thanks dragon328, wonder would the cloud providers like VSTECS (with link to AWS) benefit from the AI demand first before the AI DC providers like YTLP....
@probability, I would expect the AI infrastructure (AI data centre) to be ready first before the big cloud like AWS could use it to provide cloud computing services. But it will be more or less the same timing, as I would expect big cloud like AWS will be able to secure customers months before the AI data centre is ready
nice write-up an insightful..... from my perspective SEA Ltd and its Shopee will play a big role in the AI DC build-up alongside the digital banking license that the partnership was also accorded.... not sure how the digital banking plays alongside the DC but Sea Ltd is YTL's first customer for the DC .... whatever the terms are, i can only see upside hereon while having powerseraya as "fixed deposit"... lets hope for the best :-) peace
Yes raymondroy, PowerSeraya still delivers solid earnings like a fixed deposit. Unfortunately the market is full of short term traders and short-sighted investors who are taking the weak Q3 result to take profit and punish the stock.
The long term prospects are very good, the current share price weakness is a good opportunity to accumulate for long term investment.
Otb cant be seen at ytlp forum today. Isnt that a good enough sign of whats coming? With all due respects Dragon328,your timing skills cant compete with Otb.
Dan Zanger, a renowned trader known for his record-breaking returns, emphasizes the importance of caution after a prolonged bull run, even for the best-performing stocks. According to Zanger, when a stock or the overall market has experienced a significant and extended upward trend, the risk of a massive sell-off increases. This is because:
1. **Profit-Taking:** Investors who have seen substantial gains may decide to lock in profits, leading to increased selling pressure. 2. **Overvaluation:** Prolonged bull runs can push stock prices beyond their fundamental value, making them susceptible to sharp corrections when market sentiment changes. 3. **Market Sentiment:** After an extended period of optimism, any negative news or economic indicators can trigger fear and prompt a rapid shift to bearish sentiment. 4. **Technical Indicators:** Indicators such as overbought conditions, divergence between price and momentum indicators, and other technical signs often signal that a correction is imminent. 5. **Institutional Selling:** Large institutional investors may also start selling to rebalance their portfolios, which can exacerbate the downward pressure on stock prices.
Zanger advises traders and investors to be vigilant during these times, employing strategies such as tighter stop-loss orders, reducing exposure to high-risk positions, and being prepared to act quickly to preserve capital.
@dragon328, two years ago you made the bold call that YTL Power would be 10 bagger. This is the best call I’ve ever come across! Thank you,
Since then, data center, which was only a small part of the valuation then, has become prominent. Not to mention the subsequent development in AI Data Center.
However, it’s very challenging to put a value on the DC business. I have a few questions and hope to get your input.
First, let’s start with Hong Leong’s latest report. It has two parts: 1. YTL DC – valued at RM8.4b by assuming 200MW, and valuation extrapolated from Keppel DC REIT 300MW (its latest market cap is SG$ 3.136b or RM11b) 2. AI DC through the 60% owned YTL Communication – valued at RM12.9b based on 25x FY26 P/E
On (1), MIDF also wrote that “Rollout is expected to be gradual with the first 8MW having been commissioned early-May with further additions of 8MW per annum thereafter.”
I understand SEA has committed on 32MW. It will take delivery in stages. It just feels that it will take a few years to fully deploy the capacity.
So the valuation approach by Hong Leong analyst is a bit troubling. In (1), he applied the valuation of Keppel DC REIT (which is already up and running) on YTL DC (which will not be fully deployed at least for a few years).
I’m not sure how he worked out (2). The total value is RM12.9b/0.6 = RM21.5b. Minus the telco business which is about RM1b, he put a value of RM20b on the future AI DC without mentioning the basis.
Hong Leong analyst used MW as the basis for valuation (at 8,400/200 = RM42m per MW). I wonder, for colocation DC (i.e. 1), will lettable floor space be a better metric?
Keppel DC REIT 2023 Annual Report mentions that it has 3.066 million sf of lettable floor space. FY23 core PBT (I ignored FV changes of investment properties) is SG$166m. So it works out that PBT is SG$54 psf (or RM189 psf)
Actually, Keppel DC REIT is not entirely on co-location business. The fully-fitted segment provides better margin. Therefore, the PBT for co-location service is actually even lower than SG$54 psf
According to YTL Power company website, YTL Johor Data Center 1 is up to 72 MW with 535,000 sf. Applying Keppel DC REIT valuation (which is already at the higher end), the PBT is 535,000 * RM189 = RM101m
Even if extend to 200MW though future development, the projected PBT is only RM101m * 200/72 = RM281m Assume 15% tax (there will be some tax incentives), net profit for 200MW is RM239m
Even at 20X PE, valuation is “only” RM238m * 20 = RM4.8b, about half of Hong Leong’s RM8.4b Not to mention my approach also suffers from the fallacy of putting a 20X PE on earning streams which will take at least a few more years to develop.
