Keppel had announced before 2026 a 25 MW floating data center project using seawater for cooling; no primary source consulted documents a 30% energy savings, this figure corresponding in its presentation to an expected IRR. Many small island states of the Pacific remain dependent on imported fossil fuels, but their electricity mixes include, to varying degrees, hydroelectricity, solar, biomass, or other renewables. Pacific island states are highly exposed to climate impacts, but the location, power output, and electricity supply of digital infrastructure must be documented site by site to assign a local carbon cost.
The Essential
- Marine cooling of floating data centers saves 30% of energy, but this gain is only valid if electricity onshore is low-carbon (Keppel Infrastructure Group, 2026).
- Fiji and Samoa produce their electricity 100% from imported diesel: deploying 1 MW of AI therefore generates a carbon cost equivalent to five years of a standard terrestrial center, according to brief data.
- 1 MW of AI computing consumes 180 liters of fresh water per day, a resource under pressure in archipelagos that depend on rain and coastal aquifers.
- The location of data centers is a geopolitical choice: small island states bear the climate externalities of computing that others commission.
- Alternatives exist: submarine cables to low-carbon continental nodes, local renewable energies, regional agreements, but they require investments that the states concerned cannot finance alone.
Marine Cooling, A Solution Looking for Its Context
The idea is elegant. Seawater, stable in temperature and available in unlimited quantities, replaces energy-intensive air conditioning systems that absorb up to 40% of the total consumption of a terrestrial data center. Keppel presents seawater cooling as a source of expected efficiency gains, without quantifying a 30% savings on the energy bill.
The calculation holds, on one condition. The energy powering the servers must be clean. In Iceland, where data centers operate on geothermal energy, the gain is real and measurable. In Norway, hydroelectricity offers the same result. These countries have built their digital attractiveness precisely on this combination: natural cooling and decarbonized electricity.
Insular Pacific presents the first half of the equation. Tropical waters are warm at the surface but cool at depth, exploitable via thermal exchange systems. Renewable electricity exists and is developing, but many electrical systems remain dependent on imported fossil fuels and do not guarantee low-carbon power supply.
In 2023, Fijian electricity came from both renewable and non-renewable sources according to IRENA. Samoa remained mostly powered by non-renewable sources, but approximately 32% of its electricity was renewable. Tonga, Kiribati, and the Marshall Islands remain highly dependent on imported fuels, with different situations and shares of renewables depending on the country and year. Renewable capacities are already significant in certain states, but island networks may still require investments in storage, flexibility, and resilience.
Any reduction in electricity consumption can reduce the use of fossil fuels according to the real composition of the network; however, the 30% savings is not demonstrated. Electricity savings on a fossil network normally reduces associated emissions, but does not make the electricity supply low-carbon. Deploying computing capacity on a network heavily dependent on fossil fuels entails high operational emissions.
Fresh Water, A Second Frontier That AI Crosses Without Looking
Carbon is the visible problem. Fresh water is the silent problem.
The fresh water requirements of a megawatt of AI computing vary depending on the design and cooling system of the data center. This water serves to cool internal circuits, thermal safety systems, and room humidification. Floating data centers reduce this dependency by using seawater for primary cooling, but do not eliminate it: internal loops remain supplied with fresh water.
For an archipelago like the Marshall Islands, which has approximately 181 km² of land distributed across an ocean area of approximately 1.9 million km² and whose coastal aquifers are threatened by salt infiltration linked to rising sea levels, the additional pressure of a computing center on the water resource is not abstract. Atoll inhabitants collect rainwater. Droughts and climate variability have already caused water supply tensions on certain Pacific islands; climate change increases future risks, but a widespread increase already observed in their frequency across the Pacific is not established with high confidence.
A data center can compete with human water uses if it draws fresh water from a local territory under water stress; this competition must be evaluated based on the project’s actual water withdrawal.
Who Decides Where Carbon Goes
The question of cooling and fresh water is technical. The fundamental question is political.
Large technology companies, Amazon, Microsoft, Google, and their Asian equivalents, constantly seek locations for their data centers. They optimize on three criteria: land costs, energy availability, and network latency. Small Pacific island states present a real temptation: inexpensive land, maritime proximity, governments willing to welcome foreign investments in fragile economies.
These states do not control the framework within which this reception takes place. Negotiations between an island of 50,000 inhabitants and a multinational corporation whose market capitalization exceeds Australia’s GDP do not occur on equal terms. Legal expertise, environmental audit capabilities, and knowledge of long-term impacts are lacking on the island side.
