Artificial intelligence (AI) is rapidly emerging as a general-purpose technology with the potential to transform the economy. Its development depends heavily on advances in computing hardware (compute), which enables the training of increasingly complex foundation models and their deployment at scale.
The AI compute infrastructure supply chain is multi-layered and spans several interrelated markets. At its core are the chip designers and manufacturers, who produce the essential hardware powering AI data centres. These firms rely on specialised suppliers of materials and equipment, which are critical to the chip production process. The chips are then integrated into fully functioning server racks and deployed by data centre operators. These operators also rely on the energy and networking infrastructure required to power and interconnect the individual servers within a data centre, as well as link multiple data centres together. Once operational, cloud providers sell access to this compute power to AI developers and modelers, enabling them to train, and deploy increasingly sophisticated AI systems.
Despite covering a range of markets there are several common features which have important implications for competition:
rapid innovation and dynamic market evolution with high levels of research and development
high concentration and barriers to entry meaning at each level there are often only a few suppliers
vertical and conglomerate integration is increasingly found across the AI compute supply chain
increasing levels of crossholdings and investment partnerships
high levels of state intervention including public investment and trade barriers
high levels of demand, often outstripping supply.
Competition enforcement action across the OECD Members has been limited so far, even if some investigations are currently underway. Potential competition concerns in the AI compute sector are varied and evolving. For example, firms with significant market power may engage in exclusionary practices, such as bundling products in ways that disadvantage competitors. Market power may be also expanded through less traditional means, including acqui-hires and strategic partnerships that fall below merger notification thresholds. In addition, as certain layers of the AI stack mature and become commoditised, the risk of collusion may increase, particularly where fewer players dominate. Beyond conduct the strategic significance of AI to national competitiveness and security has led to increasing levels of government intervention across the supply chain. With industrial policy initiatives potentially shaping market dynamics in areas such as chip fabrication, data centre development and access to compute.
Competition authorities will likely need to respond using the wide range of tools available. This may start with building technical expertise across the AI compute stack through research and recruitment to enable timely and effective interventions. A particular challenge will be ensuring merger control regimes are flexible enough to capture acquisitions of nascent competitors, and scrutinising conglomerate mergers for expansion risks. This includes examining cross-layer partnerships to ensure they are not used to foreclose rivals or distort competition. Competition agencies may also consider advocating for pro-competitive policies, such as supporting public investment in open-source technologies, public compute resources, and infrastructure to help overcome economies of scale and enable competition to flourish.