Strong AI Chip Demand Pushes TSMC's July Revenue by 45% Year-over-Year

[AleksandarK]

The demand for AI accelerators is going strong, and the world's largest semiconductor manufacturer, TSMC, has just confirmed that with its July 2024 revenue report. According to its latest July 2024 data, TSMC has reported a consolidated revenue of NT$256.95 billion, or about $7.94 billion at the time of writing. This represents a massive 23.6% jump from June 2024 and a 44.7% from July 2023, when revenue came in at NT$207.869 billion and NT$177.616 billion, respectively. For revenue throughout the year, measured from January to July, TSMC booked NT$1.523 trillion, or about $47 billion at the current rate. For this 7-month period, TSMC's revenue has increased by 30.5% Year-on-Year (YoY), showing great demand and an uptick in the company's production capabilities.

Of course, this is possible thanks to the massive demand driving AI chip sales from various startups and established giants like NVIDIA and AMD. Another vital customer for TSMC is Apple, which produces smartphone and Mac chips at Taiwanese facilities. The solid financial results from TSMC suggest that other fabless chip designers in its ecosystem may also experience positive outcomes in their earnings. It's worth noting that the semiconductor supply chain operates on a long-term planning basis, with arrangements made months in advance. As such, we can expect advanced silicon solutions to reach new customers in the coming months, further driving growth in the sector.



View at TechPowerUp Main Site | Source

 

[bonehead123]

Well DUH.....said nobody, hahahahaha

I STILL say "TSMC" stands for "The Smartest Manufacturing Company" !

 

[Prima.Vera]

What is Intel or Samsung doing??

 

[Onasi]

What is Intel or Samsung doing??

Trying their best, I would assume. Turns out, cutting edge lithography is rather hard.

 

[Visible Noise]

What is Intel or Samsung doing??

Samsung has turned in another bumper result on the back of strong AI-driven demand, unveiling a 15-fold increase in Q2 operating profit.

The result exceeded market expectations, helping drive the stock 3.6% higher on the Seoul exchange Wednesday.

The conglomerate reported operating profit of 10.44 trillion Korean won (US$7.6 billion), up from KRW670 billion ($480 million) a year ago and KRW3.84 billion ($2.8 million) in Q1. Total revenue was KRW74.1 trillion ($54.1 billion), up 23%, with an EBITDA (earnings before interest, taxes, depreciation and amortization) margin of 27%, up from 16% a year ago and 22% in Q1.

Samsung profit skyrockets on surging memory demand

Demand for AI memory powered Samsung's Q2 profit surge. But its networks business slumped 20% as operators trim capital expenditures.

www.lightreading.com

So if this part of the article is true - "For revenue throughout the year, measured from January to July, TSMC booked NT$1.523 trillion, or about $47 billion at the current rate." - Samsung had more revenue in the last quarter than TSMC has had all year.

 

[Prima.Vera]

Trying their best, I would assume. Turns out, cutting edge lithography is rather hard.

Aren't all taking their tools from ASML ??

 

[ThomasK]

Aren't all taking their tools from ASML ??

The tools are, silicon design is not.

 

[TheLostSwede]

Samsung has turned in another bumper result on the back of strong AI-driven demand, unveiling a 15-fold increase in Q2 operating profit.

The result exceeded market expectations, helping drive the stock 3.6% higher on the Seoul exchange Wednesday.

The conglomerate reported operating profit of 10.44 trillion Korean won (US$7.6 billion), up from KRW670 billion ($480 million) a year ago and KRW3.84 billion ($2.8 million) in Q1. Total revenue was KRW74.1 trillion ($54.1 billion), up 23%, with an EBITDA (earnings before interest, taxes, depreciation and amortization) margin of 27%, up from 16% a year ago and 22% in Q1.

Samsung profit skyrockets on surging memory demand

Demand for AI memory powered Samsung's Q2 profit surge. But its networks business slumped 20% as operators trim capital expenditures.

www.lightreading.com

So if this part of the article is true - "For revenue throughout the year, measured from January to July, TSMC booked NT$1.523 trillion, or about $47 billion at the current rate." - Samsung had more revenue in the last quarter than TSMC has had all year.

You're missing the point here though, TSMC = Foundry and that's it, Samsung = Foundry + makes just about anything and everything else.

Aren't all taking their tools from ASML ??

