AI applications require significant expansion of global energy capacity

The increase in capacity globally will come not only from efficiency gains and expansion of existing centers, but also from the construction of new data centers.

A commitment to sustainable electricity consumption

Renewable energy sources such as wind and solar will play a critical role in meeting the increased demand for computing power as countries strive to meet the Paris Agreement’s goals for reducing greenhouse gas emissions. hyperscalers Western countries also have their own ambitious decarbonization goals.(1)

• Google aims to use only carbon-free energy 24/7 by 2030.
• Amazon plans to power its operations with 100% renewable energy by 2025. The company is also targeting net-zero carbon emissions by 2040.
• Meta (Facebook) has reduced greenhouse gas emissions from its operations by 94% since 2017. It has done this primarily by powering its data centers and offices with 100% renewable energy.
• Microsoft aims to cover all its electricity consumption with zero-carbon energy purchases by 2030. It also plans to eliminate, by 2050, all the carbon it has emitted since the company was founded in 1975.
• Apple now runs all of its stores, data centers and offices worldwide on 100% renewable electricity, approximately 90% of which comes from renewable sources created by Apple. The company has achieved this through long-term power purchase agreements (PPAs) with certain renewable power plants and through equity investments in or direct ownership of other renewable energy facilities. .

Intermittency of solar and wind energy poses a challenge

Data centers are energy-intensive and operate 24/7. Since wind and solar are intermittent energy sources, it has become clear that data centers cannot be powered directly by renewables alone, even when battery technology is used to store the energy produced. (Batteries also present their own challenges due to their cost, limited lifespan, and low efficiency.)

THE hyperscalers have solved this problem by signing virtual PPAs with renewable energy developers, whose energy is fed into the electricity grid, which still gets much of its energy from coal or natural gas plants. The data centers of hyperscalers are then powered by both green and gray electrons (coming from fossil fuels). When the electricity cost of the PPA is higher than that of the network, the hyperscalers pay the difference. When the cost of the PPA is lower, they save money.

Hydro or nuclear power could offer an alternative to dependence on fossil fuels. But hydroelectricity has geographical constraints. As for nuclear power plants, they pose additional problems, ranging from the time required to build them to public resistance to nuclear sites. For now, natural gas offers the most viable option for obtaining energy to complement renewable sources, given that it can provide energy on demand and is a much cleaner alternative to coal-fired power plants.

Multiple obstacles to creating additional capacities

Increasing capacity for timely electricity generation and transmission, while managing the overall stability of power grids, is a challenge that could slow the construction of data centers and the proliferation of AI-based solutions. Multiple additional obstacles appeared.

First, existing data center construction is already having a negative impact on power grids. This has led some data center operators to suspend new additions. In Ireland, where data centers now use 18% of the country’s electricity, no new centers will be allowed to connect to the power grid until 2028. The Netherlands has limited new center construction to two sites, and Singapore has imposed a four-year moratorium on new data center construction.

Second, adapting supply chains to meet the grand ambitions of hyperscalers proves difficult. There is currently a shortage of transformers, the heavy, complex equipment that adjusts the voltage of electricity so that it can be transported long distances and used at safe levels for data centers. Wood Mackenzie, a data analytics provider for the renewable energy sector, estimates that it now takes two years to obtain a transformer, compared to one year at the start of 2022. Since this challenge requires an increase in production and not a technological breakthrough, it may only be a short-term bottleneck.

Third, connecting renewable energy generation to the grid is also taking longer because of growing grid-connection queues. In the United States, for example, it now takes four years to assess the impact of a new renewable energy plant on the grid.(2) New plants also require new power lines to transport electricity from where it is generated to where it is used. The lead times for adding transmission lines are also long. In total, with the three to four years needed to site and permit a new project and another three to four years for construction, the entire process of commissioning a renewable power plant can take up to six to eight years. Unlike supply chain issues, long-standing bureaucratic delays can only be resolved through government action. Given the time it takes to achieve change on this front, it seems likely that these bottlenecks will persist and continue to limit capacity growth.

In response to these challenges, the hyperscalers find alternative solutions. One option is to acquire an “off-grid” captive energy source. Amazon did just that by purchasing a data center in Pennsylvania that draws power from a nearby nuclear power plant.

AI can help solve the problem it creates

Perhaps not surprisingly, AI can help solve many of the challenges associated with providing the increased energy it needs. With AI still in the early stages of development, it is too early to predict exactly how this scenario will play out. Still, it seems highly likely that AI will help discover ways to manage and use energy more effectively and efficiently.