AI Data Centers Need More Than Chips: The Power, Water and Infrastructure Challenge Behind the AI Boom

Text and Photo by T&S Editorial Team

Artificial intelligence is often presented as a race for faster chips, more powerful models, and greater computing capacity. Behind that digital race, however, is an increasingly physical problem: AI needs electricity. It needs data centers capable of handling high-density computing, requiring cooling, network infrastructure, substations, transformers, transmission capacity, and backup systems. Depending on facility design and location, it can also place significant demands on water resources.

And the infrastructure challenge does not begin or end with the data center. The semiconductor fabrication plants that produce advanced chips also require substantial electricity, highly purified water, specialized chemicals, wastewater treatment, and other supporting infrastructure.

The International Energy Agency’s latest assessment estimates that global data-center electricity consumption reached about 485 terawatt-hours (TWh) in 2025. Its central projection puts consumption at about 950 TWh by 2030, with electricity use from AI-focused data centers expected to triple during that period. The AI boom, therefore, is not only a question of computing power; it is becoming an issue of energy, water, industrial infrastructure, and environmental governance.

For the Philippines, those questions have become especially relevant as the government advances its participation in the U.S.-led Pax Silica initiative and the proposed technology and manufacturing development in New Clark City, Tarlac. The government now emphasizes that the New Clark City development is primarily an advanced manufacturing park rather than a hyperscale data center complex. Earlier public descriptions, however, placed artificial intelligence much more prominently at the center of the project.

That evolution in messaging deserves examination. More importantly, changing the terminology from an “AI hub” to an “advanced manufacturing park” does not make the underlying resource questions disappear.

AI Runs on Physical Infrastructure

AI may appear almost weightless to someone entering a prompt into a smartphone or computer, but the computing behind that interaction is anything but weightless. Training and operating advanced AI systems requires servers, accelerators such as GPUs, networking equipment, cooling systems, and enormous amounts of supporting electrical infrastructure.

According to the IEA, global electricity consumption from data centers is expected to approximately double from 485 TWh in 2025 to around 950 TWh in 2030, with AI-focused facilities growing considerably faster than the wider data-center sector. The IEA also reports that the expansion of AI infrastructure is increasingly encountering physical bottlenecks, including grid connections, transformers, gas turbines, advanced chips, and other equipment necessary to build and operate large computing facilities.

This creates a fundamental constraint on AI expansion: buying more chips does not automatically create more computing capacity. The electricity has to reach those chips; the heat they generate must be removed; network connectivity must be available; and the local power system must be capable of supporting the load. Furthermore, much of that infrastructure can take far longer to plan and construct than the data-center equipment itself. The IEA notes that data centers can sometimes become operational within two to three years, while the broader power infrastructure needed to support them often requires significantly longer development periods.

Power May Become One of AI’s Biggest Constraints

Data-center electricity consumption remains a relatively small share of total global power demand, but the global percentage can hide important local effects. Data centers tend to cluster, meaning a new concentration of facilities can create major demand within a particular electricity market, transmission area, or community. The IEA expects data centers to account for about 3% of global electricity demand by 2030, but their impact will be substantially higher in some countries and regions.

This matters because electricity supply is only part of the equation. Large power users also need transmission capacity, substations, transformers, reliable grid connections, backup power, energy storage, sufficient generation during peak demand, and infrastructure capable of supporting future expansion. Renewable energy can reduce the carbon intensity of electricity consumption, but a commitment to renewable power alone does not answer all of these infrastructure questions. Solar and wind generation still require transmission, interconnection, and, depending on the system, storage or complementary sources of firm electricity.

For countries seeking AI investment, the question is therefore not simply how much renewable energy could theoretically be produced. A more meaningful question is whether the electricity system can reliably support a large new industrial or computing load without shifting unacceptable costs or risks onto surrounding communities and other electricity users.

Water Use Cannot Be Reduced to One Viral Number

Water use has become one of the most debated environmental effects of AI data centers, and it is also one of the easiest subjects to oversimplify. Not every data center consumes the same amount of water. A peer-reviewed 2025 study by researchers, including scientists from Lawrence Berkeley National Laboratory, found that water consumption at the workload level could vary by more than 10,000 times depending on the technology and operating conditions.

Important variables include:

• Server efficiency and utilization
• Cooling technology and local climate
• Infrastructure efficiency and electricity generation
• The age of computing equipment and the water requirements associated with the local electric grid

This means claims that assign a universal water footprint to every AI query should be approached with caution. The better environmental questions are site-specific: Where will the water come from? How much will actually be consumed, and how much will be recycled? What cooling system will be installed, and what happens during dry periods? Could the facility compete with residential, agricultural, or ecological water needs? And what is the indirect water footprint of the electricity that supplies the facility? These questions require actual engineering and environmental assessments, not generalized assumptions about AI.

