Keppel DC Reit

[weijian]

KDC REIT's preference offering was ~150% oversubscribed, with nearly as much excess applications as the valid acceptances. Not a surprise considering the share price is consistently ~10% above subscription price.

KDC REIT is trading >NAV and <4% dividend yield, an outlier among S-reits. What more, many of its assets have relatively short leases and benchmarking it to other industrial REITs (which also have short leased Sporean assets), that is even more impressive. The difference is because KDC REIT has been able to do its financial engineering magic - using "expensive equity" to buy assets that give relatively high return of capital + return on capital. That is a very lethal combination.

VBs used to warn about KDC REIT been expensive. But if we look back at its share price over the last few years compared to other deeply discounted-high dividend yield S-reit peers, it has probably outperformed. So, we could argue that "been expensive" is probably KDC REIT's real moat.

Virtual Dialogue with Securities Investors Association (Singapore) (SIAS) for the proposed acquisition of Keppel DC Singapore 7 (KDC SGP 7) and Keppel DC Singapore 8 (KDC SGP 8), and the proposed entry into new master lease agreements and new facility management agreements in relation to Keppel DC Singapore 1 (KDC SGP 1) and Keppel DC Singapore 2 (KDC SGP 2)

UL: How do you see distribution per unit (DPU) panning out in the near future with this acquisition, and in the long run?

AL: We will see immediate proforma accretion to DPU, which will stabilise at approximately 7%, assuming 100% interest in KDC SGP 7 and KDC SGP 8, land lease title of approximately 25.5 years and with tax transparency being obtained. We have delivered healthy reversions in the past two quarters and remain optimistic about delivering continued value for our Unitholders.

https://links.sgx.com/FileOpen/Transcrip...eID=827855

[weijian]

I used to be affected by the "short leases" on Spore industrial land but then I figured out that even without a 20-30year lease, the specs/fitting of the industrial property would probably be largely obsoleted after 20years. So even if one had bought a freehold industrial property, unless it is in a land scarce geography, else it wouldn't be too much of a difference with a leasehold industrial property.

MINUTES OF THE EXTRAORDINARY GENERAL MEETING (“EGM”) OF THE UNITHOLDERS OF KEPPEL DC REIT

HH, a unitholder, sought clarification about the 25-year land tenure for Keppel DC Singapore 7 and Keppel DC Singapore 8. CEO replied that the duration of the land tenure was consistent with other industrial land, which were generally granted for 20 to 30 years. Towards the end of the land tenure, an occupant will generally be able to apply to the authorities for a lease extension. CEO highlighted that similar arrangements were in place for other Singapore data centres within Keppel DC REIT’s portfolio, and that extensions have been obtained by the Manager in prior instances.

https://links.sgx.com/FileOpen/Keppel%20...eID=830771

[weijian]

This isn't looking very good as even Keppel Ltd's very own Christian Tan got a thumbs-down in the KIT AGM this year. Looking at Keppel DC REIT's past AGM results, the dissent has been building momentum since the AGM in 2022 (year end FY21) and finally the lead independent director has been voted down this year. So this is not a day1 issue and has remained "unresolved" for some time. What exactly is happening?

Some of Keppel DC REIT's institutional investors are not happy and I suppose such a re-election failure is one of the first for Temasek-linked companies. Keppel wants to be an asset manager and having satisfied institutional capital shareholders are critical to its reputation to keep attracting more LPs for its private funds.

Keppel DC Reit lead independent director Kenny Kwan fails to get re-elected at AGM

Low Huan Ping is appointed as director; other resolutions are passed with more than 90% in favour

https://www.businesstimes.com.sg/compani...lected-agm

Prof Mak's linkedin post shedding more light: https://www.linkedin.com/posts/yuen-teen...88480-awjG

[weijian]

Pretty informative to reduce my personal ignorance on the different type of DCs...

