Blog

Why Electricity and Heat Have Become the Binding Constraints on AI Infrastructure, and Which Markets Benefit

Mobility Foresights Technology & Energy Research | Hub Analysis | September 2026

AI infrastructure is now limited less by chips than by power and heat. The International Energy Agency estimates that data centers used about 415 TWh of electricity in 2024 and projects that figure could more than double by 2030, with AI the main driver. Rack densities have risen from roughly 10–20 kW in conventional data centers to well above 100 kW for current AI systems, and roadmaps point toward several hundred kilowatts and eventually 1 MW per rack.

Those densities are forcing three architectural changes at once:

• A shift to 800-volt DC power distribution, announced by NVIDIA and partners in 2025 for next-generation AI factories.
• A move from air to direct liquid cooling.
• A search for on-site and dedicated power, from gas turbines to restarted nuclear plants and small modular reactors.

This hub brings together 61 Mobility Foresights market studies across six areas: 800V DC and rack-scale power architecture; server and rack power conversion; power distribution, protection and backup; liquid and immersion cooling; facility cooling and fire protection; and on-site power, gigawatt campuses and cost risk.

Key Takeaways

• 800V DC is the next power architecture. It cuts copper, conversion losses and space, creating new markets for DC switchgear, breakers, busway, connectors and converters.
• Every conversion stage is being redesigned for efficiency. GaN and SiC devices, high-current inductors and advanced power-management ICs are gaining share.
• Liquid cooling is now standard for AI racks. Coolant distribution units, manifolds, quick disconnects and leak detection are scaling from niche to volume.
• AI loads are unusually volatile. Synchronized training creates power swings that require new protection, monitoring and energy-storage buffering.
• Power access decides where AI is built. Gigawatt campuses are securing gas, nuclear and fuel-cell supply directly, reshaping utility and energy-equipment markets.

How We Segment AI Data Center Power and Cooling

1. 800V DC and Rack-Scale Power Architecture

Conventional data centers convert AC power several times between the grid and the chip, and 54-volt rack busbars cannot carry the current megawatt racks require without very large copper. The industry's answer is high-voltage DC distribution. Four reports cover the architecture:

• Our 800 V HVDC data center market report sizes the transition.
• The AI rack power architecture market study compares 48V, ±400V and 800V designs, including open-hardware disaggregated power racks.
• The AI infrastructure power market report sizes all power equipment serving AI facilities.
• The AI rack-scale infrastructure market study covers integrated rack systems that combine compute, power and cooling.

The 800V chain creates a new set of component markets:

• Conversion: the 800V DC-DC converter market for AI data centers and 800V bidirectional DC-DC converter market for AI data centers reports cover step-down conversion and conversion that can also feed stored energy back into the system.
• Switching and protection: the 800V DC switchgear market for AI data centers and 800V DC circuit breaker market for data centers studies cover DC switching and protection. The 800VDC data center ground-fault location systems market report covers fault detection in DC systems.
• Distribution hardware: the 800V DC busway market for AI data centers, 800V DC data center power cable market, 800V DC data center power connector market and 800V DC power rack market for AI data centers studies cover the physical distribution hardware.

2. Server and Rack Power Conversion

Inside the rack, power must be converted to the sub-1-volt levels that GPUs use, at currents of hundreds or thousands of amperes. Every percentage point of efficiency matters at gigawatt scale. Four reports cover the supply side:

• Our AI server power market report sizes server power.
• The AI server power supply market and AI data center power supply unit (PSU) market studies cover the PSUs converting rack power to board level.
• The industrial power conversion and AI PSU optimization market report covers efficiency optimization.
• The AI datacenter converter market study sizes conversion stages across the facility.

Board-level conversion is where semiconductor innovation concentrates:

• The intermediate bus converters (IBC) market and AI and cloud data center DC/DC power market reports cover intermediate conversion.
• The AI datacenter voltage regulator market study covers voltage regulators next to the processor.
• The AI server power management IC (PMIC) market and AI datacenter power management market reports cover the controllers that coordinate them.
• The AI server power MOSFET market study covers switching devices.
• The AI hardware power inductor market and high-current power inductors for AI data centers market reports cover magnetics, where high currents and space limits are pushing new materials and designs.

3. Power Distribution, Protection and Backup

AI racks draw enormous power, and that power changes quickly. Synchronized training jobs can swing a data center's load by large amounts within seconds, which stresses equipment and can affect the grid. Three reports cover the response:

• Our AI power burst detection market report covers monitoring systems that detect these swings.
• The AI data center power monitoring IC market study covers the sensing chips behind them.
• The AI data center supercapacitor backup power market report covers fast-response energy storage used to smooth the swings and ride through short disturbances.

Distribution:

• The AI data center power distribution unit market report provides the market view of rack PDUs.
• The AI data center power distribution units (PDU) market study provides the product and technology view.
• The in-rack power distribution system market report covers busbar and shelf-level distribution.
• The digital busway (busway monitoring systems) market study covers overhead busway with built-in metering.
• The data center high-current busbar connector market report covers the connectors that join them.

Protection and backup:

• The high-current inrush protection and advanced protection devices market report covers inrush and fault protection.
• The data center battery energy storage circuit protection elements market study covers protection for battery systems.
• The UPS for AI data center market report covers uninterruptible power supplies, now mostly lithium-ion.
• The data center modular battery energy storage system market study covers larger battery systems that provide backup and increasingly grid services.