DCF is a better valuation approach, but there is insufficient details to use that method.
But I see some similarities with property development business, where analysts will estimate the contribution of individual projects. I also observe that typically the projected contribution is around 5% to 15% of the GDV.
Can the same concept be applied here, where the DC business valuation is somewhere between 5% to 15% of the project cost?
I just read the Hong Leong report on 23 Aug 2023. It mentioned “Data centre of 48MW (32MW undertaken by SEA Limited) is guided to provide RM100m PBT when fully commissioned (in stages over 4 years)”
The RM100m PBT is very close to my estimate above of RM101m based on floor space by referencing Keppel DC REIT. The difference is company website mentions 72MW and analysts talk of Phase 1 being 48MW.
@klee, let's put this straight, I am not competing with anyone here in promoting this stock. Mr. OTB has my respect and he just happened to be promoting the same stock.
My intention of releasing this article was to give a full picture of the earnings potential of YTL Power in the long run, and hopefully those who were still holding the stock would not sell off just because of one weak quarterly result. But apparently the market was full of short term traders and investors with very short span of investment horizon.
I am not sure what Mr. OTB has done but it would not bother me, as he has his own followers to guide and answer to. My intention has always to discover undervalued gems and promote long term investment for 2 years to 10 years or beyond.
@observatory, thank you for your comments above, it helps to prompt for more studies into the valuation of data centre businesses of YTL Power.
Firstly, I will try to answer some of your questions on Hong Leong report.
On YTL DC valuation, Hong Leong used the valuation of the consensus target price for Keppel DC REIT to apply to a total of assumed 200MW of co-location data centres for YTL Power. I think it is quite fair as Keppel DC REIT is the only listed entity of regional co-location data centres, which are similar to what YTLP is developing in Kulai with SEA Ltd being the first customer.
Yes it may not be a like-to-like comparison as you said, but we also do not know how much of the ~300MW data centres of Keppel is in operation, how much has how many years of lease left and how much EBITDA margin for each. While YTL DC1 only has 1st phase of 8MW in operation for SEA, but the timeline for subsequent phases of 8MW in 2025-2026 are pretty much fixed. I cannot be sure of the total committed capacity by SEA, some reports say 36MW while others say 48MW. YTL DC1 can actually be expanded to 72MW by putting additional 1MW in each of the 24 halls in DC1, so I would expect SEA to gradually commit more after the 36-48MW.
I agree with you that it is not a good comparison with Keppel DC REIT but that is the only reference right now.
That's the reason I chose to estimate the potential earnings from the co-location data centres and assume gradual commissioning of the data centres up to 100MW in modules of 36MW each year by 2027.
For 36MW of co-location data centre, I assume capex of RM1.5 billion, 80% gearing, project cashflows of 15 years, project IRR of 15%, I get first year PBT of RM119 million rising to RM250m in year 15 as project loan gets paid off.
That's how I got the net profit projection of RM19m from co-location data centres in FY24 rising to RM200m in FY2027 as YTLP secures enough customers for a total 100MW co-location data centres. I feel this better reflects the earnings potential and valuation of co-location data centres for YTLP, which should add valuation to YTLP every year to 2027, rather than based on a lump sum at current year.
As to valuation method using the lettable floor space, I am not too sure if it is appropriate, as each data centre building design is different. For example, YTL DC1 is designed such that each of the 24 halls has space to accommodate 1MW more of data centre equipment if needed. That would immediately add 50% IT load to the floor space (from 2MW each hall to 3MW each hall).
Data centre buildings are not like commercial office space where we typically value it per square foot basis.
DCF valuation is a good method to value the AI data centre business but one needs to have sufficient data for a DCF valuation. The AI data centre is a relatively new business, and hence analysts struggle to come out with a proper valuation using traditional method like DCF or PER.
Hong Leong used a PER of 25x on projected FY26 earnings of AI data centres. CLSA used 3x multiples of sales, or implied PER of 29x with reference to CoreWeave valuation.
Macquarie used DCF valuation for YTLP's AI data centre, by making many assumptions, including 100MW AI data centre in 2026, another 100MW in 2029 and another 100MW in 2032, US$3.42/Hour/GPU of leasing rate etc. The DCF valuation comes to RM50.5 billion for this total 300MW AI data centres, and RM30.3 billion for YTLP's 60% stakes. Again some would argue too high, but it all depends on assumptions made.
