AI Changes Everything for Data Centres

Goodman Group results last week included an interesting one-liner from the CEO. You get these one-liners occasionally – like when the Nvidia CEO said last year that they can’t make enough GPUs.

The GMG CEO said – “The amount of leverage that’s going into the sector, the amount of debt, is next level extraordinary.”

GMG has been in the data centre business for 20 years, but it is accelerating. Another one-liner. He says they are at the start of a 10-year growth story.

GMG are in the middle of building 500 MW of data centres in Australia and Europe, including one in Sydney, and the major barrier to entry is now access to power.

How much is 500 MW? I had to look it up. When talking about data centres, they talk about megawatts, which is not a measure of electricity generation like a power plant. In data centre parlance, it is the power capacity that tells you how much computing power a site can support, which tells you how much revenue it can potentially generate.

For example, a data centre might rent out space to a colocation (colo) client (the client owns the servers but GMG provides the building, connectivity, cooling, power, security) at $150 per kW per month. So 1 MW earns around $1.8 million a year.

So 500 MW would earn $900 million annual recurring revenue, or $1.2bn at $200 per kW. Margins are usually around 60 to 70% once built. You can do the numbers. And this revenue is recurring. Good business… if you can build it.

The world’s largest data centre operator is Equinix with 250 MW capacity under management. So 500 MW is big, and GMG are clearly targeting the hyperscalers who are landing Woolworths as a customer in a shopping centre. Hyperscalers pay more.

In the “old days” (pre AI) – three years ago (!) – a rack of business use servers used around 5–10 kW per rack. The new AI training clusters (using Nvidia chips) use 30–80 kW per rack. Some use 100 kW per rack. AI tenants like OpenAI, Anthropic, Meta, etc., will pay $200–300 per kW per month for liquid-cooled AI-ready racks. At that price, 500 MW could earn GMG $1.8bn in revenue. AI has flipped the economics for data centres – it needs fewer racks, higher density and allows far higher rents. 70% of data centre demand comes from the top 10 hyperscalers.

 

The Top 3 Hyperscalers

Amazon Web Services (AWS) – The largest – spends over $100 billion annually on data centres – has ~32% of global cloud market share. Largest cloud provider and the largest data centre customer in the world. Has its own data centres and leases from Digital Realty, Equinix, CyrusOne, and Goodman. Has bought twice as many Nvidia chips as anyone else. Expanding globally.

Microsoft Azure – Number 2. Global footprint. Expanding rapidly.

Google Cloud (Alphabet) – Spending $85 billion in capex in 2025. Top 3 global tenant. Builds its own data centres and leases from REITs.

 

Other AI Tenants

Meta (building for AI and social media), Oracle Cloud, Alibaba Cloud (largest hyperscaler in Asia), Apple (1,500 MW of global data centre capacity), Tencent Cloud (Chinese), Huawei Cloud (China-backed), CoreWeave (AI-specialist cloud provider), VAST Data, Cerebras Systems (ultra-fast inference hardware dedicated to AI), IBM Cloud, DeepSeek (China – lower-cost), xAI (Musk’s AI group), OpenAI (massive AI developer now building trillion-dollar-class data centres and infrastructure), Inflection AI, Anthropic (AI lab leasing high-density AI-ready data centre space), Mistral AI (France), SoftBank / Stargate Project ($500 billion Stargate project with OpenAI and Oracle).

Other customers for data centres include business and government – some of the biggest are US financial firms (Goldman Sachs, JPMorgan, etc.) and government and defence contractors, plus corporates like Netflix, Zoom, Epic Games, Tencent.

 

The Long Game in Data Centres

AI server demand is doubling every year. Power is the new bottleneck for data centres. That makes it hard to enter the market.

It is amazing – here we are worrying about share prices on a daily basis – meanwhile, the companies we are worrying about are building out infrastructure on 10, 20, 50-year horizons. They simply couldn’t care about the share price today. Maybe we shouldn’t either. The Man in the Moon, objective, obvious recommendation, is “Buy on weakness for the long term”.

GMG and NXT are the best data centre exposures in Australia and, as such, offer two “AI plays”. Which is why we hold GMG and NXT in the Marcus Today Growth Portfolio.

 

Get the Full Picture

Find out how GMG and NXT fit into our Growth Portfolio and why data centres are one of the most powerful long-term growth stories.

Sign up to Marcus Today

 

AI Summary of Data Centre Exposures on the ASX

NextDC (ASX: NXT)

• Overview: Australia’s largest listed pure-play data centre operator.
• As of 2024, operates 13 data centres across major cities including Sydney, Melbourne, Brisbane, Perth, and Canberra.
• Raised AUD 1.32 billion in 2024 to accelerate expansion in Sydney and Melbourne.

---

Goodman Group (ASX: GMG)

• Overview: Industrial REIT pivoting aggressively into the data centre sector.
• Working on a AUD 10 billion data centre development program, backed by a massive AUD 4 billion equity raise.
• Knight Frank explicitly cites Goodman as a key ASX player to capture data centre growth.

---

Infratil (ASX: IFT)

• Overview: Infrastructure investor with diversified holdings.
• Owns 48% of CDC Data Centres, the largest private data centre operator in Australia and New Zealand.
• CDC is expanding rapidly, valued at ~AUD 17 billion after major institutional investments.

---

Macquarie Technology Group (ASX: MAQ)

• Overview: Formerly Macquarie Telecom Group – provides telecoms, cloud, colocation, and data centre services.
• Operates five Intellicentre data centres in Sydney and Canberra, and is building the AI-ready IC3 Super West facility.
• Its facilities are certified under the federal government’s strategic hosting framework.