NXT & AJD - NEXTDC & Asia Pacific Data

[Snow Leopard]

Data Centres:

NextDC build them, sell them to Asia Pacific Data (or whomever) booking a profit, lease them back again and then operate them, hopefully booking a profit.

Just really getting going but having a had run through the accounts I think I can see growth in there so I have spent a few dollars from my play fund (actually my Oz Super Ann!) on NXT recently.

Do Your Own As Usual

Best Wishes
Paper Tiger

[Snow Leopard]

The share price of NXT has risen steeply over the last few days, much to my surprise, and is up a few cents more today.

Best Wishes
Paper Tiger

[Snow Leopard]

NEXTDC:

Investor Update (http://stocknessmonster.com/news-item?S=NXT&E=ASX&N=739027) and A We Landed A Biggie (http://stocknessmonster.com/news-item?S=NXT&E=ASX&N=739025) announcements today.

Just have a look at the update and see what you think.

Best Wishes
Paper Tiger

[Snow Leopard]

This continues to go against the flow on the ASX, setting a new high again today.

Best Wishes
Paper Tiger

[Snow Leopard]

Announcement (http://stocknessmonster.com/news-item?S=NXT&E=ASX&N=743197) today that NXT have sold off their holding in AJD.

They are doing better than expected and need the cash to fund the growth.

Best Wishes
Paper Tiger

[Snow Leopard]

Up 7% to a new high today and with good volume behind it.

It was also announced that a NextDC had done a deal with Dimension Data (http://www.nextdc.com/media/news/588-nextdc-welcomes-dimension-data-to-its-national-ecosystem.html) for DD to resell the NextDC product.

One presumes that the two are in some way connected.

Best Wishes
Paper Tiger

[G on]

I have been waiting for the retail offer book to arrive and so far nothing. Anyone know how I can get the info online? In the past the company website can have the info but I get nowhere.

[mark100]

Try calling the share register, Link Market Services

[Snoopy]

Up 7% to a new high today and with good volume behind it.

It was also announced that a NextDC had done a deal with Dimension Data (http://www.nextdc.com/media/news/588-nextdc-welcomes-dimension-data-to-its-national-ecosystem.html) for DD to resell the NextDC product.

One presumes that the two are in some way connected.


You are a bit lonely on this thread Snowey. It looks like even you have been asleep for ten years as well, so I thought I would drop in.

I see NXT has just had a $1.321 billion dollar plus capital raising?

From the road show presentation for the April 2024 capital raising:
https://www.listcorp.com/asx/nxt/nextdc-limited/news/entitlement-offer-investor-presentation-3017856.html

From slide 10:

1/ Acceleration in demand has grown NEXTDC’s forward order book to a record 68.8MW. (from slide 14, one year earlier, December 2022, the order book was just 8MW)
2/ NEXTDC’s contracted utilisation has stepped up and now exceeds built capacity (from slide 16, 141MW built, 150MW contracted)

From slide 13:
3/ Power density per rack forecast to rise from 36kW (pre AI) to 80-100kW (AI on stream)
4/ AI workloads are power intensive with average power density 2.5x larger than non-AI workloads.

From slide 14:
5/ Historically, the global data centre market has grown at a 15% CAGR (2017A – 2023A) supported by digitisation and cloud migration mega trends. Emergence of AI megatrend and the pace of adoption is forecast to accelerate market growth to a 19% CAGR (2024E – 2027E).




Snoopy Comments

I don't know how accurate NXT's predictions on AI driven demand for datacentres will be. But points 4 and 5 drive home the point that off site processing (e.g. compiling an AI response) takes a lot more energy than 'simple storage'. At least twice as much by the look of things. Now creating an AI reply is 'data creation', not just 'data recovery'. So that answers my question posed on the Infratil thread
https://www.sharetrader.co.nz/showthread.php?1360-IFT-Infratil&p=1061062&viewfull=1#post1061062

where I was interested in the difference between 'data storage' and 'data creation' and the processing requirements of each.