As already mentioned, yes, some of their main tools from ASML, but how these tools are used in the fabs differ between the three. This is why you can't take a chip design taped out for Samsung and bring it over to TSMC without spending at least six months making changes to the design so it fits with TSMC's tools. At one point there was a design alliance between Samsung, GloFo and IBM, but even that required some extra work if you wanted to move foundry.
TSMC appears to have the best people when it comes to taking a design from one of their customers and turning it to a successful, mass produced chip. That said, TSMC has had its fair share of issues over the years too, but seemingly never as bad as Samsung and to some part Intel.
Also keep in mind that the tools from ASML is just one part that is involved in these fabs, there's tooling from a bunch of other companies involved as well, plus materials suppliers that can make a difference.

 

[Onasi]

@TheLostSwede
The article even says:

As expected, it was the device solutions unit, which includes the memory and foundry segments, that powered the result.

The business, which recorded a KRW4.4 trillion ($3.2 billion) loss last year, posted operating profit of KRW10.8 trillion ($7.9 billion), with sales of KRW28.6 trillion ($20.9 billion), up 94% year-over-year and 23% sequentially.

The company said the ramp-up of AI investments by hyperscaler customers meant robust demand for conventional memory products DRAM and SSD as well as its high-bandwidth memory (HBM) line.

Like, yeah. We all know HBM is like crack coated muffins with how much it is in demand. But nothing suggests Samsung is any closer to getting any significant contracts for their new foundry nodes. In fact, I don’t think they had any at all since Ampere.

 

[Visible Noise]

You're missing the point here though, TSMC = Foundry and that's it, Samsung = Foundry + makes just about anything and everything else.


As already mentioned, yes, they all use some of the main tools from ASML, but how these tools are used in the fabs differ between the three. This is why you can't take a chip design taped out for Samsung and bring it over to TSMC without spending at least six months making changes to the design so it fits with TSMC's tools. At one point there was a design alliance between Samsung, GloFo and IBM, but even that required some extra work if you wanted to move foundry.
TSMC appears to have the best people when it comes to taking a design from one of their customers and turning it to a successful, mass produced chip. That said, TSMC has had its fair share of issues over the years too, but seemingly never as bad as Samsung and to some part Intel.
Also keep in mind that the tools from ASML is just one part that is involved in these fabs, there's tooling from a bunch of other companies involved as well, plus materials suppliers that can make a difference.

These are just Samsung Electronics results, not the entire conglomeration.

"The DS Division posted KRW 28.56 trillion in consolidated revenue and KRW 6.45 trillion in operating profit for the second quarter."

Roughly $21B is sales and $4.7B in profit in chips. They aren't struggling.

@TheLostSwede
The article even says:

Like, yeah. We all know HBM is like crack coated muffins with how much it is in demand. But nothing suggests Samsung is any closer to getting any significant contracts for their new foundry nodes. In fact, I don’t think they had any at all since Ampere.

"The System LSI Business posted record-high sales for the first half of the year as it saw earnings improve in the second quarter due to increased supply of key components such as systems on chips (SoCs), image sensors and display driver ICs (DDIs) for major flagship products."

"The Foundry Business saw improved earnings as a result of increased demand across applications. Due to higher orders for sub-5nm technology, the number of AI and high-performance computing (HPC) customers increased twofold from a year earlier. The Foundry Business also distributed the process development kit (PDK) for 2nm Gate-All-Around (GAA) technology to customers ahead of mass production in 2025.
In the second half, the Foundry Business expects a rebound in mobile demand and continued high growth in demand for AI/HPC applications. As a result, the foundry market is expected to experience growth overall, particularly in advanced nodes. For 2024, on the back of full-scale mass production of second-generation 3nm GAA technology, the Company expects growth to outpace the market."

 

[Onasi]

@Visible Noise
Samsung System LSI isn’t the foundry business. That would be the aptly named Samsung Foundry division, to be precise. They are both part of Samsung Semiconductors though, true.

 

[AnotherReader]

Aren't all taking their tools from ASML ??

There's a lot more to semiconductor fabrication than the tools from ASML. If that wasn't the case, EUV would have been required before TSMC's N5 node. For a long time, DUV lithography was the state of the art, and it didn't stop Intel or TSMC from progressing to ever denser nodes.

