Report: Rise of AI Is Corroding Apple's Influence Over TSMC

[MacRumors]

A detailed new report from SemiAnalysis argues that while Apple's partnership with TSMC created the modern leading-edge foundry model, the rapid rise of AI computing is changing who pays for new chip technologies and how much influence Apple holds over each new manufacturing node.

The report traces how Apple became the single most important customer in the global semiconductor industry by committing early and at scale to new manufacturing processes at TSMC, beginning with the A8 chip in 2014. It argues that Apple's willingness to absorb early costs, fund yield improvements, and align its annual product cycle with TSMC's roadmap allowed the foundry to outpace rivals and cement its dominance at the leading edge.

Apple's annual spending at TSMC grew from roughly $2 billion in 2014 to about $24 billion in 2025, while Apple's share of TSMC revenue rose from single digits to as much as 25% at its peak. For much of the past decade, Apple accounted for more than half of initial production at each new node, and in some cases nearly all of it, effectively underwriting the financing of advanced manufacturing when no other customer could do so at scale.

However, the rise of AI accelerators has created a second class of customers, such as NVIDIA, who are capable of consuming large amounts of advanced manufacturing capacity. TSMC's revenue mix has shifted significantly as a result.

Smartphones once represented nearly half of TSMC's revenue, but that share has fallen as high-performance computing demands, including AI, have grown to become the largest segment. This means Apple is no longer the only customer capable of funding new capacity, even though it remains the largest single customer by revenue.

This change is said to already be visible with upcoming manufacturing nodes. Apple's share of early production for TSMC's N2 and A16 nodes is expected to be lower than previous generations, the latter of which is designed for high-performance computing workloads rather than mobile devices.

Under SemiAnalysis's modeling, Apple's position strengthens again at later nodes such as A14, which are being designed from the start to support both mobile and high-performance products. In that scenario, Apple's share of early capacity rises again as its iPhone and Mac chips return to being the primary drivers of volume.

While Apple continues to rely on TSMC for its most advanced chips, SemiAnalysis says the company is exploring alternatives for lower-risk components and certain categories to diversify its supply chain. Intel's upcoming 18A-P process is a potential option for some Apple silicon without disrupting flagship products.

See SemiAnalysis's full report for more information.

Article Link: Report: Rise of AI Is Corroding Apple's Influence Over TSMC

 

[bollman]

AI: the death of personal computing

 

[WarmWinterHat]

AI: the death of personal computing

Just wait it out. Shareholders are going to stop pumping money into these companies sooner rather than later, since they have absolutely no path whatsoever towards profitability.

 

[1037]

I need more Ai Cat videos.

 

[Bustycat]

View attachment 2595289

Just wait it out. Shareholders are going to stop pumping money into these companies sooner rather than later, since they have absolutely no path whatsoever towards profitability.

Including Alphabet with the net income of 100 billion dollars in 2024?

 

[bollman]

Just wait it out. Shareholders are going to stop pumping money into these companies sooner rather than later, since they have absolutely no path whatsoever towards profitability.

Oh, you just wait a bit too. I'm not so sure Microsoft, nVidia, AMD and the likes are all that interested in going back to letting users "own" anything. The very concept of renting a computer when you need it is a wet dream for these tech giants.
AI is going to hit the wall hard, and there will be huge data centers doing nothing. A wonderful opportunity to make people rent whatever they need instead of selling people a computer every 5 years or so.
We are in the age of "you will own nothing and be happy" and it is not all that easy to push back on it.

 

[WarmWinterHat]

Including Alphabet with the net income of 100 billion dollars in 2024?

Alphabet isn't causing this, OpenAI, Grok and similar ilk are.

Oh, you just wait a bit too. I'm not so sure Microsoft, nVidia, AMD and the likes are all that interested in going back to letting users "own" anything. The very concept of renting a computer when you need it is a wet dream for these tech giants.
AI is going to hit the wall hard, and there will be huge data centers doing nothing. A wonderful opportunity to make people rent whatever they need instead of selling people a computer every 5 years or so.
We are in the age of "you will own nothing and be happy" and it is not all that easy to push back on it.

