How Hyperscalers Are Powering Their Data Centers

Governor Sarah Huckabee Sanders, Ruth Porat, President & Chief Investment Officer of Alphabet and Google, and Jim Anderson, Google’s Vice President of North America Partner Ecosystems, discussing Google’s third US data center and $4 billion investment plans in Arkansas. Oct 02, 2025. Source: Google.

Major cloud providers and AI computing companies are acutely aware of their outsized energy footprint and have launched a variety of strategies to secure power supply and improve efficiency. These strategies range from massive renewable energy purchases to advanced engineering for efficiency, and even direct investments in energy projects. Below we review initiatives from some of the industry’s leaders:

Google: Google has been carbon-neutral since 2007 and in recent years matched 100% of its annual electricity use with renewables. Now it aims for the ambitious target of 24/7 Carbon-Free Energy by 2030, meaning each Google data center would be powered by clean energy at every hour of the day. To get there, Google has signed over 50 renewable PPAs worldwide (totaling several gigawatts of wind and solar) and developed novel contracts like hourly firming (pairing solar, wind, and battery storage to cover round-the-clock load). Recently, Google announced its third US data center in Arkansas and committed to building a 600 MW solar project with Entergy to enhance local resilience. Google also invests heavily in efficiency: its data centers boast an average PUE (Power Usage Effectiveness) around 1.1–1.2, meaning very little wasted energy on cooling overhead. Notably, Google applied AI (via its DeepMind unit) to its cooling systems – an AI control system that reduced cooling electricity by ~30% in some facilities by smartly managing chillers and fans. On the supply side, Google is pursuing cutting-edge projects to secure firm green power, as mentioned earlier (e.g. partnering in advanced geothermal projects in Nevada, and investigating small modular nuclear reactors). Google also announced a partnership in April 2025 with grid operator PJM and its Alphabet sibling “Tapestry” to use AI for speeding up grid interconnections (a direct effort to tackle the delays in getting new energy projects online for its data centers). Most interestingly, Google uses sophisticated forecasting to make its data centers “flexible” and shift some workloads to when and where renewable energy is plentiful. Its ‘carbon-intelligent computing’ platform shifts moveable compute tasks between different data centers globally based on regional hourly carbon-free energy availability. This kind of load flexibility is still in early stages (since most data center loads are mission-critical and can’t be interrupted), but Google’s demand response pilots with utilities have shown promise in reducing peak grid stress by timing non-urgent jobs to off-peak hours and targeting machine learning workloads during periods of grid strain. All these moves underscore Google’s holistic approach: reduce the energy each data center needs, offset or directly procure renewables for 100% of that energy, and innovate to fill any gaps (with batteries, novel sources, or demand flexibility).
Microsoft (Azure): Microsoft has a bold commitment to be carbon-negative by 2030 (meaning it will remove more CO₂ than it emits, including from power use). To that end, Microsoft has contracted for green energy across the globe, from solar farms in Virginia to wind in the Netherlands. It also pioneered sustainability principles for AI: for instance, it co-founded the “24/7 Carbon-Free Energy Compact” with Google under UN auspices, encouraging hourly tracking of energy sources. To reduce its footprint, Microsoft invests in energy-efficient chip design (including custom AI accelerators that do more work per watt) and advanced cooling (experimenting with liquid immersion cooling for servers at its Quincy, Washington facility, which can cut cooling power needs). Microsoft is also exploring on-site power generation for data centers. In 2020, Microsoft deployed a pilot where servers were powered by hydrogen fuel cells for 48 hours, and in 2022 ran a 3 MW hydrogen fuel cell demonstration, hinting at a possible shift to hydrogen backup instead of diesel. And in a headline-grabbing deal, as noted, Microsoft agreed to buy fusion power from Helion by 2028, becoming the first big tech firm to invest in fusion as a future energy source. Microsoft is also backing small nuclear startups and research into grid storage to ensure it can meet its datacenters’ needs with carbon-free sources long-term. Moreover, Microsoft has deployed advanced “energy matching” algorithms for hourly renewable tracking, with pilots in Sweden to dynamically match data center electricity consumption with carbon-free energy supply.
Amazon Web Services (AWS): AWS, as part of Amazon’s overall Climate Pledge, achieved its goal of powering operations with 100% renewable energy in 2023, two years ahead of its 2025 target. Amazon has become the world’s largest corporate renewable buyer for five consecutive years, with over 20 GW of clean energy contracted globally. These include large utility-scale solar and wind deals in the U.S., Europe, India, and beyond. AWS also often builds on-site solar at its data centers (rooftop or adjacent arrays) to supplement grid power, though such installations are relatively small compared to total consumption. To increase efficiency, AWS designs its own servers and cooling systems; it maintains an average PUE of around 1.2 across its cloud data centers. Like its peers, Amazon is addressing the need for around-the-clock reliability: AWS has developed battery projects at some of its facilities (for instance, pairing a 5.8 MW rooftop solar array with a 2.5 MW battery energy storage unit at its San Bernardino Air Hub, allowing solar energy to be dispatched even when the sun sets). Amazon is also leveraging AI to optimize battery performance and energy management; at its Baldy Mesa solar-plus-storage project in California, machine learning software built using Amazon SageMaker analyzes billions of data points to optimize when to buy, store, and sell energy based on grid conditions, helping stabilize the grid during extreme weather events. Additionally, Amazon teams are developing AI models to aggregate performance data from rooftop solar installations and predict energy generation to help minimize overall energy usage. AWS supports demand response programs and distributed energy resource management through its cloud platform, enabling utilities and grid operators to control and optimize energy assets at scale. By doing so, AWS can help smooth out renewable variability and participate in grid services, effectively improving grid stability while also ensuring AWS facilities get the power they need. Additionally, Amazon has been exploring on-site generation using fuel cells and even gas turbines in some regions to have more direct control over power (though this raises emissions questions, it can guarantee supply in constrained grids). Notably, Amazon’s data center designs incorporate quick-disconnect from grid and switching to generators when needed – a standard practice – but AWS is now working with utilities on issues related to “grid inertia” so that such transitions, if they happen, don’t shock the grid.
