Hello Reader,
This week was packed due to SIJORI Week. The events I attended spanned four w.media conferences, two roundtables I moderated, and a networking event.
A small irony of these conferences is how I often don't have the luxury of sitting through all the sessions I want to, nor the opportunity to engage in deeper conversations amid the hustle and bustle of the exhibition floor.
What I did manage to do, though, was to meet over a hundred people involved in the data centre industry and hear what they had to say. It is gratifying to hear how many told me they found my content useful or felt helped by it. Others wanted to know more about my writing process.
So here's what I'll write about today: how you can be a data centre "expert" yourself, and the rise of useful AI and its implications for data centres.
You can be an expert
What does it take to know a niche really well? The journey for me started in 2012 when I was approached to update a commercial data centre report. I wrote a couple of them before I was separately approached to do some freelance writing work for DCD in 2013. One thing led to another, and I soon started writing on a regular cadence for them.
Data centres were simply a minor part of what I devoted my time to. In truth, I wrote for multiple publications at the same time, including product reviews for PC World, IT deep dives for TechRepublic, bi-weekly newsletters for Questex in the US, and even marketing technologies for Questex Asia.
If you ask me, the most important thing in learning is understanding how things work. Too often, we treat areas we are unsure about as black boxes that we then proceed to ignore. I have difficulty with that; I feel compelled to read up or ask until I get clarity.
Mind you, I'm not saying you must figure out everything. The trick is getting to the point where you know enough to know what you don't know. At that stage, you wilGUIGl have a good inkling of experts you can approach, or even what questions to ask AI.
Useful AI
If there is one takeaway from this week's conference, it will be that we are seeing the start of the rise of useful AI. SemiAnalysis's Daniel Nishball, in his keynote, reiterated how his firm spends around US$171K a week on Claude Code. In case you are wondering, this works out to US$10 million a year.
But it was what I heard from GMI Cloud's Andy Chen that truly stuck with me. In the panel discussion I moderated, Andy shared how his firm, a neocloud specialising in renting GPUs, jumped from one to two million tokens per day six months ago to over 200 million tokens today.
This got me digging into token consumption, and the figures are massive. Token consumption has risen exponentially over the last six months and looks set to continue surging. In fact, GMI Cloud expects token use on their platform to reach a trillion tokens a day by the end of this year.
In commentaries I've written over the last few months, I've observed how agentic coding platforms have become so good that they are now reshaping how software is being created. Add in more capable AI agents and it's easy to see that the consumption of tokens is set to keep rising.
What does this mean for data centres? Just 12 to 18 months ago I was dismissive about the impact of AI on data centres. As I look at what AI can now do, my feeling is we will need a lot more data centres for AI inferencing before the day is up.
What are your thoughts? As usual, you can reach me by hitting reply to this email. I look forward to hearing from you.
Regards,
Paul Mah
