AI is draining the electric grid – this energy tech will save us

As AI pushes the power grid to its limits, virtual power plants (VPPs) offer the most practical solution as a counter, writes David Maher, chief technology officer at Intertrust.

The AI boom is pushing the power grid to its limits.

Goldman Sachs predicts a 160% increase in data centre electricity demand, stating: “Such a spike in power demand hasn’t been seen in the US since the early years of this century. Data centres will use 8% of US power by 2030, compared with 3% in 2022.”

Virtual power plants (VPPs) offer the most practical solution to counter this issue.

However, this innovative approach to energy production creates an unprecedented challenge from an information technology perspective.

Governing the sprawling network of assets composing a VPP is only possible with an innovative approach to managing its underlying computing architecture.

In this article, we’ll discuss the energy challenge, offer an overview of the VPP concept, and explore key elements of the IT approach required to harness the power of an energy grid many times more complex than a traditional power plant.

AI: a different sort of ‘brain drain’ 

AI has many virtues, but energy efficiency isn’t one of them.

Generative AI models learn from vast data volumes, and improving them requires intensive data processing on massive servers residing in multiple data centres across the globe. AI search can use 30 times more energy than traditional searches, and now see many more applications using these large models are growing at a dizzying rate…and consequently using much more energy.

Technologies that might allow AI to operate more efficiently, such as quantum computing, are far off in the future. That creates a predicament as AI use is expected to continue soaring. Thanks to AI, we are rapidly expanding energy demand, reversing years of progress we’ve been making in reducing overall demand and fossil fuel dependence.

Multiplying power plant capacity—either through building new plants or supplementing existing ones—isn’t practical. VPPs, however, offer the best of both worlds: a massive energy production boost with minimal infrastructure development.

Have you read?
Iberdrola looks to generative AI to drive energy sector innovation
Managing peak demand and grid flexibility: The role of renewable energy, VPPs and vehicle-to-grid technology

More energy without massive infrastructure

VPPs are energy grids composed of thousands of disparate types of power sources, storage devices, and demand controllers, including solar panels, wind farms, batteries, and other small, independently-owned energy components. AI is used in VPPs to optimise energy demand, use, and storage.

By some estimates, VPPs could provide a volume of electricity roughly equal to that used by 50 million families. Their popularity has been growing in the U.S. since their 2020 greenlight from the Federal Energy Regulatory Commission and similar trends are emerging in places like Australia, Japan and Europe. Energy industry analytics firm Wood Mackenzie predicts investment in VPPs may surpass $110B by 2025.

However, this novel approach poses an IT/data problem, as it presents an infrastructure much more complex than your typical power plant. Rather than managing power from a single highly controlled source, power companies must manage tens of thousands of independent sources distributed across large geographies.

To call these energy sources non-uniform is to put it mildly–we’re talking about devices from different makers, operating under different specs and with varying kinds of software, many originally brought into operation decades ago. Furthermore, these devices are owned by individual households’ independent operators and not a centralised authority.

Let’s discuss some of the capabilities that, from an IT/data management perspective, are required to make a diffuse and distributed system such as a VPP work, both in the near-term and future.

A single pane of glass

One of the first capabilities required to make a VPP work is a ‘single pane of glass’ or viewport offering a unified and real-time view of all devices, software, and activities taking place across the entire system. This is no small order, considering the system complexity we’re describing. But fundamentals of an effective approach include:

• Real-time, authoritative, and secure control over diverse siloed software and devices
• A trusted, distributed, application level communications system for monitoring and controlling all devices
• Real-time and predictive analysis capabilities based on data from all devices

End-to-end security

Critical infrastructure such as the power grid is a top cyberhacking target of foreign actors, according to FBI Director Christopher Wray. The more complex a system, the greater the potential for an attacker to identify and exploit attack vectors. A VPP–which has to support hundreds of different kinds of devices, each with its own hardware/software vulnerabilities–has all the potential weak points of an IoT network and more.

VPPs face further problems in that they require close cooperation between information technology teams and the operational technology teams that control physical infrastructure, aka IT and OT integration. As these teams typically operate with a high degree of independence from each other, this disconnect creates further complexity and another tempting attack vector for cybercriminals.

Utility companies, therefore, need the ability to secure a VPP’s computing infrastructure at all points, from the cloud to the device. This requires capabilities for controlling assets remotely and regularly conducting security audits on all devices, including older ones, as well as the data exchanged between the device and the cloud. Some specific requirements include:

• Audit tracking for operational compliance and optimization
• Protection of data in transit, at rest, or in use, across the entire data fabric
• Secure IT/OT bridging that allows the IT teams to communicate and collaborate with the OT teams
• A zero-trust architecture approach that doesn’t just verify every device or user, but rather every action taken by that device or user

Flexibility & scalability

From a computing perspective, managing a VPP requires the ability to not only scale the number of devices used by the system but also to adapt the distributed infrastructure to integrate with future assets and technologies. The system must be able to flex to both current and future needs.

While it is hoped that the number of individuals opting their devices into VPPs will increase over time, the growing popularity of solar means that VPPs could see tremendous numbers of devices joining the network in relatively short periods. This means more power sources for VPPs, but it also means that the architecture must handle massive data increases in a short time. To accomplish this, the system must provide a high degree of interoperability based on open standards between all devices and software.

In summary, VPPs offer a hopeful, safe, and optimal option for addressing the strain that AI will place on the power grid, and help us avoid what may be an unprecedented energy crisis. Still, VPPs carry all the information technology baggage of a highly diverse and distributed system. However, power companies should pursue this approach, as these problems are surmountable. By following the recommendations outlined in this piece, the utility industry can get the best of both worlds–a stable, secure, and scalable power source, without the burden of building massive new power infrastructure.

About the author
David Maher is Chief Technology Officer at Intertrust and has over 30 years of experience in secure computing and is responsible for Research and Development at Intertrust. In addition, he is currently President of Seacert Corporation, a certificate authority for the Internet of Things, a developer of application security software, and Co-chairman of the Marlin Trust Management Organization which oversees the world’s only independent digital rights management ecosystem.