Meta (Facebook): Meta has powered its operations with 100% renewable energy since 2020, primarily through a portfolio of wind and solar contracts across the U.S. (in states like Iowa, Texas, New Mexico) and internationally. Meta’s data centers are highly optimized for energy efficiency; they pioneered the use of open air cooling and custom server hardware through the Open Compute Project, achieving PUEs as low as 1.07 at some facilities. Meta is unique in that it doesn’t offer cloud services to the public (its centers serve its own social media and AI workloads), but its footprint is still enormous. To manage it, Meta invests in AI for energy optimization. For example, Meta uses machine learning algorithms to optimize data center operations in real time. Since 2021, these algorithms analyze environmental sensors and dynamically adjust cooling systems, reducing overall facility energy and water consumption. While Meta focuses on large-scale renewable procurement (contracting over 15 GW of wind and solar), it is evolving its energy strategy by entering wholesale electricity markets through its subsidiary Atem Energy, LLC, which filed in 2025 for FERC approval to buy and sell power and grid services. This move will give Meta greater flexibility to manage energy costs, hedge renewables contracts, and support grid reliability, marking its transition from consumer to market participant. Moreover, Meta invests in community renewable projects near its data centers (including funds for local solar farms) to ensure the grids hosting its facilities get greener. On the R&D side, Meta has looked into advanced cooling (like immersion cooling for AI training clusters) to curb growing GPU-related energy use, and is successfully reusing server waste heat to warm nearby buildings in colder climates—for example, its Odense, Denmark data center provides waste heat to district heating systems that warm up to 11,000 local homes.
OpenAI and AI-Heavy Startups: Unlike the hyperscalers, OpenAI doesn’t own massive data centers (it primarily leases cloud capacity, mainly from Microsoft Azure). However, OpenAI’s demands are driving unique new strategies. Reports suggest OpenAI is involved in the “Stargate” project with startup Crusoe Energy, aiming to develop dedicated AI data centers with their own power sources. Crusoe (originally known for using flare gas to power Bitcoin mining) has secured 5 GW of gas-fired generation capacity to build out data centers for AI computing. OpenAI would be a major client. This approach – while ensuring power availability – raises concerns, as it relies on fossil fuels: “There are early indications that these ‘Stargate’ datacentres could exacerbate dependence on fossil fuels,” notes researcher Alex de Vries. Sam Altman, OpenAI’s CEO, has even mused about the need for new energy solutions to support advanced AI, including potential investments in nuclear energy. His backing of Helion fusion and other energy startups underscores that AI companies recognize energy supply as a strategic constraint. We may see firms like OpenAI partner to build private power plants (e.g. small modular reactors or dedicated solar farms with storage) solely for their compute clusters. In the meantime, OpenAI benefits from Microsoft’s aforementioned projects (so its strategy is somewhat by proxy, via Azure’s initiatives). Other AI compute providers (e.g. Nvidia’s AI cloud, or Hive AI) are similarly exploring co-location with energy sources. In summary, the AI sector is so hungry for power that it’s willing to invest beyond the traditional scope of tech companies – even into energy production – to secure reliable electricity.
Others: Many smaller cloud and colocation companies are also innovating. For example, Switch and Digital Realty (colocation providers) have both achieved 100% renewable sourcing through PPAs for their multi-tenant data centers. They also build on-site solar plus battery microgrids at some campuses to reduce grid draw at peak times. Switch’s large Nevada campus runs a huge solar array and one of the largest behind-the-meter battery installations in the U.S. to shave its peak demand. Efficiency-wise, companies are adopting new best practices like liquid cooling for high-density racks (which can cut cooling energy ~20–30% and allow more compute per watt) and deploying AI workload schedulers that consolidate jobs onto fewer servers when possible (turning idle servers off). On the carbon offset side, many companies also purchase high-quality carbon credits or renewable energy certificates to cover any gaps in their green power procurement, aiming for net-zero emissions for their operations.

Overall, the major players are using every tool at their disposal to manage their power footprint: procure renewables at scale, improve energy efficiency, shift loads intelligently, and invest in next-generation energy tech. These efforts not only reduce their environmental impact but also ensure stability of supply – a key business issue. As an Amazon sustainability officer put it, “Every watt we save is a watt we don’t have to generate, and every watt we generate cleanly is a hedge against future cost volatility”. Despite these strides, the sheer growth of demand means total energy use by these companies is still climbing. They are essentially in a race: can efficiency and clean energy deployment outpace the rising tide of digital demand? The strategies above are the playbook to try to win that race.