 

[TheLostSwede]

Like, yeah. We all know HBM is like crack coated muffins with how much it is in demand. But nothing suggests Samsung is any closer to getting any significant contracts for their new foundry nodes. In fact, I don’t think they had any at all since Ampere.

DS should be the semiconductor business, but if you remove memory, that leaves about US$5 billion from the non memory business, or in other words, the LSI and Foundry BU's together, produced a only quarter of the revenue of the DS unit at Samsung.

Samsung Electronics Announces Results for Second Quarter of 2024

Samsung Electronics today reported financial results for the second quarter ended June 30, 2024. The Company posted KRW 74.07 trillion in consolidated revenue and operating profit of KRW 10.44 trillion as favorable memory market conditions drove higher average sales price (ASP), while robust...

www.techpowerup.com

@Visible Noise
Samsung System LSI isn’t the foundry business. That would be the aptly named Samsung Foundry division, to be precise. They are both part of Samsung Semiconductors though, true.

This is the LSI BU, but it and the foundry BU are both listed under Samsung Semiconductor as you say, with memory being a separat BU again.

System LSI | Business-area | Samsung Semiconductor Global

Samsung's System LSI division designs core semiconductors. Explore how System LSI leads the field of system semiconductors, from everyday electronics to 5G and AI.

semiconductor.samsung.com

There's a lot more to semiconductor fabrication than the tools from ASML. If that wasn't the case, EUV would have been required before TSMC's N5 node. For a long time, DUV lithography was the state of the art, and it didn't stop Intel or TSMC from progressing to ever denser nodes.

ASML actually made a lot of that tooling as well though, but at least they had competitors, to a point when they just exceeded all their competitors in terms of equipment quality and performance.

"The Foundry Business saw improved earnings as a result of increased demand across applications. Due to higher orders for sub-5nm technology, the number of AI and high-performance computing (HPC) customers increased twofold from a year earlier. The Foundry Business also distributed the process development kit (PDK) for 2nm Gate-All-Around (GAA) technology to customers ahead of mass production in 2025.
In the second half, the Foundry Business expects a rebound in mobile demand and continued high growth in demand for AI/HPC applications. As a result, the foundry market is expected to experience growth overall, particularly in advanced nodes. For 2024, on the back of full-scale mass production of second-generation 3nm GAA technology, the Company expects growth to outpace the market."

This is all nice, but Samsung keeps proving that the yields they reach aren't good enough for their customers, so they keep going back to TSMC.
Apple, Nvidia, Qualcomm and others have tried to save some money by going to Samsung, but their nodes never deliver what they promise.
Until Samsung solves that problem, which is a pretty major one, they're always going to be a tier two foundry.

 

[Onasi]

@TheLostSwede
Yeah, I saw that link too, but the SamSemi site is a bit weird and the foundry services are a separate tab:

Foundry Overview | Samsung Semiconductor Global

Samsung Foundry creates a world beyond imagination by offering competitive processes, design technologies, IP, and high-volume manufacturing capability.

semiconductor.samsung.com

Also interesting how there are four fabs are listed as “Advanced Node” on here:

Foundry Overview | Samsung Semiconductor Global

Samsung Foundry creates a world beyond imagination by offering competitive processes, design technologies, IP, and high-volume manufacturing capability.

semiconductor.samsung.com

Which is fair enough, but I was under the impression that the Taylor site is yet to enter mass production at all.

 

[cvaldes]

Aren't all taking their tools from ASML ??

Yeah, there's a lot more to making chips than just having a fancy litho machine.

In the same way, if you buy top-tier Kitchenaid stand mixer, it's not going to make you a world-class pastry chef. Same with hammers, scalpels, paintbrushes, whatever.

There are a lot of things that determine the final product. Chip design is one but not the only factor. We saw that with Nvidia's Ampere generation with Samsung as the foundry: lots of transient spikes. And Nvidia switched back to TSMC.

Anyhow this shows how much the people who really know about chips value TSMC's capabilities.

 

[Darmok N Jalad]

It's rather interesting that the fabless companies that source from TSMC manage to have pretty good success shipping products that hit performance and delivery targets. For so many years, Intel was the king of process technology, or at least that's how it seemed since their CPUs were top performers with good margins. It makes me wonder what the collaboration looks like between AMD/Apple/NVIDIA and TSMC to end up with products that manage to compete so well. It seemingly goes against conventional wisdom on how to make the best semiconductors. When AMD dumped their fab business, it was heavily criticized, but it's like AMD knew that keeping up on process technology was not their core business, and their size made it impossible to get access to the best equipment. It's certainly panned out successfully for them.

