Underneath that buildout sits a problem that almost nobody is talking about. Every one of those data centres needs power, and not just any power: AI training clusters run at near-100% utilisation, 24 hours a day, seven days a week, for weeks at a time. That is not a load you can run on intermittent solar. Without something firm sitting behind it, you default to gas. In markets like India and Indonesia, you default to coal.
At ArkTerra Partners, we ran the modelling. On a business-as-usual grid, that 24.2GW pipeline is tracking toward 166 million tonnes of CO2-equivalent per year by 2030. That is equivalent to nearly a fifth of the global aviation industry’s annual emissions, generated by data centres alone.
Not just an emissions problem
I want to be careful about how I frame this, because framing it as an emissions problem makes it sound abstract. It is not abstract. It is a stranded asset problem. It is a regulatory risk problem. And it is a water security problem, all of which affect project returns directly.
These data centres are essentially being financed today on coal and gas grids that assume a 25-year operational life. Projects financed today could be non-compliant or physically constrained by regulatory standards that come online after investment decisions are made.
Australia’s government has already published formal expectations for data centre and AI infrastructure development, with emissions intensity standards under discussion. If we bring it closer to home in Southeast Asia, Malaysia’s Johor state, the region’s biggest data centre hub, has told developers to defer water-cooled expansion projects until mid-2027 and has halted approvals for the highest water-use facility tiers.
When we look at the water security issue, ~9GW of AI-ready capacity running on legacy cooling systems can consume ~18 billion litres of water per year. Meanwhile, dry seasons in Southeast Asia are growing longer. India’s data centre zones in Rajasthan are sitting on aquifers approaching depletion. Water is not a climate talking point here; it is a project viability constraint right now.
Why gas keeps on winning
Energy storage, in integrating variable renewable energy (VRE) generation and adding operational flexibility, is among the enabling technology solutions to help solve the energy side of the problem. We can decouple data centre emissions and the growth of AI from emissions growth.
However, gas remains the incumbent due to straightforward technical and market dynamics.
Hyperscalers sign 10- to 25-year power purchase agreements (PPAs). They need 99.99% uptime. Solar and wind are cheap, but their generation is variable. Unless it can be made dispatchable, for example, using energy storage solutions, you cannot guarantee uptime.
Meanwhile, gas peakers in APAC operate at a marginal cost of around US$30/MWh in markets with access to cheap domestic gas. They are already built. Grid operators dispatch peakers first when renewables are unavailable.
Every data centre signing a renewable energy PPA without storage sitting behind it is, in practice, signing a renewable energy PPA with implicit gas backing.
Coal retirement timelines are also extending. We already see that in India and Indonesia. Coal plants scheduled for closure from 2030 to 2035 are now getting utilisation rates from data centre demand that make early retirement economically impossible. Each new data centre campus built without energy storage extends that lock-in of fossil fuel reliance.
In the US, power producer and energy holding company AES and Google signed an energy storage-backed PPA for the search engine giant’s northern Virginia campus. AES will supply round-the-clock carbon-free energy sourced from 500MW of owned and third-party renewables, targeting roughly 90% hourly matching, in one of the world’s largest data centre clusters.
In Australia, AGL worked with energy storage technology provider Fluence (of which AES was a co-founder and remains a stakeholder) to build one of the country’s largest grid-scale battery energy storage systems (BESS) at the site of a retiring gas-fired power station in South Australia.
The technology works. The contract structures exist. The question is why APAC is not replicating them, and the answer is not technology readiness. It is a project development capital gap at the pre-feasibility stage.
Three deals that indicate what the technology stack might look like
Three significant deals signed in the past few months tell the story more clearly than any projection:
In April 2026, Meta signed an agreement with US startup Noon Energy to reserve up to 1GW/100GWh of ultra-long duration energy storage capacity for its data centres. The deal starts with a 25MW pilot project scheduled for completion by 2028, scaling to the full 1GW supply arrangement once the initial phase is complete.
Noon’s technology, a carbon-oxygen battery built around a reversible solid oxide cell, uses carbon and oxygen: no scarce metals, no lithium supply chain exposure and the company touts it as a 100-hour, or multiday, energy storage solution.
That contract underwrites a new technology that is not commercially proven in the market, and underwrites the OEM’s capacity to build out manufacturing to serve that need and to raise the required capital.
Also in April, Bloom Energy expanded its partnership with Oracle under a master services agreement for up to 2.8GW of on-site fuel cell systems, with 1.2GW already contracted and in deployment. The proposition here is speed: Bloom delivered a fully operational system to Oracle in 55 days against a 90-day target.
In early June, Google signed what it described as a first-of-its-kind agreement with virtual power plant (VPP) operator Voltus. Under a three-year Bring Your Own Capacity (BYOC) arrangement, Voltus will aggregate up to 100MW of distributed energy resources (DERs) annually within PJM, including batteries, smart thermostats, and other flexible assets, into a Google-funded virtual power plant.
Google is the first hyperscaler to sign onto the BYOC product. The significance is not the 100MW. It is the commercial structure: a large energy consumer funding distributed capacity that gets accredited in a wholesale market. That template is directly transferable to APAC grid markets once the storage and demand response ecosystem matures.
Three deals. Three different approaches to the same problem: firm power for AI infrastructure without grid dependency. None of them are yet happening at scale in APAC.
The gap is project development capital
The technology exists. The offtakers: NEXTDC, DayOne, the hyperscaler campus teams, have the problem. What is missing is a structured pathway to get first-of-a-kind enabling technology projects from pre-feasibility to signed offtake in APAC markets where the project development ecosystem is thin.
As told to Andy Colthorpe, editor, Energy-Storage.news.
Energy-Storage.news publisher Solar Media (part of the Informa Group) will host the Energy Storage Summit Asia 2026 on 1-3 July at QSNCC, Bangkok, Thailand. The conference takes place during ASIA Sustainable Energy Week 2026 (ASEW), the region’s most influential platform for driving clean energy. For more information, visit the official website.
Event advisory board member Pavina Adunratanasee will moderate the session From Growth to Maturity: Turning ASEAN Storage Pipelines into Bankable Portfolios, and take part in How Could Southeast Asia Capitalise on Industrial Loads, Recycling, Data Centres & EV Charging? as a panellist.
About the Author
Pavina Adunratanasee is founding partner of ArkTerra Partners, a catalytic finance advisory that mobilises capital to take the energy transition’s enabling technologies from technically proven to commercially bankable.