Hello Reader,

We're almost in May, the month when Singapore announced 300MW of new data centre capacity at last year's ATxSummit. Back then, IMDA said it would make up to 300MW available "in the short term."

As we approach the one-year mark with no new updates - and no fresh DC-CFA to allocate additional data centre capacity, I have a haunch. All signs point to another shift in Singapore's data centre policy, and AirTrunk's rumoured 80MW data centre might be part of it.

("Signs point to another shift in Singapore's data centre policy" here.)

How much capacity do we really need?

Of course, the next question would be: How much data centre capacity do we actually need for AI workloads? Because that's what really matters now, isn't it? If AI is going to define the next era of innovation, we probably want the infrastructure powering it right in our own backyard - just in case proximity ends up mattering more than we think.

And if you haven’t been following closely, the latest reasoning models are startlingly capable - able to perform methodical, step-by-step reasoning combined with relevant online searches, all at machine speed.

("AI battle heats up with new OpenAI models..." here.)

Back to the question. I had the chance to sit down with Dr Leslie Teo of AI Singapore for a briefing this week on the latest developments of the SEA-LION LLM project. One point from our conversation stuck with me: that the data centre capacity we'll need is ultimately unknowable.

The logic goes like this:

“

There are three big unknowns: the future capabilities of GPUs, emerging AI techniques and algorithms, and the actual processing power we’ll require. The only thing we can be sure of is this: any estimate of future data centre requirements must rely on outdated historical data - practically guaranteeing we'll get it wrong.

The reality of AI training

But do we even need all this capacity in local data centres? Beyond a relatively small number of GPUs for certain tasks, the answer is likely no. There’s nothing stopping a capable AI team from training models in a data centre across town or halfway around the world.

This is also why efforts such as the AI Diffusion framework, aimed at keeping Chinese tech firms from accessing advanced GPUs, may frustrate and slow them - but they won’t stop them. But that’s a discussion for another day.

However, I do think it's vital to maintain a healthy level of data centre capacity, if only to signal to investors and talent that Singapore is serious about building AI capabilities. And in a world that’s abruptly turned inward, discarding the trade certainties of the past few decades, can Singapore afford to squander a potential ace card?

But what of the next generation?

I’ll leave you with one final thought this week. I bumped into a senior director from a government agency who I know is highly competent from a technical standpoint. We spoke briefly about the astounding capabilities of the latest AI models.

I then posed a question.

“

"But what about the next generation?" I asked. "While senior developers will no doubt benefit greatly from AI, what happens to junior developers? And if there’s little or no room for them in the workforce, where will the next generation of senior developers come from?"

He has no clear answer for me, only conceding that things are changing rapidly.

It’s worth noting that this isn’t just about developers. As AI continues to improve in areas such as coding, writing, illustration, and performing complex tasks through intelligent agents, the impact is set to grow. And it won’t be limited to tech roles. The ripple effects will eventually reach every creative and knowledge-based profession.

So here we are, on the edge of an AI-driven future, building data centres we may or may not need, for jobs that may or may not exist. The question isn’t just whether Singapore is ready for what’s coming. It’s whether anyone truly is.

Regards,
Paul Mah.

Spotlight