### Conclusion While a power loss at a CDC Data Centre can have significant implications, the robust infrastructure and redundancy measures in place are designed to mitigate impact and ensure that services remain operational. These measures include UPS systems, backup generators, redundant power feeds, data replication, and comprehensive incident response plans. However, the efficiency of these systems can vary, and there may still be brief service interruptions or impacts depending on the specific circumstances and the effectiveness of the data center's contingency planning
AI/HPC is not resilient when it comes to power and so the market will stratify according to reliability of connectivity and power. Renewables without battery infrastructure does not lend itself to AI/HPC.
It may be the case that LLM analysis takes place in locations where power is reliable and cheap, and then deployed closer to where they are required.
From my AI app: ### Conclusion The resilience of AI systems to data center power loss is achieved through a combination of redundancy, data replication, distributed computing, and robust backup systems. By leveraging UPS, backup generators, cloud solutions, and specific AI resilience strategies like checkpointing and stateless architecture, AI operations can maintain continuity and data integrity even in the face of power outages. These measures ensure that critical AI functions remain operational, mitigating the impact of data center power losses.
From my AI app: ### Conclusion The resilience of AI systems to data center power loss is achieved through a combination of redundancy, data replication, distributed computing, and robust backup systems. By leveraging UPS, backup generators, cloud solutions, and specific AI resilience strategies like checkpointing and stateless architecture, AI operations can maintain continuity and data integrity even in the face of power outages. These measures ensure that critical AI functions remain operational, mitigating the impact of data center power losses.
What does it all mean?
Redundancy means you have more than you need in case you lose some.
Data replication means data redundancy.
Distributed computing means computing redundancy.
Robust backup systems means robust redundancy.
UPS is power backup.
Backup generators are power backup.
Cloud solutions mean something in a data centre or centres.
Checkpointing is computing redundancy (staged analysis/compute, so it can resume from the most recent point of failure)
Stateless architecture means separating the state from the compute, a form of compute redundancy.
The AI is trying to convince you that all of this redundancy will mitigate power loss. All fine and dandy in theory, but redundancy is expensive.
Most of this will be resolved in time.
Google, Microsoft and Apple already do a lot of this for their own compute but even with colossal budgets to throw at these problems, they still have plenty of outages.
New technology and developments always come with risks. Refinement takes time. Going back to the aviation theme, the first commercial passenger jet, The comet, had horrendous accidents but over the years jet reliability and safety have improved. No doubt AI systems will have hiccups.
New technology and developments always come with risks. Refinement takes time. Going back to the aviation theme, the first commercial passenger jet, The comet, had horrendous accidents but over the years jet reliability and safety have improved. No doubt AI systems will have hiccups.
"It might take a boom and consolidation to sort this out just like the internet."
This thread seems to have got off track with concerns about ongoing resilience to power outages, affecting IFT. Discussion seems to be mostly those concerned from who don't know about data centre availability.
While there is no doubt, Data Centres are vulnerable to 'long term' power outages, they are super resilient with numerous mitigations in place that are hard to explain to the lay person.
There is no doubt that a business who contractually relies on the "5 nines" 99.999% availability, doesn't have all the fail safe measures in place to guarantee that service.
Ultimately all data centres will need to go to where the power is cheap and reliable, even if it is remote. If not remote, it makes sense for HPC to be colocated with rapid (15 second for example, or faster) curtailment activities such as Bitcoin mining. The overall HPC demand will be more than most people can even imagine, and that's with HPC technology advancement over time. It might take a boom and consolidation to sort this out just like the internet. Yes datacentres have backup which is typically diesel generators onsite but as battery technology improves over time we will see this gradually migrate to battery and with combination HPC-mining you have the best of all worlds.
IMHO IFT remains the best placed risk adjusted public investment on NZX by a long shot. So many tail winds are in evidence, but particularly the AI growth into datacentre deployment, into data infrastructure growth and finally into green renewable energy demand. IFT sits across so many strong thematics / tail winds / growth areas. Beautifully positioned / poised!
