The data center trends defining the AI era

A historic surge

In just the past few months, pushed by an insatiable demand for computing power and the rapid ascent of AI, investment in data center infrastructure has shifted to a historic boom all over the world. Cloud giants, hyperscalers, infrastructure operators and nations alike are green-lighting multi-billion-dollar, multi-gigawatt developments at a velocity the industry has never seen before and reimagining the physical foundations of intelligence itself.

The scale is staggering

•  Google has announced $40 billion in AI-driven data center expansion and workforce development in Texas alone, including three new facilities slated for completion by 2027. 
•  Amazon Web Services (AWS) is preparing up to $50 billion in next-generation AI and high-performance computing infrastructure for the U.S. government, bringing an additional 1.3 gigawatts of capacity online starting next year. 
•  Stargate Project, a joint venture created by OpenAI, SoftBank, Oracle, and investment firm MGX, aims to invest up to $500 billion in AI infrastructure in the U.S. by 2029. 
•  Meta’s plan to spend “something like at least $600 billion” through 2028 on data centers and other infrastructure in the US, seems especially ambitious.

Market forecasts underscore this rapid expansion. According to Mordor Intelligence, the global data center market, valued at approximately $386.71 billion in 2025, is expected to reach $627.40 billion by 2030, representing a Compound Annual Growth Rate (CAGR) of 10.16%. In terms of installed capacity, the market is projected to grow from 118.92 thousand megawatts in 2025 to 240.05 thousand megawatts by 2030, at a CAGR of 15.08%. 

Asia's rapid ascent

In Asia, China is obviously one of the key drivers of data center demand, underpinned by rapid AI adoption and large-scale investments supported by government-backed funding mechanisms and policy initiatives. 

The China Data Center Market size was valued at USD 15.81 billion in 2024, and is projected to grow to USD 19.39 billion by 2025. Additionally, the industry is expected to continue its growth trajectory, reaching USD 41 billion by 2030, at a CAGR of 16.15% from 2025 to 2030. 

The Southeast Asia Data Center Market too, shows great growth potential. It was valued at USD 13.71 billion in 2024, and is projected to reach USD 30.47 billion by 2030, rising at a CAGR of 14.24%. 

•  Malaysia has risen to the forefront, benefiting from capacity constraints in nearby markets like Singapore, lower land and energy costs, and a favorable regulatory landscape for data centers. 
•  The Philippine data center market, too, is set for rapid expansion, with its value expected to increase from USD 735 million in 2025 to USD 2.48 billion by 2031, reflecting a CAGR of 22.5%. 
•  As is the Indonesian data center market, with a valuation that is expected to rise from USD 2.81 billion in 2025 to USD 6.08 billion by 2031, at a CAGR of 13.73%.
   

A global race for data centre dominance

Though the Asian market is booming, unsurprisingly, the US remains the world’s dominant data center hub, hosting roughly 38% of all facilities, almost 4,200 facilities. Europe is also a major player, with nearly 3,500 data centers distributed across the region. 

We see sovereign investments everywhere in the world, too, as the AI race will prove to be one of the most decisive in history on a geopolitical level. Elon Musk’s xAI, for instance, is partnering with Humain to build a 500MW data center in Saudi Arabia, set to become one of the most advanced AI compute hubs globally.

An economic shift

Just to give an idea of the power of the data center market: consumer spending typically accounts for about 70% of GDP and is usually the primary engine of economic growth. But 2025 has marked a sharp departure from this pattern. In the first half of the year, the contribution of data center investment to GDP growth matched that of consumer spending, an unprecedented shift:

While consumer spending’s contribution has been gradually declining, investment in data center construction has surged, emerging as an increasingly powerful driver of overall economic activity.

With great growth come great challenges

Exponential growth often does come with some hurdles. At the same time these challenges also present powerful opportunities for innovation and long-term resilience. 

Energy, cooling and efficiency pressures

AI’s unprecedented demand for compute, for instance, is pushing operators to rethink energy use, cooling, and overall efficiency. As global data center electricity consumption reached an estimated 415 TWh in 2024 or about 1.5% of global electricity consumption, this sharp rise is encouraging the industry to reinvent itself on many levels. 

The shift toward liquid cooling, for example, is not just a response to GPU heat output, it’s also an opening for designs that are far more sustainable than the standard infrastructure. Similarly, rising concerns about water consumption are prompting a pivot toward water-efficient systems: from closed-loop liquid systems to heat-recovery technologies that transform thermal waste into district heating. 

Ultimately, the challenges facing data centers today are directly fuelling the breakthroughs that will define tomorrow’s digital infrastructure. 
Companies piloting high-efficiency cooling, grid-aware workload management, and heat-reuse programs are already proving that data center growth and sustainability can reinforce each other rather than compete. 

