Hello Reader,
This week I attended Microsoft's AI-themed "Frontier Transformation" conference at Singapore Expo, caught up with long-time friends at GovWare 2026 and published my first podcast with Micron.
For today's commentary, I'll talk about Singapore's plan to fight AI with AI, and the multi-trillion-dollar question of the week: What's up with those cancelled IPOs?
Fighting AI with AI
The Singapore government is gearing up its cybersecurity to fight the onslaught of AI-powered threats out there. And the only way to do that effectively will be to adopt AI capabilities.
While these weren’t the exact words used by Minister Shanmugam in his keynote at the opening of the Singapore International Cyber Week (SICW), he did say - and later reiterate - that Singapore will have to "defend with speed using AI."
Announced as part of Singapore's Cybersecurity Strategy 2026, the steps he outlined are a sign of the times. Specifically, Singapore will hasten the adoption of AI-enabled cyber capabilities and invest in AI harness engineering for cybersecurity.
What I found particularly intriguing was the decision to "mostly" use the best AI models available instead of our own. This is highly pragmatic. After all, the cost of a leading state-of-the-art AI model is now in the billions of dollars, and even those models are quickly supplanted within months.
The solution? Build a proprietary platform with the flexibility to work with multiple frontier models. Then plug in the appropriate model as required. Implicit within this strategy is an architecture that ensures Singapore is not beholden to any one firm or power bloc. I’ll share more on this another day.
Where's the capital?
It started with Firmus. After announcing what would have been Australia's largest-ever IPO, it eventually cancelled the planned US$5.5 billion listing. This happened despite preparatory work such as a unilateral announcement of a deal with OpenAI in Malaysia - tellingly, the press release didn’t list an OpenAI contact.
Days later, AirTrunk announced that a planned US$1.5 billion IPO would be delayed to 2027. While others like DayOne are going ahead, the jury is still out on whether we will witness a domino effect. What is happening? Has AI demand tanked and is demand for data centres drying up?
I think there is no evidence that AI demand is slackening. Having used AI coding tools heavily for the last six months, I think we are barely touching their potential. And based on numbers revealed in recent weeks, frontier AI labs such as Anthropic and OpenAI have shown that they can generate significant revenue.
What is happening, rather, is the staggering amount of capital required to develop data centres, particularly AI data centres. Fully fitting out an AI data centre, including GPUs, can cost between US$40 and US$60 million per MW. When you read headlines of 1GW of data centre builds, know that we are talking about US$40 to US$60 billion dollars.
It's a lot of money - and the market might well be running out of ready capital to, well, capitalise on the AI and data centre boom. Indeed, Broadcom CEO Hock Tan on Monday said that capital will be the biggest constraint facing the company in AI development.
The real bubble is expectations
But I personally feel there is another issue at stake: most investors don’t really understand the nuances of AI use and the data centre buildout. In a way, the challenge isn’t so much a bubble as the expectation of one by investors hoping to make a quick buck.
Tan put it this way: "A lot of investors have no clue what they're investing in, but set very extreme expectations, only to find out that it's almost like they are in a casino gambling."
Of course, the rapid-fire development of new capabilities doesn't help. I mean, I'm using AI tools daily, sometimes from morning to night. Yet there are days I feel like I am running up a downriding escalator - so quick is the cadence of improvements. How's that?
In the meantime, new AI models are released by the day, while businesses large and small are still trying to figure out how AI can help them at the organisational level.
When technology moves faster than our ability to make sense of it, where do we even begin?
ps: I'll be in Malaysia for w.media CDC this coming week from Wednesday afternoon; do ping me if you are near the Connexion Conference & Event Centre.
ps2: Please bear with me if I haven't responded to you via email yet. Just found some emails in spam. Will get back shortly.
Regards,
Paul Mah
