Data centers are hungry for energy. Texas has a sunny solution.

After decades of flat load growth on US power grids, the nation’s electricity demand is spiking. One major factor is the explosive expansion of AI-powered data centers. Some of those data centers are big. Very big. Hyperscalers have proposed the idea of data centers as big as 5 GW.

Just in January 2025, OpenAI, Microsoft, NVIDIA, Oracle, and others announced the Stargate Project, joint plans to invest a staggering $500 billion in the next four years to expand AI infrastructure in the US. A $100 billion Phase 1 will build a data center in Texas whose footprint will be essentially as large as New York City’s Central Park.

As a result, McKinsey is forecasting a 22% compound annual growth rate (CAGR) for global data center electricity demand through 2030. The company’s mid-range scenario forecasts a quadrupling from 55 GW to 219 GW. Goldman Sachs similarly forecasts data center electricity consumption ballooning from 200 TWh annually in recent years 5x to more than 1,000 TWh by 2030.

Meeting that electricity need has data center owners and operators and grid operators alike trying to figure out how to do it. Reliability is, of course, #1. Data centers aren’t going to reach the famous “five 9s” of availability without it. But from there, how do you optimize your electricity strategy?

Three factors that influence data center power planning

As data center owners and operators look to bring new facilities online (and keep existing facilities running smoothly and efficiently), at least three factors are influencing their power planning:

• Speed. Naturally, the longer a given data center takes to come online, the longer it delays generating revenue. But even more, AI is creating a new version of a computing arms race. Companies that can get data centers online faster will give their models more time to train and acquire more learning, so they can go on to beat other models and gain a competitive advantage. Long interconnection timelines and permitting hiccups due to inadequate power planning can present real obstacles, making locations with shorter interconnection times and streamlined permitting very attractive.

• Energy cost. Power for servers, cooling systems, and other components is a major driver of both data center OpEx and total cost of ownership (TCO) when taking into account CapEx depreciation. Ongoing improvements in power usage effectiveness (PUE) are helping to ensure that more power is going toward actual computing, but AI’s thirst for energy (and lots of it) will continue to make managing energy costs a central variable for overall data center economics.

• Sustainability. All the big tech players have equally big sustainability goals. But after years of steadily lowering their Scope 2 emissions associated with electricity consumption, that familiar culprit of AI-powered computing is pushing emissions upward again. For example, Google saw its Scope 2 emissions jump 37% in 2023 vs. 2022 (2024 numbers are not yet available). Procuring clean, emissions-free power is now the name of the game to meet electricity needs while keeping climate and sustainability targets within reach.

Texas emerges as prime country for US data centers

The rise of AI-centric computing — and an associated decrease in the need for ultra-low network latency — has allowed new data center expansion to shift away from traditional hubs such as Northern Virginia and Silicon Valley. Developers can broaden their geographic lens to search for places that have plentiful, available, and affordable land on which to build new data centers… not to mention favorable electricity markets.

This has made the Lone Star State and its power market, the Electric Reliability Council of Texas (ERCOT), an attractive choice for America’s latest wave of data center expansion. It’s perhaps no surprise that the Stargate Project chose Abilene (pop: 130,000) — about a three-hour drive west of Dallas — as the location for its Phase 1. And late last year, Oncor (one of ERCOT’s major transmission and distribution utilities) said that three-quarters of the large commercial and industrial loads in its interconnection pipeline were data centers.

Overall, Texas boasts some of the cheapest industrial and wholesale power prices in the country. It also boasts some of the fastest interconnection times for large loads, such as data centers. Whereas the wait could be as long as seven years on grids in other parts of the country, a few years ago ERCOT instituted interconnection process reforms to accommodate large loads on accelerated timelines. Consequently, many projects are receiving approval to energize in as little as two years.

Three ways solar shines for data centers in ERCOT
 

One other thing that Texas has in abundance is plentiful sunshine. It leads the US in installed utility-scale solar PV capacity and last year was home to nearly half of the largest new solar installations to come online nationwide. That means a trifecta of benefits that address the three big factors we identified earlier in this piece.

1. Solar is fast

“Bring your own power” (BYOP) has become a prevailing mantra among data center developers looking to get faster approval from state and local permitting agencies and grid operators. Co-developing data centers and new generation to help meet the associated electricity demand is almost becoming standard practice these days. But which generation to build?

Nuclear is notoriously slow and expensive to get built. Natural gas is getting a look, but comes with fuel cost volatility and a hefty carbon footprint. And then there’s solar — and net-metered industrial solar in particular, a solar PV form factor unique to ERCOT that sites utility-scale solar close to the load (i.e., data center) and treats the generation like a behind-the-meter project from a net load standpoint. Importantly, these types of solar projects can be built within a 12- to 18-month timeline, so they can come online as fast as the data center itself.

2. Solar is cost-effective

Utility-scale solar famously has the lowest levelized cost of energy (LCOE) of any new-build form of electricity generation, even without US federal tax subsidies. And net-metered industrial solar in ERCOT is an even better deal, resulting in the lowest all-in power costs for industrial loads such as data centers.

That’s thanks to several contributing factors, including industrial solar’s proximity to load (which helps avoid getting basis risk priced into PPAs, since the solar and the load sit on the same side of the transmission grid) and net-metered industrial solar helping offset the demand charges that can dominate industrial power bills.

3. Solar is sustainable

Finally, there’s the carbon footprint of your power choice. Microsoft made headlines when it announced a PPA with Constellation Energy to reopen the (emissions-free) operational nuclear reactor at Three Mile Island. Of course, the US doesn't exactly have an entire catalog of old nuclear power plants waiting to be called back into service. So data centers need other, better solutions than Microsoft's headline-grabbing one-off announcement for clean power choices. On that front, solar is a no-brainer.

Seizing the opportunity through strategic partnerships

Siting new-build data centers in ERCOT and planning for BYOP solar is as solid a pairing as some Texas brisket with a side of white bread. Yet navigating the complexities of putting all the key pieces together takes some expertise. Data centers can improve their chances of success by partnering with in-state specialists prepared to create a bespoke plan for each physical location. Strategic partners can lead this aspect of planning and execution so data centers can focus on preparing for operations after interconnection.

Ultimately, as data centers plan for how to power the industry’s ongoing growth, Texas offers the fastest, most affordable, and most sustainable solutions to expedite uptime and start delivering on the massive promise of AI. A data center built on inexpensive land in Texas, with net-metered solar nearby, is most likely to thrive due to shorter interconnection timelines and a confluence of ideal conditions for AI data center development.

Jeffrey Sabins is the Chief Development Officer for Industrial Sun.