Data Centres

AI-driven expansion is concentrating compute into bigger, hotter and more resource‑intensive facilities—evidenced by Amazon’s disclosure of more than 9 billion litres of 2025 data‑centre water use and recent missile and drone strikes that damaged AWS and Oracle sites—creating strategic vulnerability tied to local power, water and transmission infrastructure.

That concentration is forcing industry shifts: developers are rethinking financing as GPUs dominate costs, enterprises are moving toward hybrid/multi‑cloud mobility and rigorous failover drills, and researchers and engineers are pursuing alternatives (from repurposing retired phones to copper cold‑plate cooling) to curb water, power and cooling burdens; watch disclosures, financing models for IT equipment, and mobility/readiness practices.

Tracking: Data Center · Data Centre

Geography: Global, Northern Virginia (Ashburn), Silicon Valley / Bay Area (Santa Clara, Fremont), New York / New Jersey metro, London (Docklands), Amsterdam (AMS), Frankfurt (FRA), Paris, Dublin, Singapore, Tokyo, Sydney, Mumbai, São Paulo, Hong Kong, Seoul, Nordics (Sweden, Finland, Norway), Iceland

1. Amazon reports 9 billion litres of data‑centre water use in 2025

Amazon disclosed for the first time that its global data‑centre operations used about 2. 5 billion gallons (more than 9 billion litres) of water in 2025 in a company blog post.

The announcement comes as AI‑driven growth fuels larger, hotter facilities and governments, researchers and local communities increase scrutiny of data‑centre impacts on water and energy.

Amazon said water consumption at data centres it directly owns and operates fell 2 percent versus 2024 despite an expanding footprint, and it published a water‑efficiency figure of 0. 12 litres per kilowatt‑hour for 2025.

The company compared that figure to Microsoft, Google and Meta but noted caveats: Google’s cited numbers focus on Gemini AI sites, and Amazon’s totals exclude construction and power‑plant water use; industry reports estimate Amazon operates about 924 data centres globally.

Key facts:

- 2.5 billion gallons (~9 billion litres) of water used in 2025

- Amazon reported a 2 percent water use decline versus 2024

- Water efficiency reported at 0.12 litres per kilowatt‑hour in 2025

- Disclosure published via an Amazon company blog post

- Industry reports estimate Amazon operates around 924 data centres

Why it matters: The disclosure signals rising pressure on hyperscalers for transparency as AI capacity grows; public and regulators now have a concrete headline figure to scrutinize.

Amazon’s efficiency metric could shape competitive narratives, procurement choices, and local permitting debates, but comparability is limited because the company excluded construction and power‑plant water use and rivals cited different site mixes.

Watch for follow‑on disclosures from other cloud providers, more granular local water‑use reporting, and regulatory interest in omitted categories (construction and upstream power water).

Local communities, utilities and planners will use these figures in permitting and resilience discussions for future campus and AI‑optimized builds.

Wionews

2. Data-centre builders move beyond real-estate loans as GPUs drive costs

Data-centre projects require far more than land and buildings, yet financing in Asia Pacific has largely defaulted to a real-estate lending template.

The article notes data centres "occupy land, sit inside buildings and generate long-term contracted income streams," but traditional real-estate lending covers a small share of total spend: land acquisition and shell plus MEP represent just 6.

2% of global capital investment, according to Fitch Ratings, while GPUs alone can account for well over half the total capital required.

As a result, developers are increasingly seeking novel funding structures and the piece frames loan-market innovation as a growing source of opportunity.

The gap between property-centric loans and heavy IT/equipment costs is forcing market participants to rethink how data-centre buildouts are financed.

Key facts:

- Land, shell and MEP equal 6.2% of global data-centre capital (Fitch Ratings).

- GPUs can account for well over half of total capital required for data centres.

- Asia Pacific financing has largely followed a real-estate lending template.

- Developers are increasingly looking to novel funding structures to fill capital gaps.

Why it matters: The capital mismatch shifts advantage to lenders and investors who can finance IT hardware and compute-heavy builds, rather than those focused only on property collateral.

Expect to see growth in non-traditional financing solutions and altered lender risk profiles as the industry adapts; monitor loan-structure innovation and terms tied to equipment value and lifespan.

Irei

3. Geopolitical strikes expose cloud fragility, push firms toward hybrid mobility

When reports emerged that a data center in the United Arab Emirates had been hit, it highlighted growing exposure of digital infrastructure to unforeseen risk.

Steen Dalgas, Senior Cloud Economist at Nutanix, argues widespread reliance on cloud services is making organizations re-evaluate assumptions about constant availability; IDC reports 88 percent of organizations are deploying or operating a hybrid cloud.

Dalgas says meeting that requirement pushes enterprises toward hybrid and multi‑cloud architectures, but also increases operational complexity: data gravity, application dependencies, governance, and latency can impede rapid migration.

The article describes a shift from redundancy to "mobility"—regular failover drills, tested recovery plans, and architectures that can move workloads across regions or providers—as AI demand grows and hyperscale capacity remains concentrated (US largest, Europe second).

