ABB: AI’s power swings are an engineering problem the Gulf can design out from the start

Reports from the United States and Europe have put a technical problem at the centre of the debate over AI’s energy use. According to reliability specialists and grid bodies in those markets, the data centres that train AI models do not draw electricity in a smooth line. They surge and fall away within seconds, and operators there have found those rapid shifts can wear out the batteries, generators and cooling systems inside the facilities and disturb the grids they connect to.
So far, the accounts are drawn largely from older power systems abroad, not from the Gulf region. But the Uptime Institute, which advises operators on reliability, reports that the strain could appear in data centres in the Middle East as well, which makes it a question worth getting ahead of for a region adding capacity at pace.
Nader Tahan, a product marketing director in technology leader ABB’s Electrification Distribution Solutions division, says the alarm in the global coverage misreads the problem.
“The headlines make this sound like a crisis, but it is really a well-understood engineering problem,” he says. “AI training loads do not draw power smoothly, they surge and fall away in milliseconds, and equipment specified for a steady load struggles with that.”
The Gulf is in the middle of significant data centre growth. In the UAE alone, data-centre power demand is forecast to roughly double by the end of the decade, pushed by developments on the scale of the Microsoft and G42 Stargate project in Abu Dhabi. Each new hyperscale facility adds a load that behaves less like a building and more like a small city.
The solution, Tahan argues, is not simply generating more electricity: “The answer is not more raw power. It is power quality: designing the protection and switching layer to absorb those swings before they reach the servers or the grid. Digital switchgears accompanied with Grid Solutions such as Grid-scale BESS can do exactly that, sensing a disturbance and acting faster than any person could.”
That equipment exists elsewhere too. What differs, in his view, is when it gets installed. In older markets, engineers are threading stabilising technology into grids and facilities built decades ago, a slow and awkward retrofit. In the Gulf, much of the infrastructure is new, so the components that absorb these swings can be specified from the outset.
“The real advantage this region has is timing,” he says. “Most of our large data centres are being built now, so we can engineer that stability in from the start rather than retrofit it later.”
The problem is real, and the industry is racing to solve it. But the reporting that has made headlines elsewhere reads differently from a region still laying its foundations. Treated as an engineering brief rather than an emergency, the volatility of AI becomes a specification to design around, not a threat to fear.