The Infrastructure Footprint Begins Before the Data Center

The debate also becomes incomplete when AI’s environmental footprint is reduced entirely to data centers. AI depends on a much larger physical supply chain: Critical minerals → semiconductor manufacturing → AI processors and servers → electrical infrastructure → data centers → networks → AI services.

Semiconductor manufacturing is itself resource-intensive. A detailed environmental assessment prepared by the U.S. National Institute of Standards and Technology (NIST) for semiconductor fabrication facilities states that chip manufacturing consumes substantial amounts of both energy and water. Lithography, etching, and deposition are particularly energy-intensive, while cleanrooms require continuous air circulation, temperature control, and contamination prevention.

Water is especially important, as NIST notes that semiconductor fabrication facilities can consume millions of gallons of water daily, much of which is converted into ultrapure water for processes such as wafer production, wet etching, solvent processing, and planarization. Water is also used for cooling and for pollution-abatement systems designed to remove hazardous gases from manufacturing equipment. While modern fabs increasingly recycle and reclaim water—and some manufacturers have adopted aggressive water-reuse targets—those improvements do not make water planning irrelevant.

Chemical and Wastewater Questions

Semiconductors are sometimes discussed as if manufacturing them simply means assembling tiny electronic components, but the actual fabrication process involves a wide range of specialized chemicals and gases. The U.S. Environmental Protection Agency identifies hazardous air pollutants from semiconductor manufacturing, including hydrochloric acid, hydrogen fluoride, glycol ethers, methanol, and xylene. Semiconductor processes also use high-global-warming-potential fluorinated gases, including nitrogen trifluoride and several perfluorocarbons, which the EPA notes can sometimes pass through manufacturing equipment unreacted and enter the atmosphere without adequate abatement.

NIST’s environmental assessment also identifies manufacturing waste streams associated with solvents, metal plating, acids, and other chemicals that require recycling, treatment, recovery, or disposal. This does not mean that a semiconductor facility automatically generates unacceptable pollution; modern fabs can employ sophisticated treatment, recycling, abatement, and environmental control systems. It does mean, however, that the environmental footprint depends heavily on which specific facilities are built, which semiconductor processes they perform, which chemicals they use, how wastewater is treated, the source and quantity of water, air pollution controls, hazardous waste management and the source of electricity. Those details matter enormously when assessing the environmental impact of a large advanced-manufacturing development.

Pax Silica Brings These Questions to the Philippines

The Philippines formally joined Pax Silica in April 2026. Pax Silica is an international initiative focused on strengthening technology supply chains across areas including critical minerals, semiconductors, advanced manufacturing, and artificial intelligence. The Philippine component has been associated with approximately 4,000 acres (1,600 hectares) in New Clark City.

From the beginning, artificial intelligence featured prominently in official descriptions of the development. On April 20, the Philippine Board of Investments announced a partnership with the United States to establish what it called the first-ever “AI-native industrial acceleration hub,” stating the planned site would host facilities involving critical-minerals processing, semiconductors, advanced manufacturing, and AI. The Philippine News Agency similarly reported on April 20 that New Clark City would host an industrial hub centered on AI innovation and investments, describing it as a Pax Silica “Golden Node”. Government promotion continued to emphasize the AI ecosystem, and by July, the proposed development was still commonly being referred to publicly as the Pax Silica AI hub.

Then the language changed.

From an “AI Hub” to the Clark Advanced Manufacturing Park

On July 23, BCDA President and CEO Joshua Bingcang discussed what was still being reported as the Pax Silica AI hub and estimated that the development could require around 3 gigawatts of power at full development. At the same time, environmental organizations, local communities, and other critics were increasingly raising concerns about impacts on electricity, water, land, and communities.

On August 6, BCDA publicly pushed back against one of the central perceptions surrounding the development. Bingcang stated that the Tarlac project was not a data-center project and said it would instead resemble manufacturing economic zones producing electronics, chips, and semiconductor components. During the August 14 Senate inquiry into the environmental and social implications of the development, the physical project was increasingly referred to as the Clark Advanced Manufacturing Park, or CAMP. By September 4, BCDA was drawing an explicit distinction between the two concepts: Pax Silica as the wider international initiative, and CAMP as the proposed physical advanced-manufacturing ecosystem in New Clark City. Trends & Spots has previously documented this shift in terminology in its report on the Pax Silica project being “re-branded”.

