Not chips—transformers are the Taiwan-made components AI data centers need most. Even Musk and Zuckerberg can’t get around them.

[karin623]

Over the past year, most discussions around AI infrastructure have focused on GPUs, advanced packaging, and semiconductor supply chains. Yet growing evidence suggests the real bottleneck lies elsewhere: power infrastructure.

Specifically, ultra-high-voltage transformers. Lead times have stretched to three or four years, upstream components remain in chronic shortage, and even billion-dollar projects cannot easily accelerate delivery. Taiwanese manufacturer Fortune Electric has quietly become a key supplier to some of the largest AI data center projects in the United States, including Stargate, backed by OpenAI, SoftBank, and Oracle.

From Elon Musk’s xAI to Mark Zuckerberg’s push for “personal superintelligence,” the future of AI may depend less on who controls the most GPUs—and more on who can secure enough electricity to keep them running.

Not chips—transformers are the Taiwan-made components AI data centers need most. Even Musk and Zuckerberg can’t get around them.

Liang-rong Chen

cwnewsroom.substack.com

 

[hist78]

Over the past year, most discussions around AI infrastructure have focused on GPUs, advanced packaging, and semiconductor supply chains. Yet growing evidence suggests the real bottleneck lies elsewhere: power infrastructure.

Specifically, ultra-high-voltage transformers. Lead times have stretched to three or four years, upstream components remain in chronic shortage, and even billion-dollar projects cannot easily accelerate delivery. Taiwanese manufacturer Fortune Electric has quietly become a key supplier to some of the largest AI data center projects in the United States, including Stargate, backed by OpenAI, SoftBank, and Oracle.

From Elon Musk’s xAI to Mark Zuckerberg’s push for “personal superintelligence,” the future of AI may depend less on who controls the most GPUs—and more on who can secure enough electricity to keep them running.

Not chips—transformers are the Taiwan-made components AI data centers need most. Even Musk and Zuckerberg can’t get around them.

Liang-rong Chen

cwnewsroom.substack.com

"Hsu said that in the past, transformer lead times were about one year; now they are three to four years. Orders placed today may not be delivered until 2030.

For the fast-moving AI industry, this is simply too slow.

With financiers like Masayoshi Son and Larry Ellison behind Stargate., wouldn’t money solve the problem?

After all, for US$100 billion-scale investments, even a one-week delay costs hundreds of millions."

 

[Xebec]

There may be an opportunity for on-site (giant) batteries to help here. i.e. when workloads are below a certain % - charge up the batteries, but allow the batteries + transformers to power workloads exceeding 100% power draw from the local transformer.

 

[EganVector]

Over the past year, most discussions around AI infrastructure have focused on GPUs, advanced packaging, and semiconductor supply chains. Yet growing evidence suggests the real bottleneck lies elsewhere: power infrastructure.

Specifically, ultra-high-voltage transformers. Lead times have stretched to three or four years, upstream components remain in chronic shortage, and even billion-dollar projects cannot easily accelerate delivery. Taiwanese manufacturer Fortune Electric has quietly become a key supplier to some of the largest AI data center projects in the United States, including Stargate, backed by OpenAI, SoftBank, and Oracle.

From Elon Musk’s xAI to Mark Zuckerberg’s push for “personal superintelligence,” the future of AI may depend less on who controls the most GPUs—and more on who can secure enough electricity to keep them running.

Not chips—transformers are the Taiwan-made components AI data centers need most. Even Musk and Zuckerberg can’t get around them.

Liang-rong Chen

cwnewsroom.substack.com

Way back in 2022, IEEE Spectrum published a special report on the transition to EV's (The EV Transition Is Harder Than Anyone Thinks) that included a section on power distribution (https://spectrum.ieee.org/the-ev-transition-explained-2658463709) that highlighted the humble transformer as a limiting factor. The relevant quote about the city of Palo Alto, CA "A big problem is the 3,150 distribution transformers in the city, Marshall indicated. A 2020 electrification-impact study found that without improvements, more than 95 percent of residential transformers would be overloaded if Palo Alto hits its EV and electrical-appliance targets by 2030."

I have no idea who will "win" the AI race, but I'm planning to invest in companies that make electricity generation and distribution products.