IFT - Infratil

[Toddy]

Thank goodness that round of capital injection is over. I'm not really up for an annual billion dollar cap raise.

Hopefully IFT will have an Airport on the block and maybe funds in the bank from Retire Australia sale to fund the next expansion round.

[kiora]

Thank goodness that round of capital injection is over. I'm not really up for an annual billion dollar cap raise.

Hopefully IFT will have an Airport on the block and maybe funds in the bank from Retire Australia sale to fund the next expansion round.

I can sympathise with & agree with what you saying Toddy.
The cap raise came right out of the blue given it was only a week or so before that they had said they had $800m available for investment.
The size of the cap raise does highlight the investment oppurtunities they see in their portfolio though.
Good for NZ inc given that 40% of their investors are NZ retail investors.
The offshore returns should help the balance of trade returns in the long run .

If we consider our share investments as a business, any interest to pay on money borrowed to take up this offer & stay pro rata is a pretty small proportion of the investment portfolio when considering the dividend returns.

[Snoopy]

It has been pointed out to me that I have made an error in my depreciation estimate calculation. The error being that I have ignored the owned land on which the datacentres are built and onto which any expansions are going to be built. We do have a bit of an obsession in NZ with 'land'. But the vast majority of the CDC land is of course in Australia where most of their business is located.

The interesting thing about datacentres is that when signals are being fired back and forth to them along optical fibres at up to 70% of the speed of light the distance from the point of data use to the point of data storage becomes less relevant. IOW whether the datacentre structure is 5km away or 50km away, the operator on the end is not going to to be able to tell much difference. This means datacentre builders can be a bit less choosy about where they build their structures. They do not have to compete with logistics distribution houses for prime space.

Looking at Googlemaps, it appears the Canberra CDC datacentre there is about 20km out of the city 'in the middle of nowhere'. So I suspect the land bank value of CDC is actually not that high - probably no more than 10% of the value of all CDC long term assets. This means my correction to overall depreciation to take account of land (which does not depreciate) is likely relatively small. Maybe applying to only 10% of the value of the long term assets?

Non current assets at EOFY2023 (the time point used for setting depreciation rates) were $A5,762.3m. 90% of that value (being the non-land assets) amounts to $A5,186m.

The above post is on the side of being 'poorly researched' by my standards. Assuming that land makes up 10% of the long term asset portfolio was just a guess. Can I back it up?

'Oneroof' has an estimated value of 23 Popes Road at Takinini, which is the home of Spark's Takanini data centre
https://www.oneroof.co.nz/property/a...pes-road/JRyhD

The June 2021 estimated value was $29.5m, made up of land $10.2m (c.f. $5.6m in 2017) and improvements $19.3m (c.f. $17.5m in 2017). Interestingly 'Oneroof' records the property being sold as a bare land purchase in 2014 for $4.42m. This would have been the time that Spark bought the property for the development of the first stage of their Takanini datacentre in 2014.
https://www.reseller.co.nz/article/1...ta-centre.html

The reported total development cost was reported as $60m back in 2014 (which would have included the land cost). Now $4.42m/ $60m = 7.4%. That means the land purchase was just 7.4% of the project cost at the time. Since opening, the Spark Takanini datacentre has undergone a multi-million dollar capability upgrade. So as far as book value goes, the value of long term assets at the Takainini site that are land may be as low as 5%.

One can't be sure that such development costs are indicative of the relative long term assets values of a new datacentre portfolio built by CDC today. It looks like land in Auckland at least has shot up in value in relative terms since the Spark Takanini datacentre was conceived. Yet despite this increasing land price effect, I believe this Spark Takanini site data is good supporting evidence that my overall guess of the land part of the CDC long term asset portfolio - being no more than 10% of the total long term assets on the books - is reasonable, and maybe even a little high.

SNOOPY

[Snoopy]

When you knocked down with yet another intellectual 'sucker punch', the only thing to do is to get up again. With the latest information and interpretations flowing through the knowledgometer, time for another crack at this exercise.

We know that Infratil own a 48.24% stake of CDC. Rather than create possible extra confusion by trying to adjust for the Infratil share all the way through, I will use the raw CDC numbers and make any Infratil ownership share adjustments right at the end.

.


Interest Calculation

Interest which is used to finance construction projects is subject to 'capitalisation of interest' in any net profit calculation. If interest is capitalised, it no longer becomes an expense.

Starting from the Infratil capital raising presentation:
https://infratil.com/news/infratil-a...-equity-raise/

Slide 14 shows a key CDC metric is 7-10 times leverage. I will use the larger number (10), because that will give me the largest interest capitalisation effect. A leverage ratio of 10 means $10m of debt for every $1m of equity is CDC's preferred new project 'funding balance'.

