AI data centers are consuming electricity at an unprecedented rate, and the trend is accelerating sharply. U.S. data center power consumption is projected to more than double to approximately 391 terawatt-hours by 2030, up from 180 terawatt-hours last year, according to Bloomberg New Energy Finance data cited in recent industry reports. This explosion in energy demand reflects a fundamental shift in how companies are building and operating AI infrastructure, driven by the race to deploy both training clusters and always-on inference systems that serve AI models continuously.

The scale of this buildout is staggering. The global AI server market, which includes the hardware that powers these data centers, was valued at USD 176 billion in 2025 and is projected to reach USD 872 billion by 2035, expanding at a compound annual growth rate of 16.3%. Global AI server shipments reached approximately 1.1 million units in 2025, reflecting simultaneous infrastructure expansion across hyperscalers like Amazon, Google, Meta, and Microsoft, as well as sovereign AI programs and enterprise operators.

What's driving this power consumption surge? The answer lies in the sheer computational intensity of modern AI chips. Nvidia's newest AI accelerators have pushed the power draw of a single server rack from hundreds of kilowatts to the megawatt range, meaning a group of about 20 server racks, each consuming roughly 120 kilowatts, now requires a cooling system capable of handling 2.4 to 2.5 megawatts of heat dissipation. This represents a fundamental change in how data centers must be designed and operated.

Why Are Tech Giants Spending So Much on AI Infrastructure?

The investment commitments tell the story. Amazon, Google, Meta, and Microsoft combined spent more than USD 416 billion on capital expenditures in 2025, a 66% year-over-year increase, with GPU-dense AI server clusters accounting for the dominant share of this spending. Beyond the hyperscalers, national AI infrastructure programs are adding another layer of demand. OpenAI, Oracle, and SoftBank secured USD 500 billion in committed U.S. AI data center investment across 2025 through 2028, targeting 10 gigawatts of compute capacity through their Stargate initiative.

These commitments reflect a structural shift in how companies view AI infrastructure. The market is moving beyond an initial phase of training-cluster buildout toward a sustained multi-year expansion driven by inference fleet deployment, where AI models must run continuously to serve user requests with minimal latency. Enterprises and regional cloud operators are now sourcing inference-specific AI servers independently of training refreshes, broadening the buyer base and accelerating fleet replacement cycles.

International players are also ramping up aggressively. Alibaba announced plans to expand its global data center capacity to more than 20 gigawatts by 2032 and unveiled its Zhenwu V900 AI chip, which delivers three times the performance of its predecessor and is scheduled for mass production in the first quarter of 2027. These announcements underscore how the AI infrastructure race is becoming truly global, with Chinese tech giants competing directly with Western hyperscalers for computing capacity and power resources.

How Are Companies Managing the Power Challenge?

• Liquid Cooling Systems: LG Electronics received Nvidia certification for a 2.5-megawatt coolant distribution unit (CDU), a large-capacity liquid cooling device that keeps massive server clusters from overheating by cooling multiple racks together rather than installing smaller cooling units at each individual rack. Market researcher TrendForce forecasts that adoption of liquid cooling in AI data centers will rise from 14% in 2024 to approximately 60% by 2027.
• Battery Energy Storage Systems: LG Energy Solution's lithium iron phosphate battery product, called JF2 AC Link, received Nvidia certification for use in AI data center ecosystems, helping stabilize power supply by absorbing sudden swings in AI server power consumption and maintaining steady voltage even when grid conditions fluctuate.
• Strategic Infrastructure Partnerships: LG and Microsoft signed a partnership agreement to supply cooling, power, and information technology infrastructure solutions as Microsoft expands its data centers worldwide, with longer-term plans to expand cooperation into prefabricated data centers built off-site in modules for faster construction.

The certification of LG's cooling and battery systems by Nvidia signals how critical these supporting technologies have become. LG Electronics previously received Nvidia certification for a 600-kilowatt CDU in July and a 1-megawatt CDU in September, showing a rapid progression toward larger, more efficient cooling solutions. LG Energy Solution also signed a contract to supply 6 gigawatt-hours of energy storage system batteries to DTE Energy, a U.S. utility that runs large-scale AI data center projects for companies including Oracle.

What Constraints Could Slow This Growth?

Despite the aggressive expansion plans, several factors could moderate the pace of AI data center deployment. Power availability and electricity cost constraints represent a significant restraint on growth. Grid interconnection timelines, permitting backlogs, and rising electricity costs are delaying AI server deployment in key markets including Virginia, Oregon, Ireland, and the Netherlands, even for buyers with committed capital. These power constraints introduce timing risk to deployments in regions facing congested grid infrastructure through 2028.

Supply chain bottlenecks also pose challenges. High-bandwidth memory (HBM) and advanced chip packaging capacity are concentrated in a small number of facilities, with HBM3E and HBM4 supply produced by SK Hynix, Samsung, and Micron with long-lead production cycles. These supply constraints limit the rate at which funded AI server orders convert to shipped units, creating periodic fulfillment lags that defer revenue recognition.

The convergence of massive capital investment, exponential growth in chip power consumption, and the need for specialized cooling and power infrastructure is reshaping how data centers are built and operated. The question is no longer whether AI infrastructure will consume enormous amounts of electricity, but whether power grids, cooling technologies, and supply chains can keep pace with demand.