Gen.Ai: Infrastructure Arms Race Powers Ahead

This story features NEXTDC LIMITED, and other companies. For more info SHARE ANALYSIS: NXT

Behind the Gen.Ai megatrend is the growing demand for data centres with major news making events putting the sector back into the spotlight, alongside the adjacent industries being swept up in the infrastructure build out.

-Oracle’s Ellison fires up the data centres debate
-Electricity generation is the key
-Bitcoin miners part of the solution
-Wilsons overweights Gen.Ai stocks

By Danielle Ecuyer

Part 1 was published on September 11: https://fnarena.com/index.php/2024/09/11/gen-ai-update-bubble-licious-or-not/

Part 2: Powering Up Gen.Ai

At the frontier of AI models and the development of the most advanced neural networks (large and specialised language models) lays the home for all the computing power, referred to as the data centre or hyperscaler, when in fact we’re talking about multiple large data centres with cloud computing services.

By now most investors are probably familiar with the concept and the race to build out more capacity to compute. 

It is a well-served topic in the FNArena data base of news stories and more literature on the topic can be found inside the dedicated section on the website with several links at the finish of this article.

But back to the topic du jour or at least what Larry Ellison, the founder, chairman and chief technology officer of Oracle, wanted to talk about in the quarterly results webcast: data centres.

Cloud services have become the company’s largest business. Oracle has 162 cloud data centres in operation and under construction, including one 800MW development which will contain acres of Nvidia GPU clusters for training large language models. 

With the signing of a Multicloud agreement with AWS (Amazon Web Services), Oracle is embedding its latest hardware and software into AWS cloud data centres.

On the webcast Ellison highlighted the company is in the process of planning the development of a one-gigawatt (1GW) data centre adjacent to the proposed site for three small modular nuclear reactors.

Asking the question to Microsoft’s Co-Pilot how big this would be, Co-pilot explained a 1GW data centre campus is planned in Northumberland, UK, consisting of 10 buildings, each with three stories and 54,000 square metres each. Or, about the equivalent of 100 football fields.

Eillison was unequivocal in his outlook for the sector that there is an “ongoing battle for technical supremacy” with a handful of companies and a nation state that would continue for the next five to ten years.

Creating the ultimate neural network is not a finite event, he explained, with ever more refined neural networks, both large and smaller more specialised models for various needs and sectors, such as biotech or finance.

Ellison is far from the lone voice on the topic just in the last week. 

According to the CEO of real estate giant, Related Companies, Jeff Blau, data centres are the “most incredible asset class” and “unique”, in part because of the tremendous demand for power.

While last week’s sale of AirTrunk to Blackstone and CPP Investments for $24bn is the largest data centre transaction on record, preceding the 10 September $750m equity raising announcement from NextDC ((NXT)) to meet upgraded capital expenditure requirements, including the acquisition of the S7 (Sydney) site for more than a 500MW facility.

Wilsons recently shone a light on the data centres industry in a catchy titled report Digital Infrastructure Deep Dive: Cloud Is Great, GenAI Will Be “Gen-sational”!

The broker quoted CEO of Digital Bridge, (specialised US digital infrastructure investors), Mark Ganzi . He forecasts the Gen.Ai market will be around three times the current installed cloud base at around 13GWs to circa 38GWs. With cloud computing expansion comes the need for increased data centre capacity.

Morgan Stanley also weighed into the Gen.Ai infrastructure debate with an aptly titled report “Underappreciated US Powering GenAI Dynamics”.

The main conclusion is there is a burgeoning “severe” shortage of data centre capacity in the US.

Electricity generation the new gold

“Over the next five years, consumers and businesses are expected to generate twice as much data as they did over the past 10 years, with major tech companies expected to invest $1 trillion in data centres. Globally, power demand is forecast to increase at a compound annual growth rate of 14 per cent over the next three years”.

That quote is from Shane Hurst, Portfolio Manager at ClearBridge Investments

Data centres are big news and big money with Goldman Sach expecting data centre power demand to advance 160% by the end of the decade.

The International Energy Agency forecasts data centres’ total electricity consumption could reach 1000 terawatt-hours in 2026 including a 60TWh increase in US data centre demand by 2026.

Share of electricity demand from data centres will increase to 8% from 3% or a 15% average compound rate in the US. As such, US utilities will experience 2.4% compound annual growth in electricity demand from 2022 to 2030 with data centres representing around 90bps of that growth.

Similar trends are in train across Europe and Asia, the above mentioned brokers highlight. Morgan Stanley observes US hyperscalers are increasing investments in Asia with over US$50bn committed. Malaysia is viewed as a “outsized” beneficiary because of its regulatory support, good data security laws and available power supplies.

NextDC’s recent equity raising is expected to help fund land acquisition in Johor, Malaysia for a hyperscale facility and land for a co-location facility in Bangkok, Thailand.

Power supplies on notice

Citi noted “it’s no mistake that some large-scale upstream firms are considering buying utilities”.

