Energy and Sustainability for AI Infrastructure
A single Google search uses 0.3 watt-hours of electricity, while a request for OpenAI’s ChatGPT takes 2.9 watt-hours. The International Energy Agency forecasts that global electricity consumption by data centers could surpass 1,000 terawatt-hours by 2026, more than twice the amount used in 2022 (We quoted this in the first newsletter of AI Infrastructure thesis series). In a 2024 white paper, EPRI predicted that if a projected high growth rate of 10% per year continues, data centers will annually consume up to 6.8% of total U.S. electricity generation by 2030.?
Energy is the Major Bottleneck for AI Infrastructure
According to Exelon CEO Calvin Butler, AI-driven power demand is set to jump 900% in the Chicago area in the following years. “About 25 data center projects that would consume around 5 gigawatts of power total – roughly equivalent to the output of five nuclear plants – are undergoing engineering studies in the territory of Exelon unit Commonwealth Edison Co.’s territory, Butler said” (News). This challenge is not limited to the U.S.; Ireland’s state-owned grid operator has halted connecting new data centers in Dublin until 2028 (News). As a result, Microsoft decided to build its own power plant with natural gas. Energy and transmission availability is the first consideration when AI data centers make site selections.
All Kinds of Energy Sources On Deck
These data centers often rely on current utilities powered by fossil fuels. “Soaring electricity demand is slowing the closure of coal plants elsewhere. Almost two dozen facilities from Kentucky to North Dakota that were set to retire between 2022 and 2028 have been delayed, according to America’s Power, a coal-power trade group.” (News). However, enterprises face pressure and responsibility to reduce their carbon footprints and environmental impact. Many clean energy solutions have been involved in supporting AI’s power hunger, including wind, off-shore wind, solar, nuclear, hydropower, geothermal power, nuclear, carbon capture, waste to energy, and more.?
Power access constraints are driving demand for on-site power generation, off-grid, or micro-grid solutions. Many companies have similar approaches with modular data centers, clean energy generation, and storage solutions. Bloom Energy (BE) and ECL use hydrogen power for AI data centers. East of Houston, ECL recently announced that it will build the world’s first 1 GW “off-grid, hydrogen-powered modular data center that operates 24/7 with zero emissions, minimal noise and a negative water footprint.” The center may also be expanded to 2 GW in the future. It will use 3D printing and water output from the hydrogen fuel cell used for cooling, thus requiring zero water and local power. Excess water can be given back to communities (News). For comparison, data centers usually consume millions of gallons of water per year.
All of the largest tech companies have energy strategies to support AI. Sage Geosystems provides carbon-free geothermal power to Meta, and Fervo Energy serves Google (News). Constellation (CEG) just signed its largest-ever power purchase agreement with Microsoft to provide approximately 835 megawatts of nuclear energy (News). Claiming to be the first mover in providing direct-connect 24/7 carbon-free nuclear power to data centers, Talen Energy has long-term agreements with AWS (News). Amazon also recently announced an investment in small nuclear reactors (SMR) developer X-energy (News). Google just signed an agreement with SMR developer Kairos Power to enable up to 500 MW and aim to bring the first one online before 2030 (News). According to the company's annual environmental report, Google consumed more than 24 terawatt hours of electricity last year. One terawatt is equal to 1,000,000 megawatts. But Google has pledged to meet net-zero emissions and run carbon-free energy every hour of every day on every grid where it operates by 2030.
Nuclear Power is Hotter than AI Now
Investment interests in nuclear energy-related deals are hot for its 24/7 carbon-free power within a compact footprint. A quick database search shows that over 30 deals in this space happened in Q3 2024 alone, including grants, equity, and debt financing. Or simply check out listed nuclear energy companies, such as Nuscale Power Corp (SMR), Nano Nuclear Energy (NNE), or OKLO (OKLO). The interest is especially high in companies developing SMRs for their smaller and more flexible modular designs. All of the SMRs have yet to start operating in scale.
The US Department of Energy (DOE) has just published an updated version of its “Pathways to Commercial Liftoff Advanced Nuclear” report, writing that nuclear capacity could triple by 2050 from 100GW to 300GW. The report highlights a significant rise in electricity demand over the past year, following decades of stagnation. This surge, primarily driven by AI and data centers, has intensified interest in nuclear energy due to its ability. However, barriers remain and the DOE has discussed success factors including partnerships (A briefing by Data Center Dynamics).
Clean Energy Alignment for Positive Impact?
Using wasted clean energy to power computing is also pursued by many and dates back to Bitcoin mining. Here is the context. The transmission infrastructure that is needed to capture and fully utilize all renewable energy generated is lagging. And, renewable energy production is asynchronous with power demand. These lead to curtailment, congestion, and negative pricing. The excess energy is cheap for users, and bringing in users also benefits the clean energy industry. Examples are Slouna (SLNH) and Crusoe. They bring modular data centers to co-locate with renewable energies as well as provide cloud computing operations and data center infrastructure. Soluna is publicly listed, and its financials have been improving. Crusoe operates modular data centers near oil fields to power them by converting methane released from natural gas flaring into electricity. Methane is a potent greenhouse gas with 82.5x the global warming potential (GWP) of carbon dioxide over a 20-year period.
