TSMC posts blowout $39 billion quarter, raising the same old AI question: is this sustainable?

Skye Jacobs

Posts: 2,018   +59
Staff
Bottom line: TSMC delivered another strong quarter, offering a clear signal that demand for AI chips remains high. The company reported revenue of NT$1.27 trillion, or about $39.6 billion, for the June quarter, up 36% from a year earlier and in line with analyst expectations. The gains were driven in part by a sharp surge in June revenue, which jumped 68%. That late-quarter spike suggests that orders tied to AI infrastructure are accelerating, particularly among customers expanding their data center capacity.

TSMC sits at the center of that buildout. The company manufactures many of the world's most advanced chips used in AI data centers, including processors designed by Nvidia and Apple. As tech companies race to expand their data center capacity, demand for those chips has remained robust.

The broader AI spending wave is enormous. Global investment in AI infrastructure, driven by companies such as Meta Platforms, is expected to exceed $725 billion this year. That spending includes servers, networking equipment, and specialized AI accelerators, much of which depends on chips manufactured by TSMC.

Still, not everyone is convinced the current pace is sustainable. Investors have begun to question whether tech companies are building more capacity than they will ultimately need. A significant portion of that spending is debt-funded, and there is still little visibility into when, or whether, those investments will generate strong returns.

Credit: Bloomberg

Chipmakers, however, continue to push back against those concerns. In June, Chief Executive Officer C.C. Wei warned that TSMC would be unable to keep up with demand from American customers for years, even as new US manufacturing capacity comes online.

That view is echoed elsewhere in the supply chain. SK Hynix has said shortages of advanced memory chips, particularly the high-bandwidth memory used in AI systems, could persist into the next decade.

TSMC's full earnings report later this week is expected to provide a clearer picture of how demand is holding up. Investors will pay close attention to the company's capital spending plans, which are often viewed as a proxy for future supply and demand across the semiconductor industry. TSMC has said it plans to spend nearly $56 billion this year, approaching a record.

Credit: App Economy Insights

The company is also expanding its footprint beyond Taiwan. It is investing heavily in the US, including an Arizona campus expected to represent a total investment of about $265 billion. The move aligns with Washington's push to bring more advanced chip production onshore and reflects TSMC's growing importance to both economic and geopolitical security.

Permalink to story:

 
I've been trying to find the numbers on this, because the cancellation rate and expected completion rate are starting to get out of hand. However, the only source I could find anything contrary to those numbers was actually an advertisement for an AI model masking itself as a news article stating that all the information about floating about data center delays were AI hallucinations and that you should invest in their AI model because it's more accurate than the others.
 
I've been trying to find the numbers on this, because the cancellation rate and expected completion rate are starting to get out of hand. However, the only source I could find anything contrary to those numbers was actually an advertisement for an AI model...
Funny, many months ago you told us that you had inside knowledge -- which "nondisclosure agreements" kept you from specifying -- that "all the AI datacenter projects were being cancelled, and most of those already built were being shut down." Yet daily the tide still increases.
 
Can't really find a reason to celebrate the successes when they arise from neglect of the consumer market. I'm sure someone up at the top is glad.
 
Funny, many months ago you told us that you had inside knowledge -- which "nondisclosure agreements" kept you from specifying -- that "all the AI datacenter projects were being cancelled, and most of those already built were being shut down." Yet daily the tide still increases.
Yes, like the projects I work on are being delayed and cancelled, I only work on the east coast US and only have direct knowledge of the projects that I and the company I'm employed by have contracts with.

And I knew you would be the one to reply. It's why I didn't say anything one way or the other in my post about what the numbers actually are. I was also hoping that anyone who read my post would be given enough information to go do their own research and find numbers in their own.

Mind posting some of the numbers YOU'VE been able to find? Maybe you have access to information I don't have and you can clarify any misunderstanding I have about what I'm seeing.
 
Mind posting some of the numbers YOU'VE been able to find? Maybe you have access to information I don't have and you can clarify any misunderstanding I have about what I'm seeing.
A rational, well-considered non emotion-fueled response from you. Kudos. The figures I've seen estimate 50% of all datacenter projects being delayed or canceled, due to higher anticipated costs or other shortages.

Your error is the conclusion you derive from that. Most of these were planned prior to the memory price explosion. When buildout costs rise to that extent, many less-urgent projects are going to affected. But even with those cancellations, datacenter spending is expected to triple to $7T by 2030, with a corresponding increase on both the number and size of such centers. The market isn't going away; it's exploding.

 
A rational, well-considered non emotion-fueled response from you. Kudos. The figures I've seen estimate 50% of all datacenter projects being delayed or canceled, due to higher anticipated costs or other shortages.

Your error is the conclusion you derive from that. Most of these were planned prior to the memory price explosion. When buildout costs rise to that extent, many less-urgent projects are going to affected. But even with those cancellations, datacenter spending is expected to triple to $7T by 2030, with a corresponding increase on both the number and size of such centers. The market isn't going away; it's exploding.

And that's why I brought up the number of estimated completion rate. That number takes any personal analysis away from me or you. The estimated completion rate of both started and proposed Data Centers in the US is way lower than the 50%. Go dig up some numbers on that and I'll show you what I found
 
Go dig up some numbers on that

"Global data center inventory surged year-over-year in Q1 2026 across all major regions, fueled largely by hyperscale and AI demands. Latin America led inventory growth at 41.3%, followed by North America at 33%.

  • - Despite substantial new supply, strong demand from AI startups, neoclouds and hyperscalers has driven vacancy rates to record lows in many key markets, such as Northern Virginia (0.3%) and Atlanta (1%) in North America...."

https://www.cbre.com/insights/reports/global-data-center-trends-2026

When discussing the health of a market, what matter is what IS being built, not what isn't.
 
$56 billion in capex approaching a record, while everyone downstream is still trying to figure out if AI inference actually pays for itself. TSMC's the only one guaranteed to win regardless of how the bubble question resolves... they get paid either way.
 
The thing about this is that everyone forgets about is TSMC had this problem long before the AI train stated rolling. They were already the go-to fab for many customers. Intel had to pay more than they wanted to get in line at TSMC, and that was BEFORE the AI boom.

Last time I checked, there is still a HUGE market for chips that have nothing to do with AI, and TSMC is still out in front of the competition as the place to build them.
 
Back