IDED Publication: Enhance regional AI competitiveness through an AI computing power exchange

Start

Publications

Künstliche Intelligenz

Analysis

Germany

Digital Infrastructure

Enhance regional AI competitiveness through an AI computing power exchange

Thursday, June 5, 2025

To enhance competitiveness in the AI sector, it is proposed to establish a trading platform for AI computing power. This would facilitate access to affordable computing capacity and stimulate competition among providers. Such a platform could reduce costs for AI infrastructure and help small businesses participate in the market. Profits from the platform could be reinvested in research and development.

Download

More information

AI requires computing power. More than any other digital technology. This computing power must not only be available but also cost-effective to give AI companies a chance to be profitable and operate sustainably. To reduce the cost of computing power and increase availability, competition is needed. AI computing power is a homogeneous commodity. To foster competition, we propose establishing a trading exchange where AI companies can meet their demands and IT infrastructure providers can offer their produced computing power.

A study by a renowned investor shows that, even before AI technologies, up to 50% of the revenues of digital companies are spent on infrastructure costs. The high losses of leading AI companies, which are temporarily absorbed by venture capitalists, also clearly demonstrate the importance of the cost of computing capacity for AI companies.

The power exchange (EEX) serves as a sensible model for such an exchange. There, a company can cover its electricity needs at the best possible price. An exchange offers contracts, known as trading agreements, in the case of AI, for computing power. If a company knows its computing power requirements in advance (for example, if it regularly trains an AI model on new data - 'finetuning'), it can purchase futures contracts to secure the price for a period ('hedging'). Additionally, a day-ahead market (trading today for delivery tomorrow) and an intraday market (trading and delivery on the same day) are sensible for computing power. Companies can purchase capacity on-demand and sell it in case of non-use.

Such an exchange creates important price signals, which form the basis for regulatory and political decisions as well as investment decisions by investors. For example, a medium-term rising price for futures contracts, with a duration of 6 to 12 months, can signal the need to create new capacities. Market volatility also arises from the price of electricity used to generate AI computing power. If this exceeds a price that is economically viable for infrastructure providers, capacity will be withdrawn from the market. Similarly, competition in chip technologies for AI can lead to new types of efficient chips entering the market, making older technologies less economical.

Fundamentally, the exchange creates competition, which, in the short and medium term, ensures a reduction in the costs of AI infrastructure. AI companies that source their capacity through the exchange have a competitive advantage, as the product costs for AI services decrease equally. This presents an opportunity for states, as many states invest significantly in new AI companies and services. The exchange would now be the pivotal point for purchasing computing capacity.

To this day, there is no IT infrastructure provider in Germany that can offer the necessary amount of AI computing power, for instance, for model training. In the states, there is no single IT infrastructure provider large enough to meet the demands of even a handful of AI companies. The exchange solves this problem by acting as an aggregator of all available capacities. It thus creates more opportunities for small and medium-sized infrastructure providers to participate in the market, create innovative solutions for efficient provisioning, and participate in the AI growth market.

Additionally, should the exchange be (partially) state-owned, the profits from the trading business can be incorporated into the state budget and further invested in research and development, such as for the development of better AI chips.

The technical feasibility has already been realized through the EU's large-scale investments via the 'Important Projects of Common European Interest' (IPCEI). The EU has supported and brought to market maturity core technologies, such as software for the decentralized and federal provision of capacity, with the IPCEI program. The appendix lists some of the projects from the program. What is missing is a competent market design as the basis for trading on capacity exchanges.

In the long term, it would also be conceivable to further stimulate demand through public organizations' procurement. They could mandate that the computing capacity necessary for the operation of AI applications be procured through the exchange when purchasing AI solutions.

In the short term, it is especially important to ensure that new AI startups have the necessary incentives to use the exchange. In the current market environment, large international providers often offer free computing power. For example, Google offers AI startups up to EUR 350,000 in free computing power to ensure that AI applications are built on Google's own closed exchange. This should be prevented at the national and EU level; until this happens, however, the profits from trading on the exchange should partially be used to create equivalent incentives on the open exchange. This may need to be realized with seed funding at the establishment of the exchange from tax revenues, which should, however, be recouped through the trading profits in the medium term. This seed funding is necessary to attract startups in the AI market to the open exchange.