How do you build a data centre?

How do you build an AI data centre? Next week, I'll have the pleasure of hosting a fireside chat with @Dan Madrigal, VP of Cloud AI Sourcing, Oracle on just this topic.

This will take place at the W.Media SIJORI Cloud & Datacenter Convention 2025.

• Location: MBS, Singapore.
• Date: 10 July 2025.
• Time: 11:05 am.

But what's the big deal with Oracle? For one, they are heavily into AI data centres - more than what most people realise.

Did you know?

Oracle went into the cloud years later than the other public cloud giants. However, it has grown quickly since, possibly avoiding the initial mistakes of its competitors.

As of Nov 2024, Oracle has 85 cloud regions around the world and is planning to build another 77 - that's more than AWS, Microsoft, and Google combined.

But rather than stick with building yet another IaaS cloud, Oracle invested heavily into high-performance computing.

• Think Oracle Exadata, RDMA over Converged Ethernet (RoCE), and GPUs. Indeed, Oracle was among the first to offer Nvidia's A100 GPUs.
• Oracle Cloud Infrastructure (OCI) was engineered for demanding AI workloads; an OCI Supercluster scales up to 131,072 Nvidia B200 GPUs.

Oracle is powering AI

According to Semianalysis, a highly respected analyst firm, Oracle made a bet at the start of the AI race by signing long-term contracts for massive data centre capacity in the US.

And it paid off handsomely.

• TikTok in the US is hosted in Oracle data centres.
• Oracle partnered with Microsoft to extend the Azure AI platform onto OCI, providing additional AI capacity.
• Oracle has signed long-term multi-GW data centre contracts for AI data centres to serve customers such as OpenAI.

Closer to home, Seimianalysis says the SIJORI region of Singapore-Johor-Batam is the world's second-largest AI data centre hub in the world.

And you guessed it, a lot of that is believed to be running within Oracle data centres. So Oracle knows AI data centres - which are quite different from traditional data centres.

Fireside chat with Dan

The reality is that very few data centre operators know how to design, build, and run an AI data centre. Because so few of them do full-on AI workloads.

But now we have Dan. Here's what I plan to ask him:

• Sourcing components for >= 120kW racks.
• Take away lessons for data centre professionals.
• Challenges around deploying liquid cooling tech.
• Supply-chain strategies to roll out AI data centres faster.

If there's one session you must block off time for, it would be this.

Limited sponsored passes are available for eligible applicants.

Sign up here.