General Purpose AI / Foundation Models and Generative AI: Technical Analysis and Use Cases

On February 11, 2025, the European Union announced a new initiative to mobilize €200 billion for investments in artificial intelligence (AI) [1]. This ambitious move is driven by the rapid advancements and widespread adoption of foundation models, a field currently dominated by US-based companies. The term “General Purpose AI”, synonym of foundation models, was introduced in the 2023 AI Act by the EU to describe these versatile models, which are trained on extensive datasets and can adapt to a wide range of tasks [2].

In this article series, we explore the terminology and landscape of General Purpose AI, highlighting relevant use cases in the field of Energy Economics. General Purpose AI technology gained significant attention with the launch of ChatGPT, an AI chatbot developed by OpenAI and later integrated into Microsoft’s ecosystem as Copilot. These models have revolutionized the AI landscape, expanding its applications beyond specific use cases to a broad array of tasks. This adaptability is why the EU Commission refers to them as “machines capable of performing a wide range of intelligent tasks”, thinking abstractly, and adapting to new situations.

This emerging branch of AI presents challenges for energy infrastructure, as these models consume significant amounts of energy not only during their training but also during their usage. With the growing user base for such models, their energy consumption has become a critical consideration in terms of generation and distribution of electricity to meet this new demand.

The public release of ChatGPT in late 2022 marked a turning point in the field of artificial intelligence, transforming what had long seemed like an abstract technological promise into something tangible and widely accessible. Suddenly, millions of users were interacting with AI in everyday tasks—from drafting emails and summarizing documents to generating code and creative writing. This wave of interest came along with a flood of similar products, such as Google’s Bard, Microsoft’s Copilot integrated to Microsoft Office, and open-source projects like Meta’s LLaMA and Mistral AI’s Mixtral 8x7B, solidifying the presence of AI for the “large public”.

The roots of this breakthrough stretch back to 2017, when researchers introduced the Transformer architecture in a landmark paper titled Attention Is All You Need. This innovation enabled the creation of large language models (LLMs), capable of processing and generating human-like text at an unprecedented scale. As these models grew in complexity and capability—scaling from millions to hundreds of billions of parameters (visualization of the increasing size of foundation models over the last years) – they began to underpin a wide range of AI systems. Recognizing their broad applicability, researchers at Stanford coined the term “foundation models” in 2021 to describe these powerful models, trained on vast datasets and adaptable to many downstream tasks.

While the U.S. has so far led the development of these models, the field is becoming increasingly global, with new high-performance offerings emerging from China, such as DeepSeek, an open-source model that rivals the capabilities of its Western counterparts.

We will publish further parts of this article series on foundation models in the coming weeks. Their focus will be on locating foundation models in the AI landscape and walking through their technical operating principle step by step. We will further analyse why this technology is considered such an immense progress in AI research and describe the main factors that led to this breakthrough in the last 5-10 years.

Applications in Energy Economics

Traditional AI technologies, such as machine learning and deep learning, are already widely used in many applications within energy economics. Studies like a  have listed over 40 use cases in 2020, and more recently, a study  by PwC in 2024 identified 54 AI solution providers with over 60 products related to the energy sector [3,4]. As shown in Figure 2, existing AI applications cover a broad range of fields in energy economics.

Given the extensive use of AI-based applications, what changes can we expect with the introduction of foundation models, and what specific impacts or new use cases will these technologies bring?

Improve existing AI solutions

Foundation models have already demonstrated their ability to improve state-of-the-art traditional AI approaches. They offer the potential to enhance any application where AI has been used so far. Promising energy-related areas listed in a EESC study from 2025 [5] include:

• Predictive Maintenance
• Grid Management
• Battery Manufacturing

New Use Cases and Automation

Foundation models, particularly large language models, outperform previous AI approaches in several areas, enabling automation that was previously challenging. Their impressive information retrieval capabilities and natural language understanding open new possibilities, such as:

• Processing unstructured documents automatically
• implementing chatbots for customer services
• Coding assistance for developers

These tools, while not specific to the energy sector, will significantly impact various processes in the coming years, unlocking new levels of automation.

