NextDC And The AI Boom

Australia | Mar 06 2024

NextDC reported a beat on earnings, and its outlook is little more than gobsmacking, driven by AI demand. Can the company pull it off?

-NextDC notes overwhelming AI-based data centre demand
-Expansion plans considerable
-Cost, time, and return on investment in the spotlight
-Not the only player in town

By Greg Peel

NextDC ((NXT)) builds and operates data centres in Australia, but is also in the process of expanding to New Zealand and South East Asia.

In Australia, the company currently operates three centres In Sydney, with two more in planning (known as S1, S2, S3, etc.), three in Melbourne (M1, etc), with fourth in planning, two in Brisbane, two in Perth and plans for Port Hedland, Adelaide, Darwin, and the Sunshine Coast.

In technical terms, a data centre is a big building housing rows and rows of bloody big computers that suck loads of power and generate loads of heat, thus needing to be cooled with more power. The centres store and crunch data – loads of it.

There are different kings of data centres.

Enterprise centres serve the IT infrastructure requirements of corporations and are tailored to that corporation’s needs.

Colocation centres are used by multiple corporations to house computing hardware, servers, and supporting infrastructure, such as power, cooling, and networking equipment, in an off-site location.

Edge centres, are smaller, decentralised facilities that provide compute and storage in a location closer to where data are being generated and used. They are situated near their intended users, allowing for real-time data processing and analysis.

The biggie, however, is hyperscale centres. Hyperscale centres, also known as cloud data centres, are massive, centralised, and custom-built facilities which support primarily cloud service providers and large internet companies with enormous compute, storage, and networking requirements.

Hyperscale centres provide the capacity for AI. That’s why there’s a lot of excitement surrounding them, and why the rush is to build more.

The problem for data centre developers is first you have to buy the land, then the centres take a long time to build and cost an awful lot of money. Once built, capacity is typically leased on long term contracts, particularly for hyperscale, maybe ten to fifteen years, thus it takes a while for return on investment to convert into earnings.

Is it all just hype?

That’s what many thought when Microsoft launched ChatGPT and it seemed science fiction had caught up with reality, driving an AI frenzy on Wall Street, and leading to sceptical comparisons with the dotcom bubble/burst of 1999-2000.

But leading AI chipmaker Nvidia has silenced the critics, posting quarter upon quarter of significant earnings “beats” and blow-away guidance for the next quarter, all driven by demand for its AI chips. That story parlays into a need for data centres.

Perhaps the greatest takeaway for analysts following NextDC’s first half result release, which featured beats on revenue and earnings, was management’s upbeat commentary at the post-result conference call. Barrenjoey was compelled to describe it as “potentially one of the most upbeat calls we’ve attended this reporting season”.

NextDC’s cloud-driven assets have grown at an order of magnitude of some 2.5x each generation, note analysts.

Generative AI demand, however, goes a step further still, with NextDC suggesting GenAI demand will be 3-5x the cloud market. This could see the company fielding orders in the hundreds of megawatts, ie really big, which they could aim to satisfy with colocation or pure hyperscale data centres, or “AI Factories” -- AI-dedicated assets, customised to AI’s needs.

As an example, management noted demand is such that one customer alone could account for all of S4’s planned 300MW capacity, if management was happy to take that concentration risk.

But to recap on what was noted above, Wilsons points out any material contract win would be multi-year, with any scalable earnings contribution many years off.

And there’s the matter of return on that investment. It would make sense that as the data centre market reaches scale, like any new technology, equivalent revenues per contract would diminish over time as prices come down.

Barrenjoey had assumed exactly that in its valuation of NextDC, but management has suggested second and third generation data centre returns could potentially match or even exceed first generation. The company is expecting, for example, M2-M3 returns could meet or exceed M1, with Barrenjoey noting the land for M1 was acquired at premium, perhaps future builds will be more efficient given scale, and the pricing environment remains favourable.

If you build it they will come. Management noted because demand still significantly exceeds supply, it had seen significant improvement in unit pricing for both enterprise and hyperscale customers. Management also noted NextDC has never had a pipeline of size and scale it has now.