Some big tech offtakers are willing to pay higher prices to secure renewable energies and meet their sustainability objectives. Investments in renewables could bring attractive returns to investors. Furthermore, Google and Microsoft, along with Constellation, AES, and LevelTen, formed the Granular Certificate Trading Alliance in December 2023 to support the 24/7 renewables ambition – each hour of electricity consumption must be matched with concurrent renewable energy production, which is more challenging than offsetting with 100% renewable energy on an annual basis.
Integration and Orchestration with Grid
AI processes run continuously, but training is latency-insensitive. Orchestration and optimized scheduling of power sources have crucial benefits in cost and carbon footprint reductions. On the other hand, data centers could participate in demand response programs. This integration enables them to adjust their energy consumption based on grid signals, enhance their use of alternative power sources, and contribute to overall grid stability. Digital infrastructures and services for integrating data centers with their communities are in strong demand. Lancium brings very large loads – such as AI data centers – to strategic locations with abundant renewable energy. Applied Digital (APLD), backed by Nvidia, also designs and develops next-generation data centers placed near renewable energy sources in North Dakota.
According to TechCrunch, Sidewalk Infrastructure Partners (SIP), “the Alphabet spinout that focuses on building and backing new approaches to complicated infrastructure problems in areas like power, broadband, and waste management,” launched Verrus earlier this year. Verrus data centers are designed to be grid-aware, carbon-aware, and compute-aware, ensuring they contribute positively to the electrical grid and environmental goals while supporting complex computing tasks. For example, AI training could be bundled into batches distributed across time frames where there is less demand. Furthermore, AES and Tapestry, Google’s moonshot for the electric grid, have partnered to virtualize AES’ planning and operations into an AI-centric digital semi-autonomous grid and published a 20-year framework. This is how AI can help!
AI Helps with AI’s Power Hunger Issue
AI itself has a big role to play in boosting the energy efficiency of data centers, but that takes time. The Uptime Institute Global Data Center Survey 2024 reveals that:
Processing machine learning, large language models, drug discovery, graphics rendering, and modern AI computing demands significantly more power than traditional computing devices. A typical CPU in a data center uses approximately 300 watts per hour, whereas a Nvidia H100 GPU can use 700 watts per hour (comparable to an average American household).
Next-generation AI data centers require purpose-built power-centric design and infrastructure to support significant high power density consumption and energy efficiency. Most operators recognize AI’s potential to boost energy efficiency through dynamic resource allocation in response to real-time workload demand and heat patterns. Despite many operators planning to host the technology, trust in AI for use in data center operations has declined for the third year in a row. Globally, over 200 companies develop tools that manage energy consumption, environment, power, capacity, and assets. About half of them offer software or appliances that manage power and cooling in data centers. This space might be a better target for capital-light startups and investors, and it has been underinvested in the last 2 years. Domain experience is crucial. Phaidra is particularly a startup to watch.?
The largest portion of data center energy consumption is related to cooling and heat management - about 40–50%. We digged into cooling tech and companies here.
Back-Up Power and Energy Reservoir with Grid
Another energy demand is backup/supplemental power systems. Data centers need proper batteries for their special requirements, with very fast discharging and recharging speeds without the need for any thermal settling or cooling, according to Natron Energy. The CEO sees that batteries will be used not only for back-up power, but also as a buffer and an energy reservoir between AI compute and the grid. Natron just announced plans for a $1.4 billion giga-scale sodium-ion battery manufacturing facility in North Carolina. The company has received significant public and private financing from investors like ARPA-E, ABB, Chevron, United Airlines, and Khosla Ventures.?
Latitude Media also asks, “Can data center customers pull LDES out of the valley of death?” Long-duration energy storage company Eos Energy, which makes a zinc-based battery, secured up to $315.5 million from Cerberus Capital Management this year. Eos has a pipeline of buyers, which grew by around 51% last year, spurred by the AI boom. And projects in the pipeline are getting bigger, with longer discharge durations. In the past, the bulk of that pipeline has been made up of independent power producers and utilities (focusing on cost), but that balance is starting to shift.? Data center developers and hyperscalers (prioritizing customization) are expressing more interest.?
It’s worth mentioning that not all battery companies are booming. Moxion Power, which targeted the portable energy storage market, just went bankrupt despite raising $124M in total from big names such as Microsoft, Energy Impact Partner, and Amazon’s Climate Pledge Fund. This is a good lesson for tech startups and investors about aiming at the right market with product-market fit and strong growth levers.
10X Increase in Startups Eyeing Data Centers Last Year
Like Jensen Huang said: “The next industrial revolution has begun. Companies and countries are partnering with NVIDIA to shift the trillion-dollar traditional data centers to accelerated computing and build a new type of data center — AI factories — to produce a new commodity: artificial intelligence.” But TechCrunch reports that “data center tech is exploding but adoption won’t be easy for startups.” Sophie Bakalar, a partner at venture capital firm Collab Fund, says there are a lot more startups eyeing data centers – a 10x increase in founders over the last year. Barriers for startups include competition, trust, scalability, and too high a stake in adoption.?
This post is just a quick run-through of major trends on this landscape. We are tracking hundreds of companies related to next-gen data centers aligned with energy generation, sustainability and efficiency. Subscribe to our weekly updates and explore more collaboration with our investor network. We’ll continue to talk about more ways to boost AI computing energy efficiency such as in semiconductors, interconnectivity, and machine learning modeling or data processing efficiency.
Let us know your investing interest, criteria, and what data/signal you look for in companies. Welcome to recommend your portfolio companies that should be on our radar for our investor network.
Apply to Join Global League Club - We collect and clean data/signals, so you focus on deal-making.