Versatility and Speed of Deployment

A major breakthrough of foundation models lies in their versatility. Previously, it was necessary to identify a task that could be automated by data-driven models, gather data, and train a model for that specific task. Foundation models can adapt to a given use case with minimal adjustments and retraining, changing the way we interact with them. This speeds up the deployment of AI for new use cases. For example, based on existing models like ChatGPT, one can directly develop new AI applications tailored to their needs without the need for extensive infrastructure and model training.

Furthermore, the interaction through natural language processing (NLP) makes the use of these models accessible with minimal expertise. Previously, specific AI developer skillsets were required to use these models, whereas today a more high-level and user-centric expertise is generally sufficient to develop an AI tool on top of an existing foundation model.

Examples of Existing Foundation Model-Based Products in the Energy Sector

Apart from the famous ChatGPT by OpenAI, several products built on top of foundation models are already available on the market. The following are a selection of such AI products:

• ThinkOwl: Offers various solutions for the automation of customer processes, including spoken and written customer conversations and (partial) automation of ticketing systems.
• Elevait: Provides solutions for the automation of document processing and the management of incoming data, supporting both customer service and internal knowledge management.
• Energy Concierge: Utilizes a chatbot and various customer touchpoints, such as WhatsApp, to enable supplier switching within 24 hours.

More information incoming in our dedicated article: “General Purpose AI, Foundation Models, Generative AI: Applications in Energy Economics”

Challenges for the Energy Sector

The democratization of general-purpose AI heralds a wealth of opportunities for innovative tools and applications. However, the growing adoption of products based on foundation models brings significant challenges, particularly for the energy sector.

Energy Consumption

Energy consumption has always been a significant cost component of developing AI models, with big tech companies building data centers equipped with high-end GPUs for training purposes. The era of foundation models, however, comes with a much higher scale of consumption. These models are not only more energy-demanding for training but also for inference (any request to the model), which was previously negligible. With the democratization of these models through apps like ChatGPT or Copilot, the number of requests has surged, making energy consumption for AI a serious consideration not only for tech companies, but even on an energy system scale, both in terms of generation and distribution of electricity.

The energy demand of data centers in Europe already correspond to the electricity demand of Austria. This demand is expected to more than double again by 2030 [6]. This will then correspond to around 5% of the current electricity requirements of continental Europe.

Impact on Grid Infrastructure

Training large AI models can place considerable stress on the energy grid. Studies like the one from the University of Chicago in 2024 warn about the ‘AI Gold Rush’: massive investments in new data centers driven by the interest in foundation model products. Their projections indicate that the concerned grid infrastructure will have to rapidly evolve to meet the new data centers’ load growth [7]. As part of the new InvestAI EU initiative, €20 billion is planned for new AI data centers. Building these training data centers in Europe might become a considerable challenge for grid operators.

The capacity of European data centers amounted to 10 GW in 2023. This is roughly equivalent to the peak load of Austria. It is predicted that the output of data centers will more than triple by 2030, driven by general purpose AI training and usage [6]. This output of 35 GW corresponds to around 9% of the peak load of continental Europe.

Additional Challenges

Other challenges worth mentioning include data privacy and the lack of EU-based foundation models. A more exhaustive list of challenges and descriptions will be available in our upcoming dedicated article.

If you’re interested in more information about your specific use case or this topic in general, please get in touch with us.

Literature:

[1] EU launches InvestAI initiative to mobilise €200 billion of investment in artificial intelligence – European Commission

[2] General-purpose artificial intelligence

[3] pwc-studie-ki-softwareloesungen-in-der-energiewirtschaft.pdf

[4] Pub_20200624_Kuenstliche-Intelligenz-fuer-die-Energiewirtschaft.pdf

[5] generative-ai-and-foundation-models-eu-uptake-opportunities-challenges-and-way-forward.pdf

[6] the-role-of-power-in-unlocking-the-european-ai-revolution.pdf

[7] Exploding AI Power Use: an Opportunity to Rethink Grid Planning and Management | EPIC