On December 16, 2025, Google signed two energy deals that reveal more about the future of AI infrastructure than any product launch.
In India, it agreed with ReNew to enable a 150-megawatt solar project in Rajasthan, purchasing the project’s energy attributes under a long-term arrangement. In Malaysia, it signed a 21-year power purchase agreement with TotalEnergies to supply its data centres with 1 terawatt-hour of renewable electricity from a solar plant in Kedah.
Looked at separately, the deals appear routine. Taken together, they signal something more strategic: Google is securing the right to expand in markets where electricity, not compute, is becoming the binding constraint.
Power, not processors, is the new bottleneck
Global data-centre electricity use reached about 415 terawatt-hours in 2024, roughly 1.5% of global demand. By 2030, the International Energy Agency expects consumption to approach 945 terawatt-hours, driven largely by AI workloads.
The issue is not the global total. It is where and how fast demand shows up.
Data centres can be designed and commissioned in two to three years. Power systems cannot. Generation, transmission and permitting typically move on far longer timelines. The result is a structural mismatch: hyperscalers can scale compute far faster than grids can scale supply.
In Asia, that mismatch is now visible.
India: solar abundance, grid friction
ReNew’s Rajasthan project is expected to generate roughly 425,000 megawatt-hours per year and come online in 2026. ReNew has been explicit that Google’s long-term procurement is essential to making the project bankable.
Rajasthan’s appeal is obvious: land, sunlight and scale. But the constraint is not generation alone. It is transmission, timing and local acceptance.
India’s data-centre electricity demand is projected to rise from about 0.8% of total power demand in 2024 to 2.6% by 2030, growing far faster than overall electricity consumption. Cooling adds another pressure point. Estimates cited by Reuters suggest a 1-megawatt data-centre load can require more than 25 million litres of water per year, intensifying scrutiny in already stressed regions.
Clean-energy contracts help hyperscalers demonstrate intent. They do not, by themselves, solve the physics of peak demand on coal-heavy grids. India’s clean generation share is rising, but fossil power still dominates marginal supply.
The Rajasthan deal therefore functions as both capacity enabler and political insurance.
Malaysia: concentration risk meets long-dated contracts
Malaysia’s challenge is not scarcity of projects but density of demand.
Government estimates suggest data centres could require 19.5 gigawatts of generation capacity by 2035, equivalent to more than half of Peninsular Malaysia’s current electricity use. To manage that surge, the country plans significant additions of gas-fired generation even as it tries to rein in coal reliance.
Against this backdrop, Google’s 21-year PPA with TotalEnergies stands out less for its size than for its duration.
The contract operates under Malaysia’s Corporate Green Power Programme, a virtual PPA structure. Electricity still flows from the national grid; the renewable contract reshapes the economics and investment incentives behind it.
This distinction matters. During demand spikes, fossil generation still fills the gap. Coal’s share of Malaysian power has risen at times precisely because data-centre loads have grown faster than clean supply.
The value of a long-term PPA, then, is not exclusivity of electrons. It is predictability—for developers, regulators and investors.
Why power contracts are becoming competitive armour
These deals matter because sustainability has quietly become a gatekeeping mechanism.
In Malaysia, authorities have already rejected a significant share of proposed data-centre projects that failed to meet energy or water standards. In India, grid congestion, land use and local resource politics increasingly shape what gets approved.
At the same time, hyperscalers are operating under stricter internal standards. Google’s 24/7 carbon-free energy target requires matching electricity use with clean supply on an hourly, local basis—a far higher bar than annual renewable offsets.
That combination—regulatory scrutiny outside the fence and granular carbon accounting inside it—turns long-term power access into a strategic moat.
The uncomfortable reality
Asia’s data-centre boom is still leaning on fossil power.
Solar PPAs do not run servers at night. Virtual contracts do not eliminate peak-hour emissions. Governments are adding gas capacity because reliability demands it.
Even Google’s industry-leading efficiency—its global fleet operates at a power usage effectiveness of about 1.09, well below industry averages—does not negate absolute growth in electricity demand.
The transition is therefore not clean versus dirty. It is fast versus slow, and planned versus chaotic.
What comes next
Expect site selection to flip.
Instead of asking where data centres can be built, hyperscalers will increasingly ask where power systems can absorb them. Renewable potential, grid headroom and regulatory credibility will matter more than latency alone.
Google’s December deals illustrate this shift. They de-risk clean-energy finance while de-risking future expansion. They are less about virtue signalling than about operating permission.
In the AI economy, the decisive advantage may not be the model or the chip.
It may be the quiet certainty that the next megawatt will be there—without breaking the grid.