If they do (and that's an awfully big if), then I will politely bow out of computing, outside of work (which my employer handles), and keep using what I have, fully offline, until it breaks. I won't rent personal computing, ever. No exceptions.

As for these AI data centers, they are not packed with regular gear that is useful in the situation you are describing. They are very, very power-hungry, and very AI-focused..not general-use chips, and they have a short life span. They will become warehouses, factories or abandoned. If they do stay as data centers, it'll be a full gutting and retrofit before they can be reused as you describe.

 

[sw1tcher]

View attachment 2595289

Just wait it out. Shareholders are going to stop pumping money into these companies sooner rather than later, since they have absolutely no path whatsoever towards profitability.

The U.S. govt could then step in.

The U.S. has already taken equity stakes in Intel, MP Materials, U.S. Steel, Lithium Americas, Trilogy Metals, etc.

For instance, if Oracle has trouble financing their datacenter build out, do you believe the U.S. won't step in when Oracle is the operator of TikTok U.S. and a big partner in the Stargate project?

 

[Nyx1]

Make your own fab Apple is only way

 

[d5aqoëp]

“The fall of AI cements Apple’s influence over TSMC once again”

Headline soon.

 

[WarmWinterHat]

“The fall of AI cements Apple’s influence over TSMC once again”

Headline soon.

Unfortunately I think it will be the "The fall of Taiwan cements China's control over TSMC."

 

[Alameda]

A detailed new report from SemiAnalysis argues that while Apple's partnership with TSMC created the modern leading-edge foundry model, the rapid rise of AI computing is changing who pays for new chip technologies and how much influence Apple holds over each new manufacturing node.

The report traces how Apple became the single most important customer in the global semiconductor industry by committing early and at scale to new manufacturing processes at TSMC, beginning with the A8 chip in 2014. It argues that Apple's willingness to absorb early costs, fund yield improvements, and align its annual product cycle with TSMC's roadmap allowed the foundry to outpace rivals and cement its dominance at the leading edge.

Apple's annual spending at TSMC grew from roughly $2 billion in 2014 to about $24 billion in 2025, while Apple's share of TSMC revenue rose from single digits to as much as 25% at its peak. For much of the past decade, Apple accounted for more than half of initial production at each new node, and in some cases nearly all of it, effectively underwriting the financing of advanced manufacturing when no other customer could do so at scale.

However, the rise of AI accelerators has created a second class of customers, such as NVIDIA, who are capable of consuming large amounts of advanced manufacturing capacity. TSMC's revenue mix has shifted significantly as a result.

Smartphones once represented nearly half of TSMC's revenue, but that share has fallen as high-performance computing demands, including AI, have grown to become the largest segment. This means Apple is no longer the only customer capable of funding new capacity, even though it remains the largest single customer by revenue.

This change is said to already be visible with upcoming manufacturing nodes. Apple's share of early production for TSMC's N2 and A16 nodes is expected to be lower than previous generations, the latter of which is designed for high-performance computing workloads rather than mobile devices.

Under SemiAnalysis's modeling, Apple's position strengthens again at later nodes such as A14, which are being designed from the start to support both mobile and high-performance products. In that scenario, Apple's share of early capacity rises again as its iPhone and Mac chips return to being the primary drivers of volume.

While Apple continues to rely on TSMC for its most advanced chips, SemiAnalysis says the company is exploring alternatives for lower-risk components and certain categories to diversify its supply chain. Intel's upcoming 18A-P process is a potential option for some Apple silicon without disrupting flagship products.

See SemiAnalysis's full report for more information.

Article Link: Report: Rise of AI Is Corroding Apple's Influence Over TSMC

Yeah, right. “We’re going to ignore the $24 billion customer,” said no company ever.

If nvidia drives some new tech at TSMC, it doesn’t inhibit TSMC from maintaining their business trajectory with Apple.

