The Portuguese Competition Authority (AdC) has published a new short paper examining competition concerns related to access to chips used in the training and execution of Artificial Intelligence (AI) models. The paper forms part of the AdC’s ongoing analysis of generative AI and its impact on digital markets.
Access to AI chips — as well as related hardware infrastructure, particularly through cloud computing services — is a critical input in the development and deployment of AI systems. The AdC notes, however, that the AI chip value chain displays structural characteristics that make it inherently prone to high levels of concentration.
AI chip production depends on a complex and globalized value chain. Both the design and manufacturing stages exhibit significant concentration, largely due to structural economic factors such as substantial economies of scale, high fixed investment costs, technological sophistication, and the need for advanced manufacturing capabilities.
These structural features may create constraints along the supply chain, leading to supply rigidity and elevated prices for AI chips. In turn, such conditions can reinforce barriers to entry and expansion, particularly for startups and AI developers seeking to train and deploy advanced models. The concentration observed upstream may therefore have downstream implications for competition in AI model development and related markets.
The AdC further highlights the increasing vertical integration within the AI ecosystem. Leading chip designers and major cloud service providers have expanded their activities across different layers of the AI value chain, developing integrated ecosystems that combine hardware, software, and cloud services.
Simultaneously, the sector has witnessed an increase in strategic partnerships and cross-investment arrangements among leading market participants. The AdC notes that such developments may give rise to several competition concerns. In particular, these arrangements may facilitate privileged access to sensitive technical and commercially valuable information, contribute to the alignment of incentives among operators, and strengthen technological dependence through lock-in effects. They may also reduce interoperability between competing systems and increase the likelihood of foreclosure strategies, including the leveraging of market power into adjacent or downstream markets.
The AdC cites as an example CUDA, the development platform of NVIDIA. The widespread adoption of CUDA and the associated network effects contribute to strengthening NVIDIA’s position in the AI chip market, potentially increasing switching costs and reinforcing technological dependency.
The paper also considers the potential role of public high-performance computing (HPC) infrastructure in mitigating supply-side bottlenecks. Access to publicly funded computing resources could, in principle, alleviate constraints faced by smaller market participants.
However, the pro-competitive impact of such infrastructure depends critically on the establishment of transparent, objective, and non-discriminatory access criteria. Without appropriate governance mechanisms, public HPC resources may fail to deliver meaningful competitive benefits.
This short paper forms part of the AdC’s broader analytical work on generative AI. It follows previous publications addressing data access and use, the opening of AI models, AI labour markets, and a comprehensive Issues Paper released in 2023.
Through this series of publications, the AdC aims to identify emerging risks to competition in digital markets at an early stage, thereby promoting informed and preventive action consistent with its enforcement mandate.
As generative AI continues to reshape digital ecosystems, access to essential inputs — including AI chips — is likely to remain a focal point of competition policy scrutiny.