Key facts:

- A data center in the United Arab Emirates was reported hit

- Steen Dalgas, Senior Cloud Economist at Nutanix, wrote the article

- IDC: 88 percent of organizations deploying or operating hybrid cloud

- Hyperscale capacity concentrated: US largest share, Europe a distant second

Why it matters: Consequence: organizations that can orchestrate multi‑cloud mobility and run regular failover drills gain resilience; those tied to single regions or providers are more exposed.

Expect more investment in cross‑region workload portability, operational testing, and governance tooling, and rising strain on teams managing complexity, data gravity, and jurisdictional controls as AI and concentrated hyperscale footprints increase systemic risk.

Datacenterdynamics

4. Missile and drone strikes hit AWS and Oracle data centers in the Gulf

Missile and drone strikes during regional retaliation reportedly struck three AWS data centers in the UAE and Bahrain and, a month later, damaged an Oracle data center in Dubai, disrupting services used by banks, payment platforms, and other entities.

Iran then declared eighteen major technology companies—including AWS, Microsoft, Meta, Google, and Oracle—as legitimate military targets. Authors Jason Vogt and Nina A.

Kollars argue the rapid expansion of AI-driven data centers and construction of digital megacampuses make these facilities strategic, high-value fixed targets dependent on local power, water, and data transmission infrastructure.

AI facilities differ from traditional centers: training clusters over 100,000 GPUs can require 150,000 megawatts, a single training run could need up to a gigawatt, and that power intensity raises new risks for defense planning and cascading civilian disruptions.

Key facts:

- Three AWS data centers in the UAE and Bahrain were struck.

- An Oracle data center in Dubai was damaged a month later.

- Iran declared eighteen major technology companies military targets.

- There are now more than ten thousand data centers worldwide.

- AI training clusters over 100,000 GPUs can require 150,000 megawatts.

Why it matters: Attacks on fixed data center sites can cascade across banks, payment systems, and civilian services, increasing incentives for hyperscalers, colo providers, and host nations to harden facilities, diversify locations, and bolster contingency planning.

Because AI campuses demand extreme power density and specialized architectures, utilities, grid operators, and defense planners must factor data centers into resilience and national security strategies—especially in regions like the Indo-Pacific where new hubs and megacampuses are changing the geography of critical infrastructure.

Mwi

5. Google-backed researchers explore turning retired phones into mini data centres

Researchers backed by Google are exploring whether thousands of retired smartphones can be repurposed into low-cost, energy-efficient computing clusters, linking them into small, connected data centres.

The idea, reported by Firstpost, is described as exploratory and early-stage, focused on reusing consumer devices for distributed compute rather than building traditional facilities.

The report frames the effort as a feasibility project rather than a deployed service, emphasizing the potential to aggregate many handsets into cooperative clusters.

Details on implementation, timelines, or commercial partners beyond Google support were not provided in the article.

Key facts:

- Project described as backed by Google

- Concept: repurpose retired smartphones into connected computing clusters

- Report says the approach targets aggregating thousands of phones

- Story published by Firstpost, timestamped 25 minutes ago

Why it matters: If technically and economically viable, repurposing retired phones could reduce electronic waste and create low-cost, distributed edge compute capacity.

Watch for pilot deployments, orchestration and lifecycle-management solutions, and announcements about partnerships or use cases that validate performance and reliability.

Firstpost

6. Engineers develop copper cold plate that slashes data-center cooling energy

US server energy use more than tripled between 2014 and 2023 and may double or triple again by 2028, potentially representing up to 12 percent of the nation's grid load.

Almost half of data-center energy goes to cooling and auxiliary systems, so University of Illinois at Urbana-Champaign engineers and Fabric8Labs developed a direct-to-chip pure-copper cold plate with jagged fins.

The plates were shaped using topology optimization to maximize heat transfer while reducing hydraulic losses.

Manufactured with electrochemical additive manufacturing capable of 30–50 micrometer detail, the plates deliver up to 32 percent better cooling and cut pressure drop by as much as 68 percent; researchers estimate cooling costs could fall to about 1.

1 percent of total energy use in a high-density next-generation data center.

Key facts:

- US server energy use more than tripled between 2014 and 2023

- Server energy may double or triple again by 2028, up to 12% of grid load

- Almost half of a data center's energy goes to cooling and auxiliary systems

- Designers: University of Illinois at Urbana-Champaign and Fabric8Labs

- Cold plate uses pure copper, jagged fins, and topology optimization

Why it matters: Lowering both heat transfer resistance and pumping losses directly addresses the thermal bottleneck that limits higher rack densities; operators from hyperscalers to colocation providers could see material reductions in cooling and pumping energy and improved PUE.

Adoption depends on scaling electrochemical additive manufacturing, integrating liquid-distribution and chip-level plumbing, and verifying reliability and lifecycle impacts for large fleets; monitor manufacturability, supply chains for copper, and pilots in high-density AI/HPC deployments.

Sciencealert