MINUTES OF THE ANNUAL GENERAL MEETING

Next, VT had a query about GV7 DC. Using it as a point of reference, he asked whether it was possible to switch a DC’s purpose from AI-training to AI-inference. CEO provided colour on the various purposes that a DC could be utilised for. He went through key distinguishing factors, including how DCs purposed for AI-inference and cloud uses would prioritise low latency, whereas DCs used for AI-training would prioritise cheap availability of utilities and land. In summary, the locality of a DC and its accompanying resources/attributes, including connectivity and latency, is relevant to its eventual purpose. CEO further shared that new releases of AI-deployment models such as Deepseek might underlie demand for AIinference DCs, which requires low latency to bridge end-users (likely based in urban centres). As Keppel DC REIT owns assets which are situated in highly-connected urban centres, CEO opined that these DCs should be able to continue servicing cloud and AIinference clients.

JT, a unitholder, sought the Manager’s comment regarding (i) the possibility of an AI-bubble; and (ii) Microsoft’s plans to pullback on DC leases in Europe and US. On these observations, CEO noted that these remarks were made in the context of oversupply in AI-training DCs. On a forward basis, technological developments suggest that AI-inference DCs may become more relevant and in demand, whereas pullbacks may be more applicable to AItraining DCs. CEO then reaffirmed that Keppel DC REIT does not have exposure to AI training DCs.

AGM MoM: https://links.sgx.com/FileOpen/Keppel%20...eID=845489

[weijian]

The acquisition of this 2 data centers makes up ~20% of the KDC Reit's asset size and so it is pretty huge but when your share price is trading at ~30% above NAV, the financial engineering work is made much easier despite higher yen borrowing rates.

The equity fund raising (private placement) is also done at a low-to-mid single digit discount to last traded price, reflecting market appetite for the hottest asset class in the world now!

Keppel DC Reit buys two freehold data centres in Japan for US$1.1 billion

[SINGAPORE] Keppel DC Reit and Keppel on Tuesday (Sep 1) said that they have indirectly agreed to collectively buy 90 per cent stakes in two Tokyo data centres for about 171 billion yen (US$1.1 billion).

The two parties will thus jointly own nearly all of Tokyo Data Centre 4 and Tokyo Data Centre 5, which are freehold, hyperscale fully-fitted co-location data centres in Inzai City in Greater Tokyo.

Upon completion of the sale in the fourth quarter of 2026, Keppel DC Reit will hold an 88.62 per cent effective interest in each data centre; Keppel, through its interest in Keppel Japan KK, will hold a 1.38 per cent effective interest.

The existing operator, an “established global data centre owner and operator”, will retain a 10 per cent interest in each asset.

The Reit manager also plans to raise at least S$600 million through a private placement on Sep 10. The 280.1 million new units will be issued at between S$2.096 and S$2.142 apiece. This represents about a 2.5 per cent and 4.6 per cent discount, respectively, to the volume-weighted average price of all trades on Monday.

https://www.businesstimes.com.sg/compani...-1-billion

[swakoo]

... there is a risk trailing every data centre valuation today: the threat of quickly becoming outdated.

The prevailing AI narrative demands liquid cooling and massive rack densities that simply did not exist when older colocation assets were built. Upgrading them requires serious capital.

That is the tension in buying colocation assets during an AI boom. The Tokyo assets are fully fitted, entirely leased and cash-generating. That makes them financeable, but it also means they are not the frontier of AI infrastructure.

Keppel DC Reit’s Tokyo drift: The easy yen trade is over

In the various Keppel DC announcements, I couldn't find any mention about the power and cooling capabilities of the 2 acquired Japan DCs. Whether they are Ampere (2020), Hopper (2022), Blackwell (2024) or Vera Rubin (2026) capable. The power and cooling requirements are quite significantly higher with each new generation of GPU.

However, Nvidia CEO did say their GPUs are fungible and durable.

[BlueKelah]

The acquisition of this 2 data centers makes up ~20% of the KDC Reit's asset size and so it is pretty huge but when your share price is trading at ~30% above NAV, the financial engineering work is made much easier despite higher yen borrowing rates.