Policy and public scrutiny

Government scrutiny and sustainability regulation, though often perceived as burdens, are also helping steer the industry toward a healthier, more transparent future. Initiatives like the Climate Neutral Data Centre Pact or targets such as the EU’s 2030 efficiency goals, create incentives for modernisation rather than unchecked expansion. 

Public expectations for cleaner, more responsible infrastructure are further motivating operators to adopt renewable energy contracts, improve power-usage effectiveness, and invest in next-generation cooling. 

Ultimately, the challenges facing data centers today are directly fuelling the breakthroughs that will define tomorrow’s digital infrastructure. 
Companies piloting high-efficiency cooling, grid-aware workload management, and heat-reuse programs are already proving that data center growth and sustainability can reinforce each other rather than compete. 

More efficient software and hardware

The data-center industry’s sustainability strategy extends well beyond power optimisation, increasingly relying on intelligent software, advanced chip design, and AI-driven automation to reduce the energy required per computation. AI systems not only contribute to rising demand but also enhance efficiency through real-time monitoring, predictive maintenance, digital twins, and dynamic power management. 

Examples

•  NTT’s Rhine-Ruhr facility cut cooling energy use by nearly 20%.

•  At the algorithmic level, efficiency has also surged, with Nvidia noting a 100,000× improvement in inference energy efficiency over the past decade.

•  Google reduced Gemini’s electricity use per query by 33× in a single year.

Innovations like Nvidia’s machine-learning system for powering down idle processor cores further trim consumption, while next-generation hardware from initiatives such as DARPA’s AI Next campaign, the European Processor Initiative, and companies including NVIDIA, Intel, AMD, Cerebras, Lightmatter, and Axiado is enabling major reductions in energy per computation. These include custom AI accelerators, neuromorphic and photonic processors, and more efficient AI chips such as Google’s Trillium TPU, which delivers a 67% energy-efficiency improvement over its predecessor.

Cooling Innovations

The thermal footprint of the AI world is one of its most urgent challenges. Cooling accounts for up to 40% of a data center’s total energy use, and the escalating heat loads from dense, GPU-heavy racks are pushing legacy air-based systems to their physical limits. In response, the industry is entering a period of rapid reinvention, exploring both radical new frontiers (that aren’t quite ready yet for scaling, for many reasons) and transformative and highly efficient refinements to established technologies.

Radical experiments (not yet ready for commercial scaling)

Radical infrastructure experiments are pushing the boundaries of what future data centers could look like, from orbital compute nodes to wind-powered modules beneath the sea. While these concepts promise unprecedented efficiency gains, they remain very far from commercial readiness at the moment.

Space-based data centers

Some innovators are looking beyond Earth for radical new approaches to power and cool next-generation infrastructure. Space-based and even lunar data centers are emerging as a bold frontier, driven by ambitions to harness abundant solar energy and the natural cooling advantages of the vacuum.

•  Florida-based startup Lonestar Data Holdings plans to launch the first Moon-based data center dubbed the "Freedom Data Centre"; 

•  Elon Musk recently announced that SpaceX is planning data centers in space, scaling up Starlink V3 satellites into orbital compute nodes; 

•  Google’s Project Suncatcher and Jeff Bezos are chasing the same vision, with solar-fed, vacuum-cooled AI infrastructure beyond Earth;

•  Other global players like Red Hat and Axiom Space, along with U.S. company Starcloud, are also launching initiatives to establish their presence in orbital data centers, highlighting the competitive nature of this emerging field.

Promising as this highly experimental niche may seem, the sector still must overcome some pretty steep technical, economic and regulatory hurdles before becoming mainstream. Challenges like high launch costs, radiation exposure to the hardware, long-term maintenance in orbit, upgrades, system reliability and establishing high-bandwidth communication links with Earth, should be addressed before the concept becomes commercially viable. 

Underwater data centers

Underwater data centers are also emerging as an experimental contender in this area: seawater provides nearly free cooling.

•  The newest project from Shanghai Hicloud Technology in China, a wind-powered underwater data center deployed off Shanghai, recently completed its first phase, claiming a total capacity of 24 MW and a power-usage effectiveness (PUE) below 1.15. 

•  The approach remains challenging, and that helps explain why Microsoft quietly ended its long-running experimental Project Natick in 2024 despite early successes.

Underwater modules are difficult and costly to maintain or upgrade, hardware replacement undersea is logistically hard, and sealed server capsules pose scaling and reliability problems that have so far prevented broader commercial deployment.