If we look at the limited CDC balance sheet information we are given (AR2024 p80) total assets rose from $5,872.4m (EOFY2023) to $6,820.7m (EOFY2024), which is a rise of $948.3m.

Let's assume that all of that increment represents new data centre builds, being constructed over FY2024. Now, new assets are funded by a combination of debt and equity. With a 10:1 leverage ratio, we are looking at 'new project funding units' of $948.3m/11= $86.2m. In this instance this means $86.2m of equity funding, with the balance $948.3m-$86.2m = $857.1m funded by debt. Interest which is associated with this $857.1m in debt funding is interest that becomes capitalised.

As before, I assume total debt, across the year had a representative value of $4,251.1m. Let's say that USPP funding and other bank funding was secured at a rate of 4.5% (a revised pure guess on my part). That means the annual interest charge for CDC would be: 0.045 x $A4,251.1m = $191.3m.

However, this calculation does not take into account the proportion of interest that was capitalised. The amount of interest capitalised represents $857.1m/$4,251.1m = 20.2% of the outstanding debt. 20.2% of the representative interest bill is: 0.202 x $191.3m = $38.6m. So $38.6m is the amount of the pool of net interest that I would expect to be capitalised, following on from all of my assumptions.





Depreciation and Amortisation calculation

My average modelled depreciation time remains at 40 years. The buildings may be on the books with a longer life than that, perhaps 50 years. But all of the cooling equipment will depreciate much faster than 40 years. So 40 years is my compromise 'combined total depreciation schedule number'. Amortisation is assumed to be zero over the period, as no goodwill would have been booked on the balance sheet for a 'green field build' data centre.

https://infratil.com/news/infratil-a...-equity-raise/

From slide 7:
"388MW under construction across current footprint"

From slide 13:
"Operating capacity 302MW", "388MW expected to be completed by the end of FY2026."

So we can think of the 'under construction' section of future data centre roll outs as being completed over 3 years. That averages out at 388MW/3= 129MW to be completed over each of FY2024, FY2025 and FY2026. So we could say the paid for work under construction represents ( 129MW/(129MW+302MW) =) 30% of 'long term datacentre assets'.

Non current assets at EOFY2023 (the time point used for setting depreciation rates) were $A5,762.3m. 90% of that value (being the estimated non-land assets) amounts to $A5,186m. 70% of that non land value (being the depreciable assets that have finished construction) is $3,630m. Assuming these assets depreciate on average over 40 years, this gives an annual depreciation charge of ($3,630m/40=)$90.8m.







iif/ Recalculating Net CDC Profit after once again revising interest charges

According to 'forest' who attended the Infratil capital raising presentation, CDC presently earns between 10-15% on its different assets. So I will 'plumb down the middle' and take my asset return percentage being 12.5% as my estimate. I therefore estimate the income generating ability of those CDC assets on the books at EOFY2024, which I will call EBITDAF, to be: 0.125 x $1,403.4m = $175m. However, it seems that in reality the EBITDAF earning ability of CDC was a little above this lofty goal, coming in at $271m (although this is admittedly a definitively before tax figure. It could be that the return touted by 'forest' of $175m is NPAT).

I now have two operational NPAT figures for CDC over FY2024 to present:

Using NPAT = 0.7 (EBITDAF - I - DA); (assuming the Australian corporate tax rate of 30%)

For no capitalisation of interest charges, I get:
'Operational NPAT' = 0.7 ( $271m - $191.3m - $90.8m ) = -$7.8m, of which the IFT share is 0.4824 x -$7.8m = -$3.3m

For capitalisation of interest allowed, I get:
'Operational NPAT' = 0.7 ( $271m - ($191.3m - $38.6m) - $90.8m ) = $19.3m, of which the IFT share is 0.4824 x $19.3m = $9.3m


I am -still- not guaranteeing that I have got my operational profit calculation right. But even with some more imagined future tweaks, I see no way to turn a $9.3m profit into a (0.4824x$201.9m=)$97.4m profit (the NPAT declared in AR2024 p80). That means the $97.4m that Infratil 'earned' from CDC must be largely (over 90%) from asset revaluations, based on what CDC 'expects to earn' when the current CDC datacentre construction program is complete.

(The actual operational NPAT is adjustable and determinable by the gearing of CDC, which will have been decided upon by Infratil and its ownership partners. It could easily become more profitable by having less debt in its underlying structure. But to avoid paying back some of Infratil's injected capital as a 'tax bill', Infratil have kept the interest bill high. This is what accountants call 'tax efficiency'.)