My own unfortunate 'experience' of AI has been through the Inland Revenue department phone systems. Admittedly this is a fairly crude voice recognition decision tree system, where you have to enunciate your problem, then go through various push button options until eventually you get to speak to an operator. But I always find my blood is boiling by the time I finally get through to a human. It would certainly be better if you could ring up, state your problem, and the AI could get you to your destination in one step. I think AI could do this. But given the IRD don't really like you speaking to humans, would they do this?

The main saving grace with the AI we have right now, is that, thanks to technology advancement in the future, today's AI should be the worst AI we will ever have.

SNOOPY

[Snoopy]

From the road show presentation for the April 2024 capital raising:


I love the fine print section on 'investment risks'. It is a legal butt covering section where every disaster under the sun is explained in graphic detail. I can see why it is put in small print in the appendices. Because if anyone who was considering taking part in a capital raise like this 'read it all', they would be so terrified that there is no way they would participate. Consequently NXT would not raise any money.

Nevertheless, it does mean the promoters do have to at least think about what operational matters could go wrong. So in the spirit of being kept informed, here are some notes on 'datacentre risks' from 'Appendix B'.

------------------------

Demand Uncertainty
The market for data centres is characterised by rapidly changing technology and industry trends, frequent new product, and competitor introductions, changing laws (e.g. data and privacy), as well as changing customer demands. NEXTDC is currently exposed to favourable industry trends in relation to data centre outsourcing and co-location cloud provision which remains a key driver of customer demand. However, there are no assurances that such demand will continue or that existing customers will renew their data centre requirements through NEXTDC at all or at the same level as in prior periods.

A reduction in customer demand or increase in competitive supply may have a material adverse effect on NEXTDC’s financial position and performance. This could include a reduction in revenue:

i/ As a result of NEXTDC having to lower product prices in response to changing customer demand or competitive supply OR
ii/ NEXTDC retaining excess capacity over a longer period to the detriment of its return on capital and the subsequent valuation of its investment in its data centre facilities and therefore the value of its securities.

Data Centre Interruptions
NEXTDC promotes its 100% uptime certification and service excellence delivered through its Tier IV and Tier III facilities as a key reason for customers to entrust their critical IT infrastructure with NEXTDC. A data centre or system interruption or outage at any of NEXTDC’s data centres may have a material adverse effect on NEXTDC’s business, operations, reputation and financial performance and therefore the value of its securities.

Termination of Customer Contracts
The majority of NEXTDC’s customer contracts by number are on NEXTDC’s standard terms and conditions. But the majority of NEXTDC’s top 10 customer contracts are based on highly negotiated individual customer contracting terms rather than NEXTDC's standard terms and conditions. This means in practice that the legal terms which govern the relevant arrangements differ significantly from customer to customer. Some of these material customer contracts to which NEXTDC is a party contain provisions which may give the customer a right to terminate the contract under certain scenarios (including for convenience).

Competitive Dynamics
NEXTDC competes against other local and global data centre owners and operators, as well as alternate business models such as traditional on-premises solutions. Similarly, cloud service providers have grown in size and market share, with many having built their own data centres in addition to relying on third-party providers such as NEXTDC. An increasing use of self-built data centres has the potential to reduce demand for data centre services from third-party providers such as NEXTDC. This could lead to decreased revenue and profits for NEXTDC, as well as a potential decrease in the value of NEXTDC’s data centres.

----------------------

SNOOPY

[Snoopy]

From the road show presentation for the April 2024 capital raising:
https://www.listcorp.com/asx/nxt/nextdc-limited/news/entitlement-offer-investor-presentation-3017856.html


There is a lot of speculation attached to the datacentre market, especially about the future. 'Construction cost' is one aspect of datacentres we can feel more confident about, because those numbers are 'in the past'. Yet even then, datacentre providers can be coy about exactly how much money they have spent on various projects. In the interest of transparency, I think it is worthwhile recording in one place (i.e. here) from Australasian sources exactly what we do know 'for sure' about the cost of building a datacentre.





