Mapletree Industrial Trust Selling Data Centers. They Should.

Six years ago, I wrote that Mapletree Industrial Trust (MIT) was becoming a data center REIT, and indeed, today more than 50% of its assets are data centers. It was meant to be a positive development, but alas, not all data centers are the same.

While Keppel DC REIT has soared 17% since Digital Core REIT started trading, the other two data center REITs have lost money even after factoring in dividends. I reckon MIT’s Singapore industrial properties are what saved it, with the counter “only” down 5%. Had MIT relied purely on its US data centers, I think it would have performed just as badly as Digital Core REIT, down 42% since its IPO in end-2021.

You might also like

5 Singapore Stocks Taking Concrete Steps to Boost Shareholder Value

October 16, 2026

Not Just the Banks. 5 Singapore Stocks Down 10% in a Month. Buy the Dip?

October 13, 2026

Why is this happening?

In the tech world, we’ve seen software stocks take a beating while AI stocks rallied. The same split is happening in the data center world too.

During MIT’s recent AGM, the minutes revealed that management was asked why it couldn’t simply ride the AI boom instead of selling these US properties. The answer, essentially, is that some of the affected buildings don’t have enough power.

Management said repositioning these buildings could require significant capex on upgrading electrical infrastructure and redevelopment, a process that could take four to five years, with no certainty of landing a tenant at the end of it.

So there are data centers built for the software era that simply can’t serve the AI era. MIT is being punished for holding legacy assets that can’t participate in the AI boom.

MIT has 54 data centers in the US. Of these, only three are clearly built to handle the massive computing needs of hyperscalers. The rest are mostly older or more conventional facilities, and it’s unclear how many can cope with the power and cooling demands of today’s AI chips. MIT doesn’t release enough information for a full assessment, so the true number of AI-ready properties could be higher than three.

Nevertheless, MIT has put up 22 data centers for sale, or 40% of the portfolio by property count. Not all of them will necessarily go, since MIT has said divestments will be selective. It could also be that MIT is testing market interest with a wide net before deciding what’s actually worth selling.

Of the 22 data centers marketed, most are low powered, below 10MW. I used 10MW as the cutoff because below that level, a facility generally isn’t well suited to the massive AI clusters hyperscalers are now building. Here’s a rough reference for data center power tiers.

• <3MW: legacy/small data centre
• 3–10MW: useful conventional/colo DC; can run AI, but not ideal for huge clusters
• 10–20MW: increasingly credible for sizeable AI workloads
• 20MW+: attractive for serious AI deployment
• 50–100MW+: hyperscale AI-campus territory

For the data centers MIT is looking to sell where public information is available, most appear to have less than 10MW of capacity. I couldn’t identify a single one with clearly disclosed IT capacity above 10MW.

That doesn’t make them useless data centers. But they were largely built for an earlier era, when companies needed relatively small amounts of computing power. Today’s AI giants increasingly want data centers capable of supplying tens or even hundreds of megawatts to massive clusters of GPUs.

Another sign of weakness in MIT’s data center portfolio is its relatively low occupancy rate of 82.5%.

CBRE’s H1 2026 report says vacancy across the eight primary North American data center markets fell to a record low of 1.4%, or about 98.6% occupancy. JLL is even tighter, saying North American vacancy has remained around 1% for the third straight year, implying roughly 99% occupancy.

Even Digital Core REIT reported a 97% occupancy rate.

America is experiencing one of the biggest data center shortages in history, yet MIT’s American data center portfolio is only 82.5% occupied. That suggests a real mismatch between its data centers and demand. Or put simply, these data centers just aren’t riding the AI boom.

I believe MIT was trying to get into the data center space because it only had industrial properties in Singapore to begin with. Getting into data centers was a growth strategy, a way to stay relevant in the digital world. But it was never likely to buy data centers in Singapore, where the market was already dominated by Singtel, Keppel, Digital Realty, Equinix, AirTrunk and others. Singapore also had limited power and land, and even imposed a moratorium on new data center construction for a period. Data centers here are simply too precious to sell. That’s also part of why new data centers sprung up in Johor, close enough to Singapore to serve the same demand.

That’s likely why MIT went shopping for data centers in the US instead. The 2021 purchase of 29 properties came with 87.8% occupancy, dragged down by one asset, 250 Williams Street NW in Atlanta, which was only 63.5% occupied. Excluding that single property, the other 28 data centers were 98.4% occupied at the time of acquisition. MIT bought reasonably good occupancy. It just appears to have underestimated how quickly technology and tenant needs would move on.

What’s Next?

The next best move is to sell off the legacy data centers. MIT needs to raise the quality of its asset base. Its Singapore industrial properties remain valuable even without the data center angle, so those are clear keepers. It’s the data center side that needs the cleanup. Better to have a smaller data center exposure that’s actually AI relevant than a bigger one that isn’t.

Selling opens up another challenge though. Where does MIT deploy the proceeds? Buying more Singapore assets is unlikely. Even Keppel DC REIT had to look to Japan. Still, I find it hard to beat premium Singapore real estate, so the more likely path ahead may simply be to sell and return the proceeds to shareholders as dividends.

Join our Telegram to stay updated on the latest investing news: https://t.me/realDrWealth