As the full extent of data centre demand is laid bare, the corresponding demand for increased electricity generation is a hot topic. None more so than the recent World Nuclear Symposium in London where industry consultants TradeTech observed a record 800 attendees including financiers and delegates from the data centres industry.

Apart from Ellison’s latest comments, AWS recently acquired a data centre campus from Talen Energy for US$650m in Pennsylvania which is directly powered by the adjacent Susquehanna Steam Station, one of the largest US nuclear power plants. The proposed data centre will have capacity up to 960MW.

Crestone: “In our view, the only way AI and net zero can co-exist is if the aggressive lean into AI technologies is matched with an equally aggressive lean into renewables and other low-carbon sources of energy”.

RBC Capital points to 100% renewable energy power commitments for data centres by Amazon, Google, Meta, Microsoft and Apple. There is no industry standard.

Increasingly, RBC Capital sees demand for natural gas growing to fill the gap between renewables and other energy supply sources. Morgan Stanley also believes the market is under-appreciating the potential role of natural gas. This broker proposes data centre developers may choose to locate facilities next to gas fired power generating plants.

ClearBridge makes the case that AI’s demand for power could increase investment in solar at a compound annual growth rate of 16% compared to 8% currently and wind investment up to 31% p.a. from 18% p.a. up to 2030.

Morgan Stanley estimates the carbon footprint for Gen.Ai is greater than the market currently anticipates which will underpin more growth in decarbonisation solutions as 2030 carbon neutral goals need to be met.

Part of the energy mix for data centres, as defined by the broker, is the energy intensity of the equipment and the cooling of the centres.

Breaking down the electricity usage, 40% of demand comes from the server infrastructure, 40% from cooling the data centre and 20% for lighting and storage.

While traditional air cooling has been employed, RBC Capital stresses the newer energy density for Gen.Ai renders air cooling impractical, making liquid cooling more viable.

With the aim for more sustainable data centres, this broker believes demand for efficient liquid cooling will rise.

nVent, a global provider of heat management solutions, is quoted by RBC Capital stating an estimated 5% of data centres use liquid cooling and it is growing three times faster than the legacy technology air cooling.

The broker estimates a total addressable market of US$2.5bn-US$4bn and a compound growth rate of 30% to 40% over the next five years which exceeds the pace of overall data centre investment.

Wilsons points to DUG Technology ((DUG)) as offering potential exposure to this new trend. More details below.

Bitcoin to AI, when only a switch will do

Morgan Stanley has investigated a different solution to the long lead times for data centre developments and potential supply shortages with the conversion of bitcoin mining centres to a HPC (high performance computing) data centre.

The broker believes it is “plausible” and investors are assessing Bitcoin companies to create a REIT structure for bitcoin mining assets to be converted into data centres.

Post the Morgan Stanley symposium focused on crypto and Gen.Ai, the broker believes there is “value arbitrage” for conversion. On the broker’s calculations, a US bitcoin mining facility ranges from US$2-US$3/watt of capacity while the value to a hyperscaler of 3-years of time savings to power up a new data centre would be greater than US$10/watt.

Morgan Stanley understands around 80% of bitcoin company management teams are looking at the opportunity versus around 50% previously.

With a shortage of data centres in 2025/2026/2027 driven by grid connection challenges, converting existing bitcoin mining capacity is viewed as a growing opportunity. A long-term lease structure between a hyperscaler and bitcoin mining firm is envisaged as a possible solution.

Bringing it back home 

Wilson’s focus on Gen.Ai ranks the preferred stocks in Australia.

NextDC is the top pick (pre the recent equity raising) as the purest data centre play and of the highest quality.

FNArena daily monitored brokers have an average target price of $19.975 with a Buy rating equivalent.

Next is Macquarie Technology Group ((MAQ)) as the company’s Macquarie Park is viewed as attractive to global enterprises and governments.

Morgan Stanley has a target of $100 with an Overweight (Buy equivalent) rating. Wilsons is equally positive with a $97.27 target price.

Data#3 ((DTL)) is the third preferred by Wilsons as it partners with global technology companies to procure, deploy and manage IT systems for enterprise and government clients.

FNArena’s average target price is $9.117 with an equivalent Buy rating, one Sell and one Hold. Wilsons has a positive rating with a $9.23 target.

DUG Technology ((DUG)) is a seismic data analysis business for the global oil and gas industry but notably it has approved and pending liquid immersion technology it has been using in its High-Performance Computing for around 10 years. 

FNArena’s Stock Analysis currently shows two Buys and an average target price of $3.495. Wilsons is Buy equivalent rated with a $4.11 target price.

****

More reading on similar topics:

https://fnarena.com/index.php/2024/09/09/nextdc-building-a-generative-ai-future/

https://fnarena.com/index.php/2024/09/05/macquarie-technology-just-a-flesh-wound/

https://fnarena.com/index.php/2024/05/08/rudis-view-opportunity-in-data-centres/

Find out why FNArena subscribers like the service so much: “Your Feedback (Thank You)” – Warning this story contains unashamedly positive feedback on the service provided.

FNArena is proud about its track record and past achievements: Ten Years On

Click to view our Glossary of Financial Terms