What the public record shows is that early government communications unquestionably promoted the New Clark City development using AI-centered terminology, before later messaging became much more specific about manufacturing and explicitly rejected the idea that massive hyperscale data centers would operate at the Tarlac site. This change occurred during a period of growing public opposition and government scrutiny. While there is no publicly available document proving BCDA changed the terminology specifically to defuse public opposition, the shift in public framing raises legitimate questions.

The 3-Gigawatt Question Has Not Disappeared

One of the most important questions is the previously announced 3 GW power estimate. BCDA described this as an initial estimate for the development at full build-out. That estimate was made before BCDA’s more forceful August clarification that massive data centers would not be part of the New Clark City site.

If hyperscale data centers are excluded, does the approximately 3 GW estimate still apply to the proposed Clark Advanced Manufacturing Park?. If it does, what combination of activities produces that projected demand?. Possibilities discussed publicly around Pax Silica have included semiconductor manufacturing, electronics production, advanced manufacturing, critical-mineral processing, AI-related hardware, logistics, research facilities, and supporting infrastructure. A 3 GW figure should not be interpreted as proof that hidden hyperscale data centers are planned, as that conclusion would go beyond the available evidence. However, an estimate of that magnitude deserves a transparent explanation—and a revised resource projection should be made public if the project scope has materially changed.

Pax Silica’s Water Estimate Needs Context

BCDA has publicly estimated that industries in the proposed development could require between 65 million and 90 million liters of water per day. The agency says it does not plan to draw this requirement from groundwater used by nearby communities and agriculture, proposing instead a surface-water harvesting and storage system intended to provide up to 120 million liters per day, with potential expansion to 300 million liters per day.

This is an important clarification, but it is also still a proposal. A proposed supply system does not, by itself, answer questions about seasonal availability, watershed impacts, drought conditions, future climate variability, industrial wastewater, competing water uses, treatment requirements, or actual consumption once tenants are known. These issues become particularly important if semiconductor fabrication or other water-intensive industries eventually locate within CAMP. The distinction between a hyperscale data center and an advanced manufacturing park matters, but it does not eliminate the need for detailed water analysis.

No Pax Silica-Specific Environmental Impact Assessment Processed

Perhaps the most important unresolved issue emerged on August 26, when Department of Environment and Natural Resources officials confirmed during a House budget hearing that no environmental impact assessment had yet been applied for or processed specifically for Pax Silica. Acting DENR Secretary Juan Miguel Cuna also stated that the agency did not yet have a clear project description to intelligently assess the environmental impact.

DENR’s Environmental Management Bureau clarified that an Environmental Compliance Certificate is in place for BCDA’s broader New Clark City development plan. However, that is not the same as having completed environmental assessments for each industrial activity that could eventually operate in the proposed manufacturing park. The absence of a project-specific assessment does not prove that the project will cause unacceptable environmental damage, but it does mean claims about its environmental sustainability cannot yet be treated as settled. The final environmental footprint depends on incomplete information.

Manufacturing Is Not an Environmental Free Pass

If New Clark City will host manufacturing for the AI supply chain rather than large data centers, the environmental analysis must shift accordingly—it cannot simply end. Semiconductor fabrication can require substantial amounts of electricity and water, involving ultrapure water production, specialized gases, solvents, acids, and other chemicals, and can generate wastewater and hazardous waste that require sophisticated control systems. Critical-mineral processing and electronics manufacturing have their own distinct resource and waste profiles. Calling something a “manufacturing park” tells the public very little about its environmental footprint unless the public also knows what will be manufactured, by whom, using which processes, with what resources, and under which safeguards.

The economic case for Pax Silica should not be dismissed without examination; government officials argue that the initiative could help elevate the Philippines in global semiconductor and AI supply chains, attract investment, and create higher-value jobs. BCDA has projected more than 130,000 jobs and between US$40 billion and US$70 billion in investment at full development. But economic potential does not answer environmental questions. Projected employment does not determine sustainable water availability, and projected investment does not establish the capacity of the electricity system.

The Bigger Lesson From the AI Boom

Pax Silica illustrates a much broader global challenge. The AI industry is frequently described in terms of chips, algorithms, computing performance, and investment, but its expansion is increasingly shaped by something much more physical: power plants, electrical grids, transformers, cooling systems, water, semiconductor fabs, chemical supply chains, wastewater infrastructure, fiber networks, and land.

The question facing the Philippines is larger than the debate over the names “Pax Silica,” “AI hub,” or “Clark Advanced Manufacturing Park.” The real issue is what will be physically built, what resources it will consume, what environmental impacts it could create, and who ultimately bears those costs and risks. For a development of this scale, those answers should come before irreversible commitments, not after them.