 

[Edsel]

AI is the 21st century version of the Public Library. Replacing card catalogs, aisles & miles of physical books, cataloged research documents, and basement dwelling research assistants, AI is slowly consolidating mankind’s accumulation of knowledge that will be vastly more attainable than the systems we all grew up with.
Apple, Inc. and the rest of the consumer product industry will now need to focus on how, and when, to integrate this vast knowledge into our laborious 20th century method of research. What products and tools will they create to integrate and communicate this seamless flow of information into our daily lives? It will look nothing like the devices and tools we use today.

 

[WarmWinterHat]

AI is the 21st century version of the Public Library. Replacing card catalogs, aisles & miles of physical books, cataloged research documents, and basement dwelling research assistants, AI is slowly consolidating mankind’s accumulation of knowledge that will be vastly more attainable than the systems we all grew up with.
Apple, Inc. and the rest of the consumer product industry will now need to focus on how, and when, to integrate this vast knowledge into our laborious 20th century method of research. What products and tools will they create to integrate and communicate this seamless flow of information into our daily lives? It will look nothing like the devices and tools we use today.

You're name is appropriate...Currently, AI is the 21st century's Edsel, and physical books and research assistants aren't going anywhere.

The best way to "integrate" such services, is let the end-user decide. Stop shoving AI down our throat at every turn. It doesn't fit every situation and every workflow, and it should be force-fed like it does.

 

[bradman83]

View attachment 2595289

Just wait it out. Shareholders are going to stop pumping money into these companies sooner rather than later, since they have absolutely no path whatsoever towards profitability.

I would imagine this would end similarly to the .com bubble of the 90’s. When that bubble burst thousands of startups went away but the strongest players, like Amazon, eBay, Yahoo, Google, Microsoft, and such lived to tell the tale and consumed the market share of these defunct firms. AI will likely be similar. Well funded, established firms with competent AI solutions like Microsoft/OpenAI, Google, and xAI will likely survive while smaller players get swept aside.

 

[WarmWinterHat]

I would imagine this would end similarly to the .com bubble of the 90’s. When that bubble burst thousands of startups went away but the strongest players, like Amazon, eBay, Yahoo, Google, Microsoft, and such lived to tell the tale and consumed the market share of these defunct firms. AI will likely be similar. Well funded, established firms with competent AI solutions like Microsoft/OpenAI, Google, and xAI will likely survive while smaller players get swept aside.

Google will survive. I don't think OpenAI, in it's current form, will. It will need to massively scale back it's spending and ambitions, and rid itself of Altman.

No idea about xAI, other than random internet people, I've never heard a single person even mention Grok. Most don't even know what it is. I've never used it, nor have I ever used X.

 

[swm]

so apparently the RAM prices are skyrocketing because someone bought - of course using money that doesn't exist "yet" - all memory chips to be manufactured in advance for GPUs that haven't been made. and those GPUs will be installed into datacenter that haven't been built, maybe didn't even left the drawing board. And all those nonexistent datacenters will be powered by the power infrastructure which will be upgraded by adding 20+ nuclear reactors to satisfy the increased power demand that isn't actually there.
all this in order to generate the profits that are mathematically impossible.
never mind that the construction time for a multi-GW NPP takes about a decade.
the hardware which depreciates in <3 years will be totally relevant by then.

 

[WarmWinterHat]

so apparently the RAM prices are skyrocketing because someone bought - of course using money that doesn't exist "yet" - all memory chips to be manufactured in advance for GPUs that haven't been made. and those GPUs will be installed into datacenter that haven't been built, maybe didn't even left the drawing board. And all those nonexistent datacenters will be powered by the power infrastructure which will be upgraded by adding 20+ nuclear reactors to satisfy the increased power demand that isn't actually there.
all this in order to generate the profits that are mathematically impossible.
never mind that the construction time for a multi-GW NPP takes about a decade.
the hardware which depreciates in <3 years will be totally relevant by then.

..data centers that are getting blocked right and left because of how disastrous they are to communities, and the push-back is picking up speed, on the right and left.

