ASML will ship a total 60-70 production level EXE:5200 systems in the next four years. TSMC will need more than that for full HNA-EUV implementation at a given process node? Intel and Samsung will also need them for logic and then comes the memory makers.
Has any High-NA sales been recognized yet? Intel reported mirror replacement at SPIE. The trend of EUV recognized sales per quarter is so far declining this year.
 
Last edited:
I wonder what "faster to N-2" means. I believe the original schedule was N-3 in 2028 and N-2 in 2030.

I know they said their customers want US manufactured AI chips, and I don't doubt that is true. But I also have to wonder if Intel launching 18A and proclaiming it as "the most advanced semiconductor process ever developed and manufactured in the United States" has something to do with it. Intel's claim is carefully nuanced, but true and I'm not sure that TSMC likes being shown as second best in any way, shape or form.

Intel is not wrong and 18A will be in HVM before TSMC N2 and Samsung 2nm. TSMC N2 will of course have critical mass with 99% of the customers using it. Still, it is quite an achievement. And lets not forget BSPD. Intel has always been an innovative company. Hopefully Lip-Bu gets Intel back on track in all regards. So far so good in my opinion, he has exceeded my expectations. Thursday's call will be a big one!
 
$65B in Arizona thus far, only $100B more to go based on the latest press.
I know thats the plan so far. Have they already spent that? yes I know CHatGPT says so :).

first fab is running. Second fab has output planned in 2 years. third fab is being started. Tools are the most expensive part of a fab and the tools for 2 and 3 have not been purchased.
 
I know thats the plan so far. Have they already spent that? yes I know CHatGPT says so :).

first fab is running. Second fab has output planned in 2 years. third fab is being started. Tools are the most expensive part of a fab and the tools for 2 and 3 have not been purchased.

That is what TSMC said. What they have actually spent thus far I do not know. If they build the 6 giga fabs as planned it will be more than $165B in my opinion. TSMC is now buying more land in Phoenix, this was mentioned on the last call. I really am impressed that TSMC is investing so much in Arizona. I was half expecting Lip-Bu to sell the Intel Ohio land to TSMC. :ROFLMAO:
 
Morris Chang genius continues to pay off, with TSM being one of the world's greatest companies and a great investment. His legacy continues, which shows TSM is built for the short and long term. A true founder of the modern world economy.
 
Morris Chang genius continues to pay off, with TSM being one of the world's greatest companies and a great investment. His legacy continues, which shows TSM is built for the short and long term. A true founder of the modern world economy.

TSMC is having an amazing run. It is their connection to customers and the ecosystem that makes them unbeatable. I do not think TSMC would be such a dominant force if it were not for CC Wei. CC is a true competitor whereas Morris wanted to be everyone's friend.

It also helps that competition has tripped over themselves in trying to compete with TSMC.

I do see Intel Foundry as being a viable competitor in 5-10 years but TSMC has the HVM economies of scale factor that will be very hard to match.
 
Intel basically shot itself in the foot otherwise TSMC wouldn't have the tech crown and getting foundry customer would have been lot easier than it is now.

If we could turn back the clock and make Intel’s 10nm process deliver on time, assuming everything else remained the same, would Intel have great products, fast growing revenue, and strong profits today?

My guess is no. It might have only delayed the troubles Intel faces today by two or three years, but fundamentally, Intel was already on the wrong track.

For example, even while using what were once considered inferior manufacturing processes from TSMC or Samsung, companies like TSMC, Apple, Qualcomm, MediaTek, and Samsung made huge profits from smartphone related products for many years. Intel, meanwhile, had no competitive product in that space and still doesn’t today.

In 2007, global PC sales totaled around 360 million units, compared with 122 million smartphones.

Fast forward to 2024: global PC sales were about 255 million units, while smartphone sales reached 1.24 billion units.

The PC market that Intel relies on is either stagnating or shrinking over the years. Unfortunately, Intel didn’t miss the mobile revolution, it consciously walked away from it.

Another example is in AI processors. Nvidia, AMD, and Intel all rely on TSMC for manufacturing, yet Intel’s AI products still lag behind in competitiveness.

Intel’s long delay in 10nm manufacturing is an easy target to blame, but the more serious problem is that Intel failed to develop and deliver the right products to compete in the right markets.
 
Intel took a lot of effort kicking out Transmeta from the thin and light laptop market. But they failed at mobile.

Intel had the superior solution in the mobile market, DEC StrongARM aka Intel XScale, but they never knew what to do with it. It was used in the iPAQ and other such devices. Intel got rid of the division around the time the iPhone came out. They claimed it was low revenue. Hah. Thanks a lot Ottellini.

In retrospect it was obvious. Every time computing moves to a smaller form factor the smaller form factor eventually dominates the older one in units sold and market size. Like DEC failed to move from minicomputers to microcomputers properly, Intel failed at moving from microcomputers to mobile.
 
Last edited:
Intel had the superior solution in the mobile market, DEC StrongARM aka Intel XScale, but they never knew what to do with it. It was used in the iPAQ and other such devices. Intel got rid of the division around the time the iPhone came out. They claimed it was low revenue. Hah. Thanks a lot Ottellini.

"Intel didn’t miss the mobile revolution, it consciously walked away from it."
 
If we could turn back the clock and make Intel’s 10nm process deliver on time, assuming everything else remained the same, would Intel have great products, fast growing revenue, and strong profits today?
It would have cause they would not be jumping to TSMC and if they had the same lead IFS would be in a better position.
 
If we could turn back the clock and make Intel’s 10nm process deliver on time, assuming everything else remained the same, would Intel have great products, fast growing revenue, and strong profits today?

My guess is no. It might have only delayed the troubles Intel faces today by two or three years, but fundamentally, Intel was already on the wrong track.
For my money, you are partly right. Intel needed more volume to fill the fabs but the idea that it would have only bought Intel 2-3 years without the 10nm debacle is where I disagree.

As each process gets smaller you need a growing market to keep the fabs full. So yeah, Intel needed to move into a different market. Otellini understood this and tried, unsuccessfully, to move into adjacent markets to get that volume. Having said that, Intel would have held on to a lot more of the PC market if 10nm had been successful. Krzanich nearly killed the company by overseeing the 10nm debacle, insisting on using Intel's proprietary design methodologies for Intel's foundry efforts and not building capacity because he didn't know how to fill it, but Intel needed to find a way to bring in more product.
In 2007, global PC sales totaled around 360 million units, compared with 122 million smartphones.

Fast forward to 2024: global PC sales were about 255 million units, while smartphone sales reached 1.24 billion units.

The PC market that Intel relies on is either stagnating or shrinking over the years. Unfortunately, Intel didn’t miss the mobile revolution, it consciously walked away from it.
You can't compare units directly, because the cost of a mobile chip is quite a bit less than for a CPU. An A-18 chip is reported to sell for $45. An A-20 chip is reported to sell for $280. An Intel I9 14900KS lists for $730. From what I have heard that was why Otellini walked away from the iPhone deal. Intel's process was optimized for processor speed and producing chips on the process was far more expensive than the corresponding foundry process. At the time Intel drove the industry and cost was not an object.
Another example is in AI processors. Nvidia, AMD, and Intel all rely on TSMC for manufacturing, yet Intel’s AI products still lag behind in competitiveness.
Lip-Bu Tan seems to think that inference and agentic AI are going to be much bigger than the training market that Nvidia is currently dominating and is taking steps to get into that market. You'll have to forgive me for assuming he knows what he is doing. His track record is hard to argue with.
 
Back
Top