From the Spark Thread (NZX forum)

Post 2257 https://www.sharetrader.co.nz/showthread.php?9630-SPK-Spark-NZ-(TELCO)&p=1058688&viewfull=1#post1058688
2.3MW datacentre opened in 2014 in Takanini, Auckland, - cost $NZ60m, or ($NZ60m/2.3MW=)$NZ26.1m/MW.





From the Infratil Thread (NZX forum)

Post 3602 https://www.sharetrader.co.nz/showthread.php?1360-IFT-Infratil&p=1057693&viewfull=1#post1057693
In 2022, Canberra Data Centres (CDC) opened two 14MW datacentres in Auckland (in Hobsonville and Silverdale), for a total capacity of 28MW and costing 'more than $NZ300m', or 'more than' ($NZ300m/28MW=) 'more than' $NZ10.7/MW

Post 3750 https://www.sharetrader.co.nz/showthread.php?1360-IFT-Infratil&p=1059976&viewfull=1#post1059976.
Future datacentre roll outs are being completed over three years.

Post 3753 https://www.sharetrader.co.nz/showthread.php?1360-IFT-Infratil&p=1060160&viewfull=1#post1060160
Land value of datacentres is typically less than 10% of the balance sheet book value.





From the NextDC April 2024 Road Show presentation (link in referred text above)

Slide 17 'S4 Sydney'. Building works incorporating 10MW built capacity and critical high voltage infrastructure, $A350m, or ($A350m/10MW=)$A35m/MW

Slide 17 'S3 Sydney'
https://www.datacenterdynamics.com/en/news/nextdc-aims-raise-400m-new-build-warns-covid-19-uncertainty/

had an original build cost of $A350m (back in 2020) with an initial capacity of 12MW. 'with a total capacity of 80MW expected by the first half of 2022.' That 'first half of 2022' target was not met, as the green light to expand capacity to 50MW was only given in April 2024. Prior to this built capacity was 13.5MW
https://www.datacenterdynamics.com/en/news/australias-nextdc-acquires-data-center-in-sydney/

The $A400m of S3 'accelerated build capacity' will provide incremental capacity of (50MW-13.5MW=)36.5MW. This equates to ($A400m/36.5MW=)$A11.0m/MW

Slide 17 'M2 Melbourne'. Extending the base building and capacity to 60MW. Some history of 'M2 Melbourne' is here:
https://www.datacenterdynamics.com/en/news/nextdc-secures-new-data-center-locations-in-brisbane-melbourne/

Land to build two datacentres: In Melbourne (M2, initial capacity 6MW) AND Brisbane (B2, initial capacity 1.5MW) was acquired in 2016, at a combined project build cost of $A160m. With costs allocated in proportion to capacity, I get the initial build cost of M2 to be ( 6/7.5 x $A160m =)$128m. This equates to ( $A128m/6MW=)$A21.3m/MW

This article:
https://www.afr.com/technology/why-nextdc-s-m2-is-no-longer-the-problem-child-20200513-p54sh1

suggests M2 was a 'problem child', with only 4MW of capacity being used until 2020, after which customers signed up to another 33MW of new contract options. Now 33MW+4MW=37MW. Three years on is enough construction time to install equipment to fulfill those contracts, plus a bit more. (Edit: This may be correct. But actual capacity built at the April 2024 capital raising time, slide 27 of the presentation, was only 28MW. So there is 32MW to go to reach M2 site capacity of 60MW). The cost of finishing M2 is said to be $A330m. This equates to ($A330m/32MW=)$A10.3m/MW.

The afr article concludes with a description of the business activities of NEXTDC, which provides some clarity on the NEXTDC business model:
"NEXTDC builds hybrid data centres, which cater to both the retail and wholesale customers. Retail data centre deployments tend to be smaller contracts, which rely on the data centre providing additional services like connectivity to cloud providers and on-site staff. Wholesale contracts are those where businesses with large footprints, like banks, rent the space and power but provide their own services."

In case you are wondering how NextDC fits in with the big multinational data centre players, Google Cloud, Microsoft Azure, AWS, Oracle Cloud and IBM Cloud are all customers of NextDC.