SNOOPY

[kiora]

Can AI solve the problems it creates?

"AI's power struggle: An energy crisis in the making"


"Considering the broader implications of AI's trajectory, we envisage dramatic consequences for US power demand.

The chances that we need 1 GPU per person is increasing quickly

Combined, search, co-pilots, summarisation, the automation of processes and service offerings of businesses, autonomous driving and gaming alone may increase the amount of GPUs we need over time to somewhere around 1 GPU (of current NVIDIA Hopper quality) per person. While the computational power of GPUs will increase and improve the ability to make computational networks, so will their price and power usage.

In 8 or so years, the amount of electricity needed by data centres is expected to increase to 27% of total US electricity consumption from around 5% today

Implications for Renewable Energy

The implications of the forecast are significant. To have 1 GPU for every 3 people, we need to increase the grid by around another 1/3rd. To put this into context, US electricity demand has been growing by around 0.5% per annum for many decades, but that might ramp 6x to around 3% per annum in just a couple of years. The increase in the number of gas-fired power plants and renewable energy farms needed to enable this technology will be significant relative to how many we have in place today.

None of this demand for electricity accounts for the increasing demand from our need to reduce our fossil fuel consumption. The electrification of cars, engines more generally, other transport (ships, planes and trains) and other equipment that currently uses oil will add to this demand for electricity.

This increase in demand is attributed to several factors, including electrification, manufacturing, onshoring, LNG, crypto, greater industrial loads and data centre growth. Recent advancements in GenAI are compounding and accelerating these factors, leading to the formation of the next power demand super cycle…To be clear, it takes only a fraction of what is being forecasted to be in this super cycle.”

Before these technologies (AI, semiconductors, autonomous vehicles, search, co-pilots, automation of processes, summarisation, gaming to name a few) can reach their full potential, we believe there’s a high chance that we’ll first run out of cheap power."

[kiora]

I see no one has extroplated what the $2+b of available funds to invest means yet

With the leverage of 6x for data centers & renewable energy then could easily be $12+b in new investment ie doubling the AUM that IFT already has.

This is likely to happen over the next,2? years.

With revaluations then what will EBITDA be?

IFT is shovel ready,are investors?

[alokdhir]

I see no one has extroplated what the $2+b of available funds to invests means yet

With the leverage of 6x for data centers & renewable energy then could easily be $12+b in new investment ie doubling the AUM that IFT already has.

This is likely to happen over the next,2? years.

With revaluations then what will EBITDA be?

No wonder Fisher Funds have chosen to sacrifice 3% of their holdings in great MFT to invest in rights of IFT ....both their flagship NZ Growth Fund and KFL did same ...they see more potential in IFT then MFT ...cud be short term play as they can always reverse ahead at some profits ...KFL IFT holdings nearing 20% mark so will need slow pruning ahead ...but yours 80/20 going great guns I reckon

[kiora]

"One advantage working in favour of local data centre operators’ is that “big tech” (i.e., Amazon, Alphabet, and Microsoft) are struggling to navigate local councils, power utilities, costs and complexities of development approvals, etc. This means local operators that have already secured large land banks close to major population centres are in a prime position to service the big overseas players (thank goodness for good old Aussie bureaucracy and red tape!)"

https://www.livewiremarkets.com/wire...E%20AI%20TRADE

[Jaa]

Do Infratil and its band of merry investors have FOMO too?

A team of equity research analysts at Barclays posited that a rush of investments into data centers by cloud-computing service providers appeared to have more in common with “FOMO” — that is, “fear of missing out” — than “Field of Dreams,” a movie famous for the line “if you build it, they will come.”

They highlighted a dichotomy between Wall Street estimates for AI-related capital expenditures and the additional revenue that these investments are supposed to help generate.

While Wall Street expects cloud-service providers to spend an additional $60 billion year-over-year on chips and data centers, they are only expected to reap an additional $20 billion in revenue by 2026.

The Barclays team concluded that, much like companies that laid fiber-optic cable during the dot-com years, the biggest technology firms appear to be overspending on infrastructure.

Of note, the additional capacity produced by projects already under way would be enough to power the existing internet, plus 12,000 new applications with the user base and input demands of ChatGPT, according to Barclays’ numbers. This suggests the plans of hyperscalers to build new data centers are exceeding expected demand.

Wall Street is becoming more skeptical of artificial-intelligence hype helping to power stocks

The comparison to fiber companies during the dotcom boom is the right one in my mind. Anyone remember Ziggy and Reach PCCW?

[maclir]

Maybe. But then they've been in CDC since '16.