Hello Reader,
I was in Vietnam earlier this week for a whirlwind visit to Subzero Engineering's new factory, marking my second factory tour since my visit to Schneider Electric in Barcelona back in 2024.
The drive from Tan Son Nhat International Airport was around two hours one-way, but the trip gave me a much deeper appreciation of the many components that find their way into a typical data centre.
For today, let's talk about the enormous data centre supply chain and the slippery slope of AI use.
The data centre supply chain
When we talk about data centres, we intuitively know that they are made up of many thousands of hardware components from tens, even hundreds of suppliers within a broad ecosystem.
Over the years, I've had the fortune of meeting or working with many of them, from Singapore-invented RDHx makers like KoolLogix, GPU server brands such as Asus, membrane technology makers such as Suntar, and many more.
But have we really stopped to consider the full extent of this ecosystem? I'm not talking about the obvious components that we often see discussed, such as servers, CDUs, UPS, firefighting systems, and even chillers and cooling towers.
I'm talking about the partitions and doors used in hot aisle containment, the water pipes hidden from view, the PDUs for distributing AC power within racks, the automatic transfer switch (ATS) units that keep power flowing during outages. Yes, even the external security fencing and the mantraps in the lobby.
Each of these suppliers is reliant on their own network of suppliers for materials and parts. The result is a vast web of dependencies that stretches far deeper than most people realise. Beyond the Nvidias, Vertivs, and Eatons of the world, a sizeable fraction of the billions spent on data centres cascades through layers of the economy, reaching businesses and workers most of us will never hear about.
My point? A crash or sharp downturn in data centre spending would send ripples well beyond the industry itself, affecting livelihoods and businesses that few would think to connect to a server rack.
Faster with AI, but at what cost?
We already know that AI can help people do their jobs faster. But does this come with trade-offs? Well, new research from Anthropic has confirmed what many of us had long suspected: overreliance on AI can hinder the acquisition of new skills.
A group of 52 software engineers were asked to perform a 30-minute coding assignment in Python, a language they were familiar with. In the study, they were required to write the code for a relatively advanced task using a Python library that was new to them. A 25-minute quiz was administered immediately after to test for mastery of the concepts from the assignment.
Unsurprisingly, those who made use of AI fared significantly worse in the quiz. What's worth noting is that AI use didn't automatically mean a lower score - how they used AI did. Put simply, those who scored well made the effort to understand the code produced by AI, either by asking for explanations or solving bugs manually.
In short, the higher scorers used AI to speed up their understanding of the unfamiliar library and to code better, not merely to complete the task. They took the extra effort to grapple with the underlying concepts, asynchronous programming in this case, even as they used AI, gaining both productivity and new coding skills.
The concern, as noted by the study's authors, is the potential for AI to stunt the skill development of junior engineers. They wrote: "[To] accommodate skill development in the presence of AI, we need a more expansive view of the impacts of AI on workers... productivity gains matter, but so does the long-term development of the expertise those gains depend on."
What AI taught me about writing
The Anthropic study also underscored my realisation that there are many ways to apply generative AI to writing. Yet even from this sprawl, they can be broadly categorised into two distinct camps: approaches that deepen our expertise, and those that offer little more than one-off productivity boosts.
The danger of leaning too heavily on the latter is that we stay stuck with mediocre skills. And because writing is thinking, AI becomes a crutch that ensures we never learn to stand on our own two feet when it comes to sparking fresh ideas and sharpening our judgement.
When I first tried ChatGPT, I was as fascinated with the idea of learning advanced prompting techniques as everyone else was back then. Remember the tricks? Emulating the style of a favoured author, chain-of-thought prompting, storytelling frameworks?
My problem was I could see the flaws in AI-generated copy as clear as day, regardless of every technique I tried. Moreover, the compulsive overuse of prose techniques grated on my taste. Over time, my approach shifted towards using AI to become a better writer instead. I'm glad to report that I now write considerably faster than I ever have, while producing pieces with stronger insights.
It doesn't end with writing though. The practice itself sharpened my thinking as I examined issues critically. Often, what I set out to write would morph into something else as the arguments shaped themselves. This wouldn't be possible leaning heavily on AI, given its innate ability to generate what I call "synthetic coherence."
AI will absolutely shake up industries and transform the face of work. But how we apply it still rests with us. What would your call be? To lean on AI and polish its output, or to use it as a tool for honing your own thinking? Use AI to think faster and better, not to avoid thinking at all.
Hit reply and tell me what you think.
Regards,
Paul Mah
Content collaboration
Love the way I write? I'm opening up for more content collaborations. If you are looking for event coverage, brand storytelling, or a way to share your unique perspectives with a broader audience, fill in the form
here and I'll get in touch.