From the April 2024 presentation slide 21
"In Australia, cloud service providers have preferred to outsource their IT infrastructure needs to third-party providers such as NEXTDC, as it is typically more cost-effective, and they are able to rely on NEXTDC’s dedicated expertise and resources for data centre development and services."

SNOOPY

[Snoopy]

First observation: The two costings we have for the smaller older datacentre facilities (Spark's Takanini, 2014) and the initial stage of Melbourne M2 (NextDC, 2020) are noticeably higher than present day costs on a 'dollar per Megawatt' basis. I am not surprised by this. Spark Takanini's building works I know were constructed with further 'computer power inside expansion' in mind. IOW there was some 'future proofing' in the initial Takanini build. This may have been true for NextDC's Melbourne M2 site as well. NextDC Melbourne M2 was designed to have more than one story of datacentre based on what would have previously been seen as a 'single data layer footprint'. This made M2 a pioneering design for NextDC. It is the kind of architecture that will be required, as NextDC move forward in their plans to build datacentres in Asia (Kuala Lumpur in Malaysia is already in development) . The ground floor of a civil structure for a multistory building has to be more robust (read expensive to build) than if the building were a straight single story design plan. So it is only natural that the occupiers of the first stage of such a complex have a more expensive structure over which to spread their working computer cost overheads.

Also expensive on a dollar per Megawatt basis was NextDC's S4 Sydney site. But it was made clear that part of the reason for that was the electrical wiring work needed to bring sufficient electrical power to the building site to allow the fully finished development, some years down the track, to operate. It is important to set up a site with sufficient power operating capacity for foreseeable future needs. This point was brought home to me with NextDC's Melbourne M2 site, where the finishing cost per Megawatt of the completed data centre was only around half the establishment cost base.

The three more recent developments:
1/ CDC in Auckland New Zealand (cost $NZ10.7m+ /MW)
2/ NextDC Sydney S3 (last stage cost forecast $A11.0m/MW)
3/ NextDC Melbourne M2 (last stage cost forecast $A10.3m/MW)

are quite close in development cost on a dollar per Megawatt basis. Melbourne M2 I would suggest is the more efficient build, simply because of the smaller footprint made possible by building one data centre on top of another.

SNOOPY

[Snow Leopard]

Hi Snoopy

A little piece about Data Centres & Power Consumption in Oz (https://www.abc.net.au/news/2024-07-26/data-centre-electricity-grid-demand/104140808)

Mentions NEXTDC :t_up:

[Snoopy]

Hi Snoopy

A little piece about Data Centres & Power Consumption in Oz (https://www.abc.net.au/news/2024-07-26/data-centre-electricity-grid-demand/104140808)

Mentions NEXTDC :t_up:

I can see why so many of these datacentres put 'token' solar panels on the roof. For people 'looking over the fence' it gives the impression that the great buzzing box behind the security fence is generating its own power, even if the net effect is that no solar generated electrons will ever travel out of the complex, and plenty of electrons will be required to travel the cables in the inwards direction.

Slide 16 in the April 2024 'entitlement offer presentation', showing 'Built Capacity' and 'Contracted Utilisation' I have reformatted the bar graph into a table, and extended it by forecast below.




NEXTDCBuilt Capacity (MW)Contracted Utilisation (MW)



12/20A89MW71MW


12/21A96MW76MW


12/22A129MW84MW


12/23A141MW149MW


12/24F?MW177MW


12/25F?MW211MW


12/26F?MW251MW


12/27F?MW299MW



The forecast figures, from December 2024 onwards (note 1 from slide 16 in the presentation, are based on a compounding annual growth rate of 19% as estimated in the the 1.451 Research’s DatacenterKnowledge Base, S&P Global Market Intelligence (March 2024) (Refer to investment presentation slide 14). We are talking a doubling of contracted processing capacity over four years.