 

[kalsta]

You're name is appropriate...Currently, AI is the 21st century's Edsel, and physical books and research assistants aren't going anywhere.

Nope, AI is more like the motor car in general. Yes, people will still research topics by reading books, just as people still ride horses today—but like it or not, AI is going to dominate this space in the future. Edsel’s predictions may not be perfect, but I fear they are far more astute than yours.

 

[mellonseed]

I would be glad to see Intel could be starting to manufacture Apple's chip even it's just entry level ones. This could potentially break TSMC's monopoly in cutting-edge semi manufacturing, and help bring down the price.
And Intel's EMIB and Foveros (They're nice by the way) may enable an Apple Silicon that's larger than current Ultra chips, to consumer markets considering TSMC's advanced packaging capacity are all taken up by those data center chips.

 

[WarmWinterHat]

Nope, AI is more like the motor car in general. Yes, people will still research topics by reading books, just as people still ride horses today—but like it or not, AI is going to dominate this space in the future. Edsel’s predictions may not be perfect, but I fear they are far more astute than yours.

If you're doing actual research, and not using the source directly with citations, then you're not doing research. Research requires verification. I was a organic research chemist for years before moving into pharmacy.

My wife is an attorney and other attorney's are getting raked over the coals for using AI, and believing what it says, without following up on the original source. It's one of the reasons she refuses to use it, in it's entirety, as she doesn't want to be influenced by it.

People will keep reading books, for research and pleasure.

Nope, AI is more like the motor car in general.

In the long term, perhaps. LLMs, in their current power-hungry trillion-dollar datacenters form won't be the one that does it.

 

[Ghengis LeMond]

I'll bet the annual contract w/ Google for Gemini licensing is in the $40-50 Billion range (Google search is already $18B ....... seems about right.) Now, add in everything spent since 2019 on investment and development of the disaster that is Siri, Apple LLM development, 3rd party integration exploration (Anthropic/OpenAI/etc.) and exiting leadership buyouts ........ (total = $1 Trillion?) and you have more than enough evidence to declare that Apple is no longer an engineering company capable of competent product development.

^^^ This operating model reminds me of the way HPE works in the enterprise space.

 

[kalsta]

If you're doing actual research, and not using the source directly with citations, you're not doing research. Research requires verification. I was an organic research chemist for years before moving into pharmacy.

I would say the same about art. I fear this next generation will grow up thinking that they are creating art when they enter a prompt into a chatbot—when really they’re just playing a computer game—and stealing real artists’ work at the same time! I find this very sad. Nonetheless, this is the way the world is heading.

People will keep reading books, for research and pleasure.

I certainly hope so!

 

[Ethosik]

Oh, you just wait a bit too. I'm not so sure Microsoft, nVidia, AMD and the likes are all that interested in going back to letting users "own" anything. The very concept of renting a computer when you need it is a wet dream for these tech giants.
AI is going to hit the wall hard, and there will be huge data centers doing nothing. A wonderful opportunity to make people rent whatever they need instead of selling people a computer every 5 years or so.
We are in the age of "you will own nothing and be happy" and it is not all that easy to push back on it.

Yep. I called this from a million miles away. Cloud gaming is/was a test bench. If you can (reasonably) play games at 4k60 on any device and have manageable latency, ANY production workflow is possible.

 

[bollman]

As for these AI data centers, they are not packed with regular gear that is useful in the situation you are describing. They are very, very power-hungry, and very AI-focused..not general-use chips, and they have a short life span. They will become warehouses, factories or abandoned. If they do stay as data centers, it'll be a full gutting and retrofit before they can be reused as you describe.

Yes, and no. They may not be perfect for running vanilla windows/macOS, but I hardly doubt that is what they're aming for. Let us not forget, almost all of it will be "powered by AI" since you will have to talk or otherwise interact with the "computer" in other ways than a WIMP environment. Windows 11 (or 12 or whatever it will be called) is pushing hard for interacting with voice instead of your usual stuff.
Of course this will be marketed as the new good thing where the hardware is really cheap, but unable to do anything without a direct connection to these datacenters.