After studying the valuation and revenue calculations made by CLSA, Ambank and Macquarie, I decided to make my own calculations of potential revenue and earnings to be contributed by AI data centres to YTLP. You may challenge my calculations and assumptions made, but those are my assumptions based on indications from the company as well as from various analysts.
You can see my assumptions are more conservative than some of the analysts, eg. I assume a leasing rate of US$2.50/hour per GPU compared to US$3.42/hr per GPU assumed by Macquarie and US$3.00/hr per GPU assumed by CLSA.
I have assumed a total of 200MW of AI data centres to be commissioned by mid 2027, while Macquarie has assumed a total of 300MW while CLSA has assumed a total of 150MW AI data centres.
I just received RHB latest report on YTL Power where RHB raised target price from RM4.69 to RM6.68.
In valuation AI data centre, RHB analyst assumed 15x EV/EBITDA, total 60MW of AI data centre only, 14% project IRR, 60% stake and USD3 billion capex to come out with a valuation of RM18.5 billion.
Had he assumed a total of 200MW, the valuation would balloon to RM61.7 billion !
I was sick for a few days, hence I did not post anything in I3. dragon328 is the best analyst in YTLPower, I always consult him when I have any doubt on YTLPower. dragon328 is the person that I respected a lot, he is always better than me. Thank you.
Dear dragon328, do u think profit from powerseraya on coming Q42024 will match back Q42023? Or even break new high? What is your forecast? Thanks for your sharing.
@dragon328, thank you for the very useful explanation. It saves me lots of efforts and possibly dead ends in my research.
Your assumed capex of RM1.5b for 36MW co-location data center (or US$9m per MW) is in line with what I read. The cost for building a green field data center in US is in the range of US$7m to $12m.
I also checked out the RHB report you mentioned which says “we impute a higher AI- DC valuation assuming 15x EV/EBITDA (from 12x previously; still below global peers’ average of 18x) with a 60% ramp up (from 20% ) in 100MW DC, USD3bn capex, and 14% IRR”
The AI-DC capex is US$3,000m/100MW = US$30m per MW.
As I understand (correct me if I’m wrong), the AI DC requires much higher capex because YES Communications will own the servers and sells the service
Next, I tried to do some sanity check. Nvidia list price for H100 is about US$30k. A back of the envelope shows: Power consumption of H100 is 700W. Number of GPU per MW = 1,000,000/700 = 1,429 At USD30k per GPU, the GPU cost alone per MW is US$43m
The GPU cost alone is much higher than RHB’s US$30m per MW assumption. But I suppose YTL Power can get a good deal from Nvidia. Besides prices drop over time. Right?
Can you share a bit more details on your DC and AI-DC valuation, like the CoE, CoD, WACC? Given computing equipment become obsolete fast, what are the replacement capex assumptions? Would you consider sharing your spreadsheet in your next blog? That will be really wonderful.
@Edison Cheah, I expect PowerSeraya earnings for Q4 FY2024 to be similar to the RM700m achieved in Q3 FY2024. I expect the same level of quarterly profit to continue into 2026 then drop to S$600 million in FY2027 & FY2028 or quarterly net profit of S$150m or RM525m.
@observatory, I had also struggled for a while on the capex figure until I read CLSA report on YTL Power. Apparently, not all the power requirement for AI data centre is for powering the GPUs, CLSA report said about 67% was for GPUs and the rest for cooling & others. In my calculations, I have assumed 67.5% of power for GPUs.
In your example above, for a 100MW AI data centre, the required number of H100 GPUs may be 100,000x67%/0.700 = 95,714
Assuming US$30k per H100 GPU, the required capex would be 95,714 x US$30k = US$2.9 billion close to the figure of RHB's US$3.0b.
@emsvsi, the equity IRR would be very high as typically AI data centre developers look to get back all returns within 3 years, as the product life of GPUs is short.
I didn't calculate the IRR but you may refer to CLSA or Macquarie reports for reference
@observatory, I do not have a spreadsheet for AI data centre project, as I only aimed to calculate the potential net profit contribution to YTL Power.
I assumed a product cycle of 3 years, after which YTLP would re-invest in the newest technology chips and lease out for another 3 years. This is similar to what Macquarie analyst has assumed.
This book is the result of the author's many years of experience and observation throughout his 26 years in the stockbroking industry. It was written for general public to learn to invest based on facts and not on fantasies or hearsay....
probability
14,500 posts
Posted by probability > 2024-05-23 18:22 | Report Abuse
thanks dragon328, wonder would the cloud providers like VSTECS (with link to AWS) benefit from the AI demand first before the AI DC providers like YTLP....