 

[mb194dc]

It's amazing how many billions of chips have been sold, before there's even a decent front end use case.

One that'll generate the hundreds of billions needed to make the numbers work. To me, the whole llm technological approach to "AI" is fundamentally flawed. But we'll see I guess.

 

[Wirko]

As already mentioned, yes, some of their main tools from ASML, but how these tools are used in the fabs differ between the three. This is why you can't take a chip design taped out for Samsung and bring it over to TSMC without spending at least six months making changes to the design so it fits with TSMC's tools.

Yes. We're talking about ultra-expensive machines here, made for one of about six known customers, counting Micron+Kioxia+Hynix DRAM makers too (the EUV ones at least). Entire compact robotised factories in fact. This means that each one is made to order, to exact customer's specifications, and contains subsystems, software and other IP provided or co-developed by the customer. It just can't be any other way.

At one point there was a design alliance between Samsung, GloFo and IBM, but even that required some extra work if you wanted to move foundry.

Hm, wasn't that Samsung + IBM + AMD, with GloFo also remaining a member aften the spinoff from AMD?
IBM still develops stuff, about two years ago they demonstrated a 2 nm test chip, but I don't know if they still are a part of any alliance. Probably they sell the IP to any interested party.

 

[cvaldes]

It's amazing how many billions of chips have been sold, before there's even a decent front end use case.

One that'll generate the hundreds of billions needed to make the numbers work. To me, the whole llm technological approach to "AI" is fundamentally flawed. But we'll see I guess.

It's incorrect to assume that every single chip that rolls off of TSMC's manufacturing line is an AI chip.

There are many other usage cases for AI other than LLM; many of these non-LLM applications have been used by large corporations and governments for several years now.

Judging AI solely by the consumer chatbots (like ChatGPT) that are heavily covered in mainstream news is not a sane way of assessing the viability of the AI industry.

It's worth point out that something like Apple's A-series or M-series SoCs are not strictly AI chips anyhow. They have machine learning transistors (and have since 2017) but the entire chip isn't solely dedicated to AI. These chips also have a CPU, GPU, security processor, media encoding/decoding engines and other differentiated silicon. In fact, the Neural Engine (Apple's name for their ML cores) is probably no more than 10-15% of the total die area.

This would be the same with other mobile SoCs whether it be Samsung, Qualcomm, whoever.

 

[kapone32]

DS should be the semiconductor business, but if you remove memory, that leaves about US$5 billion from the non memory business, or in other words, the LSI and Foundry BU's together, produced a only quarter of the revenue of the DS unit at Samsung.


View attachment 358203

Samsung Electronics Announces Results for Second Quarter of 2024

Samsung Electronics today reported financial results for the second quarter ended June 30, 2024. The Company posted KRW 74.07 trillion in consolidated revenue and operating profit of KRW 10.44 trillion as favorable memory market conditions drove higher average sales price (ASP), while robust...

www.techpowerup.com

This is the LSI BU, but it and the foundry BU are both listed under Samsung Semiconductor as you say, with memory being a separat BU again.

System LSI | Business-area | Samsung Semiconductor Global

Samsung's System LSI division designs core semiconductors. Explore how System LSI leads the field of system semiconductors, from everyday electronics to 5G and AI.

semiconductor.samsung.com

ASML actually made a lot of that tooling as well though, but at least they had competitors, to a point when they just exceeded all their competitors in terms of equipment quality and performance.


This is all nice, but Samsung keeps proving that the yields they reach aren't good enough for their customers, so they keep going back to TSMC.
Apple, Nvidia, Qualcomm and others have tried to save some money by going to Samsung, but their nodes never deliver what they promise.
Until Samsung solves that problem, which is a pretty major one, they're always going to be a tier two foundry.

Intel basically said no thank you to ASML. I don't think they will do that again.

 

[TheLostSwede]

Hm, wasn't that Samsung + IBM + AMD, with GloFo also remaining a member aften the spinoff from AMD?
IBM still develops stuff, about two years ago they demonstrated a 2 nm test chip, but I don't know if they still are a part of any alliance. Probably they sell the IP to any interested party.

Nope, this was after AMD had spun off GloFo.
It was called the Common Platform or CP for short.

