Speed Built the Data Centre. Power Decides Whether it Runs.

Siemens Energy’s Steve Loo on what two years changed in APAC’s AI data-centre build-out; the items with the longest lead times, when to retrofit instead of build, and why “just add power” is the costliest misread on the table.

The building goes up fast. The power to run it does not.

That gap between how quickly a data centre can be built and how quickly it can be reliably powered is the one Steve Loo keeps coming back to. Steve is Siemens Energy’s VP of Grid Technologies Sales for Asia Pacific and Managing Director for Thailand, and we put the same questions to him at Data Centre World Asia that we put to other companies building the physical layer under the AI boom. His answers kept returning to a single point: the hard part is no longer the servers. It is the electrons, and the grid that was never designed to deliver them where the AI load now wants to sit.

Speed is No Longer the Whole Story

Two years ago, Steve says, the question was how fast a data centre could be built. It still matters, but it now shares the table with a harder one. “We are seeing very large and concentrated electricity demand coming online, often in locations where the grid was not originally designed to accommodate that level of growth,” he says.

That shift moves the pressure off the server hall and onto generation capacity, grid connections, transmission and system resilience, and onto the calendar. It also changes who needs to be in the room, and when. Developers, utilities and energy-technology providers now have to collaborate far earlier in planning, because the energy infrastructure has to be designed alongside the digital infrastructure, not bolted on after the racks are specced. “As digital infrastructure expands,” Steve says, “energy infrastructure must be planned alongside it.”

What Everyone is Waiting For?

Ask what has the longest lead time and Steve refuses the single-villain answer. “There is no single bottleneck. It depends on the project, the location and the maturity of the local power system.”

The equipment list is predictable enough: large power-generation kit and high-voltage gear (transformers and switchgear) the components that actually connect a large load and expand network capacity. But the real constraint, he argues, is sequencing. A data-centre project is a chain of interdependent steps: permitting, grid connection, construction, skilled labour, equipment delivery. “You can have the equipment available,” he says, “but if the grid connection or permitting is not ready, the project cannot move forward.”

That is also the logic behind Siemens Energy putting money into capacity rather than just quoting lead times. The company recently announced a US$1 billion investment in the US to expand production of grid and gas-turbine equipment and create more than 1,500 skilled jobs. The stated aim is blunt: increase capacity so critical energy infrastructure arrives at the pace customers need. Worth noting for an APAC reader that the factory investment named here is US-based. The regional question is whether that capacity, and the skilled labour behind it, reaches projects in Australia, Indonesia, Singapore, Malaysia and Thailand on the same timeline.

You Do Not Always Need a New Data Centre

One answer will land well with anyone holding a brownfield site and a CAPEX committee. It does not always have to be a new facility, Steve says. Many existing sites already have grid connection, power systems, cooling capacity and footprint worth reusing; the question is whether they can carry future load. “In some cases, upgrading existing infrastructure or adding new power assets may be faster, more cost-effective and less disruptive than developing an entirely new site.” The caveat is a proper assessment of the whole energy system and the grid plan before committing to either path.

That connects to where Steve sees the real waste; and it is not a component. Asked to name the most wasteful part of a data centre, he declines to point at any single technology. The waste shows up when generation, grid connection, cooling and backup power are each planned on their own: duplicated investment, and infrastructure that never gets fully used. “The most efficient outcomes are achieved when the data centre and its supporting power system are considered as an integrated whole.”

The Silver-Bullet Problem

The biggest misconception about energy technology, Steve says, is “that there is a single solution that can solve every challenge. In reality, there is no silver bullet.” A turbine, a transformer, a substation; each is one part of a larger system, and what matters is how generation, transmission, electrical infrastructure and digital technologies work together to deliver reliable power where and when it is needed.

The second misread is more operational, and more common in data-centre conversations: that powering a hall is simply a matter of increasing supply. Supply matters, Steve says, but reliability, grid integration, resilience and long-term system planning matter just as much. None of it, he adds, is something one party solves alone. Operators, utilities, governments and technology providers have to plan it together.

And data centres are not the only ones asking. Steve sees the same demand for reliable, resilient power across manufacturing, utilities and other energy-intensive, digitalising industries. What sets data centres apart is the speed and concentration of the load. AI workloads are creating very large demand on accelerated timelines. The constraint, in other words, is not unique to data centres. They are just hitting it first, and hardest.

The Monday Takeaway

If you are planning a data centre, or weighing the AI load headed for any energy-intensive site, Steve’s read points to three things. 

1. Plan the energy system at the same time as the digital one, not after; bring the utility and the grid question into the room before the design is locked.
2. Before signing off a greenfield build, assess whether an existing site’s grid connection, power and cooling can carry the load. A retrofit may be faster and cheaper. 
3. Design generation, grid, cooling and backup as one system, because the expensive waste is not a bad component, it is four good ones planned in isolation.

This interview is part of Industrial Gemba’s coverage from Data Centre World Asia. Responses have been edited for length and clarity.