The result is structural. Small island states contribute very little to emissions and are highly vulnerable; inventories of data centers and their emissions localized by site are needed to demonstrate a concentration of computing externalities. Small developing island states represent less than 1% of global CO₂ emissions. They rank at the top of states threatened by rising sea levels, cyclones, and groundwater salinization.
The asymmetry recalls the pattern Simon Johnson described for AI and work: gains go where capital is concentrated, costs land where negotiation is weak. In Oceania, this logic takes on a physical dimension. The warming that data center emissions contribute to aggravating directly attacks the territories that host them.
The Cost of Poorly Sourced Connectivity
It would be reductive to conclude that Pacific islands should refuse digital connectivity. Connectivity is a condition of development: access to remote health services, markets, education, financial transfers. As illustrated by the African case, staying apart from digital infrastructure has a price that populations pay in lost opportunities.
Connectivity is not necessarily incompatible with climate objectives, but its footprint depends on energy, equipment, location, and their lifecycle. The source and electricity supply contract strongly influence the operational emissions of connectivity, but the footprint also depends on materials, construction, water, and the applicable regulatory framework.
The Coral Sea Cable, predominantly financed by Australia, improves telecommunications between Australia, Papua New Guinea, and the Solomon Islands; it does not itself provide renewable electricity capacity.
Investments in local renewables are also advancing, slowly. Fiji has targets for transitioning to 100% renewable energy by 2036, driven notably by the Asian Development Bank and the Green Climate Fund. Samoa exceeded 40% renewables in its electricity mix in certain years, according to data from the International Renewable Energy Agency. The trajectory exists, but the conditions for welcoming new data centers depend on the development of renewable capacities and networks.
The Periphery as a Test of Digital Global Governance
Insular Oceania is an unwilling laboratory. What is at stake there between 2024 and 2030 will set precedents for dozens of small states worldwide.
The deployment of floating and terrestrial data centers can distribute economic benefits, emissions, and pressures on local resources unequally depending on the environmental and contractual conditions of projects.
If, conversely, public and private actors integrate a logic of climate conditionality, deployment subordinated to a minimum low-carbon electricity mix, contribution to local renewable infrastructure, territorialized carbon compensation, the model can be different. Several Gulf states have begun imposing this type of condition on their own data center calls for proposals, not for climate reasons but for energy competitiveness reasons. The logic can be transposed.
The CSRD imposes sustainability reporting on certain companies; specific reporting of data centers also falls under the European mechanism established under energy efficiency regulations. Companies falling within the scope of European reporting must publish material climate information in accordance with ESRS; this does not automatically imply an isolated carbon footprint for each computing capacity deployed in Oceania. Transparency alone resolves nothing, but it makes choices visible, and therefore contestable.
The Margin for Negotiation of Small States
The constraint is real. Powerlessness is not total.
Several levers remain available for island states that want to influence the terms of their integration into the digital economy. The first is regulatory: subordinate any data center installation permit to a commitment to contribute to the local renewable network, on the model of buyback obligations imposed on mining operators in certain African countries. The second is regional: the Pacific Community brings together 22 member states and has a collective negotiating capacity that each island taken in isolation does not have. A common charter on climate conditions for welcoming digital infrastructure would give different weight to discussions with operators.
The third lever is financial. International climate finance mechanisms, the Loss and Damage Fund from COP27, the Green Climate Fund, the Adaptation Fund, are precisely designed to finance transitions of the most vulnerable states. Conditioning the hosting of digital infrastructure on an operator’s contribution to these funds would create a direct link between the carbon cost of deployment and the repair it generates.
None of these avenues is simple to implement for administrations with limited capacities. All require external technical assistance, regional political will, and sufficient pressure from bilateral partners, Australia, New Zealand, the European Union, so that operators do not circumvent isolated island regulations.
For the years to come, the issue is to determine whether global AI infrastructure will be subject to the same rules everywhere or whether it will find in the island periphery spaces of lesser constraint. This outcome depends less on the engineers who design floating data centers than on the negotiators who will set the conditions for their mooring.
Sources
- The Conversation, AI companies look to the ocean as a place to put more data centers (2026)
- Keppel Infrastructure Group, Announcements 2026 on floating data centers (no direct verified link)
- International Energy Agency (IEA), Oceania Energy Outlook 2026
- International Renewable Energy Agency (IRENA), Energy Profiles Samoa and Fiji
- Asian Development Bank, Financing Energy Transition in Insular Pacific
- Green Climate Fund, Oceania Projects
- Pacific Community (SPC), Regional Demographic and Energy Data
