In a recent GDCA LinkedIn article, we explored the looming “power bottleneck”, questioning whether electricity grids can keep pace with the data centre surge driven by AI. At the time, the conversation centred on potential shortfalls and the urgent need for new generation capacity.
But as new developments from the US and Europe show, hyperscalers are not simply waiting for the grid to catch up. They are rethinking the way data centres consume, source, and even locate their power, suggesting that the AI boom may be met with smarter, more flexible energy strategies than previously imagined.
Grid Flexibility: Google’s Demand Response for the AI Era
In the US, Google is pioneering a new breed of demand response by shifting or reducing non-urgent workloads, such as certain machine learning (ML) tasks, during times of grid strain. Partnering with utilities like Indiana Michigan Power and the Tennessee Valley Authority, Google has demonstrated the ability to adapt AI processing schedules in real time, helping balance supply and demand without compromising reliability for critical services.
These “flexible demand” capabilities not only speed up grid connections for large loads but also reduce the need for new transmission lines or fossil fuel plants, a critical step as AI’s electricity footprint grows. It is a model that could be replicated globally, though for now, deployment is limited to select sites where utility and infrastructure partnerships are in place.
Reusing the Past: Microsoft and Amazon’s Power Plant Conversions
Meanwhile, in Europe, speed to power is emerging as the defining factor in hyperscaler expansion. Microsoft and Amazon are leading a trend of repurposing decommissioned coal and gas plants into state-of-the-art data centres.
The strategy offers mutual benefits:
- For tech firms, it delivers immediate access to grid connections, water cooling, and existing permits, bypassing the decade-long delays that plague new European builds.
- For utilities, it transforms stranded assets into revenue streams, often through long-term power supply agreements that can help fund new renewable energy projects.
From France’s Engie to Germany’s RWE, energy companies are marketing dozens of such sites, sometimes with direct renewable connections via “energy parks” that keep the grid as a backup. Given the size of hyperscale loads, often hundreds of MWs, these partnerships can generate long-term contracts worth billions, while accelerating Europe’s green energy transition.
Efficiency as a Growth Multiplier
A third prong of this transformation is efficiency, making every watt count. Hyperscalers are embracing technologies like liquid cooling, real-time AI temperature optimisation, and new semiconductor designs to squeeze more computing power out of the same energy input.
Liquid cooling, for example, allows chips to run hotter and faster by drawing heat away more effectively than traditional air systems. Amazon is deploying custom cold plate designs across its fleet, while companies like AmberSemi are tackling power delivery inefficiencies that can waste up to half of the electricity entering a server rack.
The potential is enormous: each 1% cut in energy loss could represent hundreds of millions of dollars in annual savings in the US alone. And with AI model sizes and training demands still climbing, these gains could translate into more capacity without equivalent increases in total energy draw … at least in theory.
The Jevons Paradox Challenge
However, efficiency alone will not solve the energy challenge. History and economics, suggest that improvements often lead to greater consumption, not less. As AI models become more powerful and accessible, demand for the best, most advanced systems will likely keep pushing total energy needs upward.
That is why the industry’s most forward-thinking strategies combine flexibility (Google’s demand response), reuse (Microsoft and Amazon’s plant conversions), and efficiency (liquid cooling and chip innovations). Together, these approaches can help bridge the gap between surging AI demand and the slower, capital-intensive rollout of new clean energy capacity.
A Blueprint for the Gulf
For the Gulf region, where data centre growth is accelerating in tandem with AI adoption, these global trends offer a strategic blueprint. The ability to:
- Secure fast, sustainable power access
- Integrate demand-side flexibility
- Embed efficiency technologies at scale
…could position the GCC not only as a capacity leader, but also as an innovator in sustainable AI infrastructure.
In the race to power the AI future, the winners will be those who can scale without overwhelming the grid, turning “power hungry” into “power smart.”
—
Want more insights on the GCC data centre market? Keep an eye out on our Events page to see where we’ll be taking the next GDCA event, and watch out on our Publications page for the launch of our latest reports.