3 min read

Why is Meta Turning to Solar Energy to Power AI Data Centers

Why is Meta Turning to Solar Energy to Power AI Data Centers
Photo by Chelsea / Unsplash

Yesterday, Meta signed its fourth major solar energy deal this year, securing 650 megawatts across projects in Kansas and Texas. The agreement was signed with energy provider AES Corporation, and the deal involves solar projects in Kansas and Texas that AES is bringing online in the Southwest Power Pool (SPP), the operator of the central U.S. electric grid and wholesale power market. 

This latest agreement adds to Meta’s already vast renewable power portfolio, exceeding 12 gigawatts, and is explicitly aimed at powering its expanding data centers, crucial infrastructure for its growing AI operations. The move underscores a critical development in the AI era: the explosive demand for compute power is driving tech’s biggest players to become major forces in the renewable energy market, racing to secure clean power for their energy-hungry AI infrastructure.

How much power do AI data centers really need?

The rise of AI has fundamentally changed the power requirements for data centers. Traditional data centers might consume 5–10 kilowatts (kW) per server rack, but AI-optimized data centers, packed with dense clusters of GPUs, require 60 kW or even up to 1 megawatt (MW) per rack — a jump of 50 times or more in just five years. This surging density translates to enormous facility-wide power needs. Small data centers might use 1–5 MW, but hyperscale AI data centers need 20 MW to over 100 MW, with designs already targeting 1 MW per rack. 

Globally, data center power consumption surged 55% in 2023 alone, according to Cushman & Wakefield, a commercial real estate services firm. Meeting the projected demand for AI infrastructure by 2030 will require trillions of dollars in investment, including an estimated $1.3 trillion for “Energizers” — utilities and power providers — to expand generation and distribution capacity.

Why are tech companies focusing on renewable energy like solar?

Major tech companies, including Meta, Google, and Amazon, have set ambitious corporate sustainability goals, often aiming for 100% renewable energy to power their operations and achieve net-zero carbon emissions. Powering energy-intensive AI workloads with renewables is essential to meeting these targets and mitigating the environmental impact of data center growth (data centers account for an estimated 4% of global energy consumption).

Solar energy has become a favored solution for hyperscalers due to several factors highlighted by the Meta deal. It’s one of the cheapest forms of new power generation, even before subsidies. Solar projects also offer a fast “time-to-power” compared to traditional power plants, often taking months rather than years to build once permitting is secured. They can also be phased in, allowing electricity to flow before the entire project is complete, aligning better with the rapid deployment timelines for new AI data centers. Securing large Power Purchase Agreements (PPAs) like Meta’s deals allows companies to directly fund the development of new renewable energy capacity, adding clean power to the grid to match their consumption.

What are the challenges in powering AI data centers with renewables?

Despite the appeal of solar and other renewables, powering AI data centers sustainably faces significant challenges. Renewables like solar are intermittent — they only generate power when the sun is shining. This requires pairing them with energy storage solutions (like batteries, which are becoming cheaper) or relying on always-available baseload power sources (like natural gas or nuclear) when renewables aren’t generating electricity. Integrating massive amounts of new, distributed renewable capacity into aging power grid infrastructure is also complex, leading to significant interconnection queues and bottlenecks. Building new transmission lines to deliver power where it’s needed can take a decade or more, far slower than the pace of data center construction.

Additionally, while addressing carbon emissions, powering AI data centers introduces other environmental concerns. The massive heat generated by dense AI racks necessitates advanced cooling solutions. Many systems still rely on water-intensive evaporative cooling, placing significant strain on local water supplies, particularly in water-stressed regions where data centers are increasingly being built due to favorable energy resources or regulations. This creates a complex trade-off between power efficiency, water consumption, and the location of energy resources.

Where are these AI data centers being built, and why?

The need for vast amounts of power and specific infrastructure is influencing where new AI data centers are located. Companies are seeking out regions with favorable conditions for energy development. Texas, where Meta has signed multiple large solar deals this year, is cited as a “hotbed” for solar development due to ample sunshine, relatively quick permitting, and speedy grid connections. These locations with abundant and potentially cheap power resources, including renewables, are becoming attractive hubs for AI infrastructure. However, this can sometimes lead to building in areas already facing challenges, such as water stress, requiring careful consideration of local environmental impacts. The strategic need for power, speed of deployment, and cost are driving a potential reshaping of the traditional digital infrastructure map.

Reference Shelf:

Meta signed another big solar deal on Thursday (TechCrunch)

AI Infrastructure to Require $7tn by 2030, says McKinsey (Datacentre Magazine)

1,000 homes of power in a filing cabinet — rising power density disrupts AI infrastructure (Goldman Sachs)