The money raised in the April 2024 capital raising is earmarked for:

i/ S3 Sydney -accelerated built capacity, $A400m, 36.5MW
ii/ S4 Sydney -Stage 1 development, $A350m, 10MW
iii/ S5 Sydney -Stage 1 development, $A300m, 0MW (building works only)
iv/ M2 Melbourne -accelerated built capacity and base building, $A330m, 32MW

That is a total 'new datacentre pipeline' of: 36.5MW+10MW+32MW= 78.5MW. From this we need to take off the 8MW in the above table 'sold in advance' to leave 70.5MW of new processing capacity on the drawing board.

149MW + 70.5MW = 219.5MW.

219.5MW of capacity will cover the 211MW projected forward demand as a December 2025. But that date is only 18months away. Either NEXTDC do not believe their externally supplied growth assumptions. Or there may be the need for another capital raise for NEXTDC from as early as mid 2025!

SNOOPY

[Snoopy]

increasing

NEXTDCBuilt Capacity (MW)Contracted Utilisation (MW)


12/20A89MW71MW


12/21A96MW76MW


12/22A129MW84MW


12/23A141MW149MW


12/24F?MW177MW


12/25F?MW211MW


12/26F?MW251MW


12/27F?MW299MW




Reflecting on the above table, with datacentre capacity doubling every four years, I do wonder where the workforce to build these new data centres is going to come from. Writing down a table as the one above is the easy bit. But building is a linear profession, where you cannot exponentially increase your output. So my reading of the above table is that NextDC is going to require twice as many builders working for them over the next four years, compared to the numbers they had working for them over the last four years. Is such a ramped up building workforce available? Particularly when datacentre competitors will be ramping up their build programs exponentially over the same time frame?

My other concern over this projected build program is how it fits in with the 'culture' of adopting new technology. I can relate to the idea of a slow burn beginning (refer Slide 16 in the entitlement offer document)
https://www.listcorp.com/asx/nxt/nextdc-limited/news/entitlement-offer-investor-presentation-3017856.html

where demand creeps upwards for two years (12/20 to 12/22) before it suddenly explodes. This is the kind of market behaviour you might expect with one of the megaproviders like Microsoft Azure decides to hit the market. This brings the market to the attention of other megaproviders who suddenly decide: "Hey, we want to be part of that action as well." All at once other megaproviders decide they want a piece of the action too and the likes of AWS, Salesforce, Workday and Servicenow stomp into town on, in this example, Microsoft's coat tails. But after all of the big boys arrive in town, what then? Does the build rate for new data centres continue to go up exponentially? My guess is no, because there will likely be an overbuild of datacentres if everyone believes it is 'easy money'. Post 2027 will likely be followed by a consolidation phase. We also know that as technology improves, more processing capacity will likely be available for less power use. This means quite a bit of future growth may be available by just upgrading existing processing space. Thus I think there is a good chance that after 2027, the exponentially increasing demand graphs, such as those that appear in slide 11, may flatten out. The future focussed graphs (slide 11 of the investor capital raising presentation) offer no comment on the post 2027 datacentre environment. But readers of those graphs will no doubt be expecting the expansionary trends will continue, given that there is no alternative commentary given.

SNOOPY

[Snoopy]

From slide 13:
3/ Power density per rack forecast to rise from 36kW (pre AI) to 80-100kW (AI on stream)
4/ AI workloads are power intensive with average power density 2.5x larger than non-AI workloads.


I have to admit I do not understand the global trend point being made here. I think I get the basic idea. Namely that if you have AI in your operation, the computer has to go away and 'construct stuff' from a database that will then be edited returned to you, the user, in an understandable format. Furthermore this process may be iterative, where you send off sequential AI requests before AI delivers the answer to the question you asked. This process is clearly going to be more 'power hungry' than just putting data in and out of a digital vault. Someone has worked out that 'on average' it is 2.5 times more power hungry.