Liquid cooling innovation: most (cost) efficient approach

Liquid cooling has emerged as the most promising and effective advancement in data center thermal management: a market projected to grow from $4.9 billion in 2024 to $21.3 billion by 2030. The shift is fuelled largely by AI workloads pushing rack power densities beyond 50 kW, far above the range where traditional air-cooling systems can operate efficiently or reliably.

As a result, liquid cooling is becoming essential not only for sustaining peak computational performance but also for ensuring long-term reliability and enabling more sustainable data-center operations.

Today, leading liquid-cooling approaches fall into two commercially mature categories—direct-to-chip (dtc) and immersion cooling—with an emerging third approach, two-phase cooling, expected to play a significant future role as chip power continues to rise.

Immersion cooling

Immersion cooling submerges entire servers in a thermally conductive fluid, offering superior heat-transfer efficiency for the next generation of ultra-high-density racks that exceed 150 kW. While promising, immersion cooling still introduces pretty steep challenges: significant upfront capital requirements, the need for specialized maintenance expertise, and substantial structural considerations, as filled immersion tanks can weigh up to four metric tons.

Direct-to-Chip cooling

Direct-to-Chip cooling circulates liquids such as water or propylene glycol directly to the heat-intensive components of a server, including GPUs and CPUs, via cold plates. This approach has become the market’s dominant solution due to its technical maturity, comparatively easier integration, and effectiveness for rack densities up to roughly 150 kW.

A strong example of a Direct-to-Chip cooling product is Arteco’s ZITREC® EC portfolio, which includes Propylene Glycol (PG)-based, water-based, and Ethylene Glycol (EG)-based formulations enhanced with Organic Additive Technology (OAT). Engineered for exceptional thermal performance and energy efficiency, these advanced coolants enable components to run at higher speeds without overheating. Their robust corrosion-protection features help extend equipment lifespans, reduce downtime, ease hardware maintenance, and lower the overall cost and complexity of thermal-management systems.

Arteco also offers ZITREC® EC in an innovative ECO version, where fossil-based raw materials are partially replaced with bio-based or recycled alternatives at the start of the value chain using a mass-balance approach. This can significantly reduce the product’s carbon footprint compared with its fossil-based equivalent.

→ Learn how Direct‑to‑Chip cooling is shaping the next generation of energy‑efficient data centers.

Two-phase liquid cooling

Another, still exploratory innovation to keep an eye on is two-phase liquid cooling, a cooling technique where a specially chosen coolant circulates around or directly on top of the chip (CPU/GPU). Rather than simply flowing and carrying heat away as in single-phase liquid cooling, the coolant changes phase, from liquid to vapor, as it absorbs heat. Because of this phase transition, two-phase cooling can absorb and move much more heat per unit volume of coolant than a comparable single-phase system. This makes it a promising candidate for future generations of extremely high-power, high-heat-density GPUs and AI accelerators.

However, it's important to realize that this approach is still highly experimental and needs to mature, with cost, complexity, fluid management, and infrastructure barriers to be overcome. The use of specialized fluorinated coolants in two-phase systems, for instance, introduces significant regulatory hurdles, including scrutiny related to PFAS, F-gas/GWP restrictions, and tighter leak control, which drives adoption hesitation.

As a result, for many data centers and high-density compute clusters, single-phase liquid cooling remains dominant today. Not only because it is simpler and more cost-effective, but also because its fluids are typically easier to source, manage, and comply with from a regulatory standpoint.

Promoted by industry leaders

The accelerating adoption of the liquid cooling approach is propelled by its compelling advantages. Chief among them are substantial energy-efficiency gains: liquid systems can reduce overall power consumption by 30–40% compared with air-based cooling. These improvements translate directly into lower operational costs and smaller carbon footprints while also enabling innovative sustainability strategies.

Because direct-to-chip liquid cooling delivers major gains in sustainability, efficiency, and thermal performance, it is no surprise that industry leaders such as Dell, HPE, and Nvidia are actively promoting its adoption. Nvidia, for instance, emphasizes that as AI models grow and reasoning workloads intensify, air cooling can no longer maintain safe thermal conditions without becoming significantly more energy-hungry and expensive to operate. 

That shift was underscored when Jensen Huang noted at the launch of Nvidia’s new Vera Rubin chips that they can be cooled with water at 45°C (113°F) and thus eliminate the need for water chillers, systems designed to chill water so it can cool air. While this challenges vendors of traditional chiller and HVAC infrastructure, it could prove especially advantageous for two-phase direct liquid cooling, which operates most efficiently at elevated temperatures. Together, these companies argue that direct-to-chip liquid cooling is a foundational technology for the next generation of high-performance, energy-efficient AI infrastructure.