4. Liquid and Immersion Cooling

Air cannot remove the heat of modern AI racks, and leading GPU platforms are designed for direct-to-chip liquid cooling. The coolant distribution unit (CDU) is the heart of the system, moving coolant between the facility water loop and the servers. Three reports cover it:

• Our in-rack coolant distribution unit (CDU) market report covers CDUs.
• The AI data center coolant distribution unit pump market study covers the pumps inside them.
• The in-rack coolant distribution manifolds (CDM) market and data center liquid cooling manifold market reports cover the manifolds that deliver coolant to each server.

Liquid cooling adds a large component and sensing market:

• The data center quick disconnect coupling market, data center liquid cooling connectors market and data center liquid cooling hose assembly market reports cover the fluid connections, where reliability standards are tightening.
• The data center liquid cooling leak detection market study covers leak detection, now essential with water near electronics.
• The data center coolant flow sensor market and data center coolant quality sensor market reports cover monitoring of flow and fluid condition.

Immersion cooling places servers directly in dielectric fluid. The two-phase immersion cooling fluids market report covers the boiling fluids used in the most efficient designs, a market reshaped by the phase-out of some PFAS-based fluids. The autonomous immersion-cooled edge computing system market study covers sealed immersion units for edge sites.

5. Facility Cooling and Fire Protection

Even with liquid cooling, facilities must reject heat to the outside. Three reports cover the heat-rejection chain:

• The data center free cooling chillers market report covers chillers that use cool outside air or water to cut energy use.
• The data center outdoor packaged cooling system market study covers modular rooftop and outdoor units.
• The embedded AI computer thermal transfer plates market report covers cold plates in compact and embedded AI computers.

Higher power density and large lithium-ion battery systems raise fire risk. The fire protection systems in data centers market report covers detection and protection design. The data center fire suppression system market study covers clean-agent and water-mist suppression suited to electronics and batteries.

6. On-Site Power, Gigawatt Campuses and Cost Risk

AI campuses now plan capacity in gigawatts, and projects such as the $500 billion Stargate initiative announced in 2025 have made power procurement a strategic function. Our gigawatt scale AI data center market report covers these campuses and their power, cooling and land requirements. Because grid connections can take years, operators are securing dedicated generation. Three reports cover the main options:

• Gas. The natural gas powered AI data center market report covers on-site gas turbines and engines.
• Nuclear. The nuclear energy powered data center market study covers power purchase agreements with existing plants, including a restarted reactor in Pennsylvania, and small modular reactor partnerships.
• Fuel cells. The fuel cell powered AI data center market report covers solid-oxide and other fuel cells that can be deployed faster than grid upgrades.

The data center-in-space infrastructure market report covers early-stage concepts for orbital computing powered by continuous solar energy. The US–Iran war impact on AI data center cost inflation market study examines how energy-price and supply-chain shocks from Middle East conflict feed through to data center construction and operating costs.

Where Value Is Created: The AI Power and Cooling Value Chain

Regional perspective

• United States. The largest AI build-out, concentrated where power is available, including Texas, the Southeast, the Midwest and the Mid-Atlantic, with growing on-site gas and nuclear deals.
• Europe. Grid constraints in major hubs are pushing capacity toward the Nordics and southern Europe. EU energy-efficiency reporting rules for data centers apply.
• Middle East. Gulf states are building large AI campuses backed by abundant energy and sovereign investment.
• Asia-Pacific. Rapid growth in Japan, India, Malaysia and Australia, with power and water availability as key constraints.

What to Watch Through 2032

1. 800V DC adoption: the first large deployments and the standards that emerge for components and safety.
2. Liquid-cooling reliability: field failure rates for quick disconnects and CDUs as volumes scale.
3. On-site generation: the balance between gas, fuel cells and nuclear, and the regulatory response.
4. Grid impacts: utility rules for large, volatile AI loads, including requirements for on-site storage.
5. Transformer and switchgear lead times: whether equipment capacity keeps pace with campus announcements.

Frequently Asked Questions

Why are AI data centers moving to 800V DC power?

AI racks are approaching several hundred kilowatts and eventually a megawatt. At 54 volts the current and copper required become impractical, while 800V DC reduces current, conversion stages, losses and space.

Why do AI servers need liquid cooling?

Modern AI accelerators produce more heat than air can remove efficiently at high rack densities. Direct-to-chip liquid cooling carries heat away through cold plates and coolant loops, and leading AI platforms are designed for it.

What is a coolant distribution unit?

A CDU pumps and conditions coolant for the servers and exchanges heat with the facility water system. It controls flow, pressure and temperature and is the central component of liquid-cooled AI racks.

How much electricity do data centers use?

The IEA estimates data centers used about 415 TWh in 2024, around 1.5% of global electricity, and projects consumption could more than double by 2030 as AI grows.

Why are AI data centers building their own power plants?

Grid connections for large new loads can take years. On-site gas turbines, fuel cells and agreements with nuclear plants allow campuses to secure power sooner and with more certainty.

What causes AI power bursts?

Large training jobs run thousands of accelerators in lockstep, so power draw can rise and fall sharply as computation phases change. Monitoring, fast energy storage and protection devices are used to manage these swings.

What do Mobility Foresights data center reports include?

Each linked report provides market size and multi-year forecasts, growth drivers and challenges, segmentation by component, application and region, competitive landscape and company profiles, and recent developments. Customization is available on request.

Work With Mobility Foresights

Mobility Foresights provides syndicated and custom research across AI data center power, cooling and energy supply. Each report linked above can be purchased individually or tailored to a specific component, technology or region. To request a sample or commission a bespoke study, contact our technology research team.