2011 Common Platform Technology Forum to Feature Leading-Edge Manufacturing Solutions from IBM, Samsung and GLOBALFOUNDRIES

/PRNewswire-FirstCall/ -- IBM (NYSE: IBM), Samsung Electronics, Co., Ltd., and GLOBALFOUNDRIES today announced the 2011 Common Platform Technology Forum will...

www.prnewswire.com

 

[mb194dc]

It's incorrect to assume that every single chip that rolls off of TSMC's manufacturing line is an AI chip.

There are many other usage cases for AI other than LLM; many of these non-LLM applications have been used by large corporations and governments for several years now.

Judging AI solely by the consumer chatbots (like ChatGPT) that are heavily covered in mainstream news is not a sane way of assessing the viability of the AI industry.

It's worth point out that something like Apple's A-series or M-series SoCs are not strictly AI chips anyhow. They have machine learning transistors (and have since 2017) but the entire chip isn't solely dedicated to AI. These chips also have a CPU, GPU, security processor, media encoding/decoding engines and other differentiated silicon. In fact, the Neural Engine (Apple's name for their ML cores) is probably no more than 10-15% of the total die area.

This would be the same with other mobile SoCs whether it be Samsung, Qualcomm, whoever.

Well obviously, but a lot of them are. You can see from the revenue being generated by Nvidia in their data centre segment.

What has happened, is that there's been a feedback loop, between day trading investors, algos, "AI" being mentioned by corporations boosting their stock price for the last year. So management at Microsoft, Google, Meta et al, have spent tens of billions on chips, because doing so will boost their stock price...

Not because there is anything particularly useful they can do with any of the hardware, none of it is demand lead on the front end. Now investors are starting to question why they've spent all these billions. You can see Google trying to shove Gemini down your throat and Microsoft desperately pushing copilot all over...

From what I can see, the technology is fundamentally flawed due to the hallucination issue, there's no intelligence there, so you end up with lots of inaccurate nonsense and non contextual results, especially from Gemini in my experience.

If we really want to pursue "AI", it seems unlikely to me it's going to come from this technological angle. Rather, an entirely new branch of technology based on a novel approach, one that starts with reasoning. Not brute forcing with lots of data but no understanding of any of it, like a $100bn+ parrot.

 

[watzupken]

I feel foundry business tends to be a laggard when it comes to showing the overall health of the industry. The reason is because the chip orders were placed perhaps as far as 6 to 12 months ago. If demand for Nvidia chip have fallen off a cliff for example, you won't see the impact to TSMC that soon. So while its been a bountiful couple of years for foundry businesses, I am not sure how long that will last.

 

[Wirko]

From what I can see, the technology is fundamentally flawed due to the hallucination issue, there's no intelligence there, so you end up with lots of inaccurate nonsense and non contextual results, especially from Gemini in my experience.

If we really want to pursue "AI", it seems unlikely to me it's going to come from this technological angle. Rather, an entirely new branch of technology based on a novel approach, one that starts with reasoning. Not brute forcing with lots of data but no understanding of any of it, like a $100bn+ parrot.

It doesn't have to be "intelligent for real". Can we people even define and completely understand natural intelligence? It just has to be sufficiently advanced. Also see the Arthur C. Clarke's three laws.

Unrelated but ... I've heard from people who own 0€ ... 50€ parrots that these birds can do more than just parroting. They have some level of understanding of human words, to the point they seem to "answer questions". See this one too. And I'm not very much surprised. Some birds are considered quite intelligent and they use a complex language of their own to communicate.

 

[TheLostSwede]

I feel foundry business tends to be a laggard when it comes to showing the overall health of the industry. The reason is because the chip orders were placed perhaps as far as 6 to 12 months ago. If demand for Nvidia chip have fallen off a cliff for example, you won't see the impact to TSMC that soon. So while its been a bountiful couple of years for foundry businesses, I am not sure how long that will last.

As long as we need electronics, the foundry industry will never really slow down though. As it is, the world relies on the chips made by TSMC, Samsung, GloFo, UMC and others, so even if "ai" turns out to be a bust, these companies aren't going to go under any time soon and it's even unlikely that they'll see any significant drop in business. However, the "market" or rather investors, might not agree and thus the share prices will drop, as the "market" is legalised gambling with other people's money and often with companies being or not being as the stake.