The equity fund raising (private placement) is also done at a low-to-mid single digit discount to last traded price, reflecting market appetite for the hottest asset class in the world now!

Keppel DC Reit buys two freehold data centres in Japan for US$1.1 billion

[SINGAPORE] Keppel DC Reit and Keppel on Tuesday (Sep 1) said that they have indirectly agreed to collectively buy 90 per cent stakes in two Tokyo data centres for about 171 billion yen (US$1.1 billion).

The two parties will thus jointly own nearly all of Tokyo Data Centre 4 and Tokyo Data Centre 5, which are freehold, hyperscale fully-fitted co-location data centres in Inzai City in Greater Tokyo.

Upon completion of the sale in the fourth quarter of 2026, Keppel DC Reit will hold an 88.62 per cent effective interest in each data centre; Keppel, through its interest in Keppel Japan KK, will hold a 1.38 per cent effective interest.

The existing operator, an “established global data centre owner and operator”, will retain a 10 per cent interest in each asset.

The Reit manager also plans to raise at least S$600 million through a private placement on Sep 10. The 280.1 million new units will be issued at between S$2.096 and S$2.142 apiece. This represents about a 2.5 per cent and 4.6 per cent discount, respectively, to the volume-weighted average price of all trades on Monday.

https://www.businesstimes.com.sg/compani...-1-billion

i think cloud computing  reveenuee is good but if datacente is for AI use then quite risky. so new placement is lower than current price is it DPS acrrtive for current holder.

but main issue is that bond rate gonna rise in Japan might affeect Keppel ngatively.

[weijian]

@swakoo,
The sellers of this 2 Tokyo data centers were not named in the Keppel DC REIT announcements but in other reports, GIC/Equinix were identified as the sellers. So if we were to search the news for this partnership, we will find that the partnership was formed in 2020 and the data centers were completed in ~2021. As such, we can roughly estimate the "generation" of their cooling/power specs.

@BlueKelah,
AI data centers can generally be classified into "training" and "inference" usage. The former "training" requires much higher specs and also customized layout/configuration, but the latter "inference" may be easier to reconfigure for other usages. If AI data center supply severely overshoot future demand, it is not impossible for this excess to affect cloud computing. For example, a hyperscaler may decide to give up some of its cloud computing leases when it comes due, and shift its cloud computing needs to its own idle data centers that had been set up earlier for inference. And also let's not forget that many of the existing "cloud computing enabled" data centers are also used for inferencing. So as AI-specific computing capacity progressively becomes available, lease demand for these co-location needs may reduce as well.

[BlueKelah]

@swakoo,
The sellers of this 2 Tokyo data centers were not named in the Keppel DC REIT announcements but in other reports, GIC/Equinix were identified as the sellers. So if we were to search the news for this partnership, we will find that the partnership was formed in 2020 and the data centers were completed in ~2021. As such, we can roughly estimate the "generation" of their cooling/power specs.

@BlueKelah,
AI data centers can generally be classified into "training" and "inference" usage. The former "training" requires much higher specs and also customized layout/configuration, but the latter "inference" may be easier to reconfigure for other usages. If AI data center supply severely overshoot future demand, it is not impossible for this excess to affect cloud computing. For example, a hyperscaler may decide to give up some of its cloud computing leases when it comes due, and shift its cloud computing needs to its own idle data centers that had been set up earlier for inference. And also let's not forget that many of the existing "cloud computing enabled" data centers are also used for inferencing. So as AI-specific computing capacity progressively becomes available, lease demand for these co-location needs may reduce as well.

yeah cloud computing is the profitable one, and i see a good future for robotics. I feel human is at the stage we can advance robotics, and most of the excitement is actually not true AI, like self-driving cars/trucks, more like automation. 

but to control real learning AI is too hard now. thats why u see anthropic  banning users using AI with malicious intent. and also the US gov banning overseas use of the Claude 5 advance models that can be used for hacking/digital attacks/etc..

anyways I only have small stake in this just to monitor whats happening, still waiting for that eventual crash to come like in 2008 to deploy more into REITs at high yields.