Rate of increase in datacentre deployment in both CDC and KAO (not material in KAO but maturing nicely and is only a couple of years off of being very meaningful) is astonishing. In active construction they appear to have more than 140% of existing capacity - that's not a pipeline, that's in active construction. The pipeline of owned sites plus the active construction is 350% increase against the existing deployed operational asset.
Assumption: I am assuming that because Infratil's investment in CDC is below 50%, CDC is not consolidated in the Infratil accounts.
I am framing my 'title question' from the point of view of new shareholders buying into IFT at 'book value' on the FY2024 balance date.
a/ Net Profit as Reported
From AR2024 p80, that is showing that the carrying value of CDC on the Infratil books is $1,403.4m.
The IFT Group share of NPAT, given IFT owns 48.24% of CDC, is 0.4824 x $201.9m= $97.40m
This means Infratil's investment in CDC is being carried on the books at a PE Ratio of $1,403.4/$97.40m = 14.4
Or looking at it another way, an earnings yield of $97.4m / $1,403.4 = 6.9%. (because Australian franking credits are not recognised in NZ, I guess this means that 6.9% is the 'gross earnings yield')
Given the touted growth prospects for CDC, both the PER and yield, as implied by book value, seem very reasonable.
b/ Net Profit as calculated
Muddying the water is slide 27 of the recent Infratil presentation. https://infratil.com/news/infratil-a...-equity-raise/
In that, the bar graph shows EDITDAF for CDC over FY2024 of $271m, of which I presume the Infratil share is 0.4824x$271m = $130.7m
Total IFT balance sheet assets for FY2024 sum to $16,109.9m. So in a 'total asset picture', the IFT stake in CDC represents only: $1,403.4m/$16.109.9m= 8.71% of all assets. The net financing expense over the FY2024 year was $366.7m. 8.71% of that figure is $31.9m. So I take $31.9m as 'allocated net interest' that must be offset against against any profit from holding CDC.
i/ We make an assessment that the depreciation and amortisation of CDC assets is systematically charged over 20 years (buildings will probably be more and computer equipment less) .
ii/ We guess that the CDC assets in operation today earn 10% on the original capital outlay. This means the price of assets on the books is $97.4m/0.1= $974m.
Now, 1/20th of that figure is $48.7m. Assuming 'F' (which is generally reserved for one off non-operational transactions), is zero, we can calculate the implied IFT share of NPAT from the EBITDAF figure quoted as follows:
0.7 x ($130.7m - $31.9m - $48.7m) = $35.1m
Comparison Conclusion
$35.1m is well below the $97.4m of NPAT implied in AR2024 on page 80. This could mean a couple of things.
i/ My estimates for 'I' and 'DA' are wrong - could be, I invite readers to comment on my assumptions OR
ii/ there is a large one off gain F which has increased the CDC profit over FY2024 (AR2024 page 80).
The term EBITDAF does not appear once in AR2024. So I am finding it very difficult to see what is going on here, and whether that $97.4m IFT share of CDC profits has been inflated by one offs. I do find it difficult to assess results when the same company publishes annual results using different yardsticks, for -in this case- CDC. Inconsistent reporting just leaves me the impression that the company that is issuing the results has something to hide. I am in the dark as to whether that $97.4m implied profit share from CDC to Infratil for FY2024 is to any extent sustainable or not.
SNOOPY
Last edited by Snoopy; Yesterday at 09:16 AM.
Watch out for the most persistent and dangerous version of Covid-19: B.S.24/7
If we make an assessment of the depreciation and amortisation of CDC assets is 20 years (buildings will probably be more and computer equipment less)
Hi Snoopy, I appreciate your detailed analysis as always. Just one thought, I am under the understanding that the clients leasing (often only parts of) a data centre own the computers. What a data centre leases out is a building with reliable electricity and cooling for the clients to put there equipment in.
Originally Posted by Bjauck
Reply With Quote
Bookmarks