So when someone says 'power density per rack is forecast to rise' I take this to mean that if normally the number of electrons being shot through the rack is 'y'MW, suddenly part of the datacentre starts running at 2.5y MW, (due to the owner of the equipment on that rack running AI). So suddenly a 100MW datacentre might become a 120MW datacentre (I am assuming the whole thing does not switch over to AI at once). However, when the datacentre was set up it was a '100MW datacentre' and the wiring to the place was sized on that basis. Come a hot day in the city, the office workers turn the air conditioning up and 'kebang' the regional circuit breaker blows.

Can anyone tell me what I am missing in this scenario? Is a '100MW datacentre' actually a '100MW datacentre', or is that figure just an estimate?

SNOOPY

[Snoopy]

Is a '100MW datacentre' actually a '100MW datacentre', or is that figure just an estimate?


A side point on this comment. Slide 27 of the capital raising prospectus on Melbourne M2 mentions the company is 'targetting a PUE' of between 1.1 and 1.28.

What does the term PUE mean? Google says:
"PUE is the ratio of the total amount of power used by a computer data center facility to the power delivered to computing equipment. Data center operators calculate instantaneous PUE using the following components: Lights and utility plugs that are associated with the data center and dedicated mechanical room."

The note behind the 1.1 number says:
"Best instantaneous power consumption ratio within a calendar year, dependent on load and optimal environmental conditions."

So at least the company is acknowledging that load is varying. But it is not clear to me if NextDC are saying that load is varying with climate (maybe the cooling system has to be turned up), usage or both. Oh dear, there is so much I don't know about the datacentre business!

The note behind the 1.28 figure suggests
"Total energy consumption ratio during a full calendar year, dependent on IT load, client design and service agreements and supports a market leading level of energy efficiency."

This is the first time in a datacentre context that I have seen 'client design and service agreements' mentioned. I wonder if there are some contracts akin to the arrangement Tiwai has agreed to on the NZ electricity grid whereby in times of peak load, the demands of certain customers can be reduced? Perhaps back up work for some customers could be suspended, while the grunty demands of AI for other customers are fully satisfied? I don't know, - I am speculating. But such contractual arrangements could make sense for NextDC.

The next comment 'supports a market leading level of energy efficiency' again sounds very sensible. Yet I have never seen it used in the context of datacentres before. I only see comments on capacity in Megawatts, which is the total billable power used by a datacentre unit, with no reference to how efficiently that power is being used to create the saleable output. That has always seemed very strange to me. If a customer were to approach NextDC with a datacentre demand, surely it would make sense for NextDC to satisfy this demand by using as little power as possible? Yet until today, I have never heard the term 'efficiency' ever mentioned. Perhaps this is because there is no incentive to save power?

Scenario: A customer signs up for 1MW of computer power (which is the amount assessed that will be needed to do the job asked). They then get billed for the use of that 1MW monthly at an agreed price (pure speculation, I have no idea how these data centre contracts are priced in practice). So the customer is happy and NextDC gets paid. There is no incentive for NextDC to look at the contract and see if the customer could do the work with less power so that NextDC could charge them less.

So much is opaque to me, on how the profitability level of these datacentres work!

Nevertheless a PUE of 1.28 seems high. It means that 28MW of power is 'wasted' as 'non chargeable energy' for every 100MW of 'chargeable power' able to be delivered. Yet if you believe what NextDC tells us in their datacentre brochure on page 8 this is a good result:
https://23251048.fs1.hubspotusercontent-na1.net/hubfs/23251048/Melbourne_FacilityOverview.pdf

"With a target PUE of 1.28, M2 embraces a level of power efficiency that surpasses industry standards, driving a significantly reduced level of power usage and a more optimal use."

SNOOPY

[Sideshow Bob]

https://thebull.com.au/18-share-tips/29-july-2024/?utm_source=ST&utm_medium=email&utm_campaign=ShareTrader+AM+Update+for+Monday+29+J uly+2024

HOLD – NextDC (NXT)
NextDC is a data centre operator. The company has been benefiting from favourable dynamics, particularly increasing demand for data centre services. How long the tailwind lasts remains uncertain. Until there’s clearer evidence of sustained demand and broader adoption of artificial intelligence models outside major technical hubs, I would be inclined to hold rather than buy or sell the stock.