Press Release 04/2026
February 20, 2026

The Portuguese Competition Authority (AdC) has today published a short paper examining competition issues associated with access to chips for training and running Artificial Intelligence (AI) models. Access to AI chips – together with related hardware resources, including those provided through cloud computing services – is a key input for the development of AI systems. The AdC’s analysis finds that the AI chip value chain displays structural characteristics that may give rise to concentrated market outcomes. 

Structural concentration and supply-side constraints 
The production of AI chips relies on a complex global value chain, with both chip design and manufacturing characterised by significant levels of concentration. This reflects structural features of the sector, including substantial economies of scale, high capital expenditure requirements and technological complexity. 
These characteristics may give rise to bottlenecks and supply rigidity along the value chain, contributing to higher AI chip prices. In turn, this may increase barriers to entry and expansion, particularly for start-ups and AI developers seeking to train and deploy advanced AI models. 

Vertical integration and integrated ecosystems 
The AdC also observes increasing vertical integration across the AI sector. Certain leading chip designers and cloud service providers have expanded across multiple layers of the AI value chain, developing integrated ecosystems combining hardware, software and services. 
In parallel, strategic partnerships and cross-investments have emerged between key market participants. These developments may have the potential to raise competition concerns where they: 

• provide privileged access to commercially or technically sensitive information; 

• align incentives across the AI ecosystem; 

• strengthen technological dependency and lock-in effects; 

• reduce interoperability; and 

• increase the risk of foreclosure or the leveraging of market power into adjacent markets. 

The paper highlights NVIDIA’s CUDA software stack as an example. Its widespread adoption, together with associated network effects, may reinforce the company’s position in AI chip markets. 

Public high-performance computing infrastructure The AdC notes that access to public high-performance computing (HPC) infrastructure may partially mitigate the identified constraints. However, any pro-competitive effects are likely to depend on access conditions being transparent, objective and non-discriminatory. 

A preventive and evidence-based approach 
This publication forms part of the AdC’s ongoing work assessing the competition issues related to generative AI. It follows earlier short papers addressing access and use of data, openness of AI models and AI labour markets, as well as an Issues Paper published in 2023. This series aims to identify emerging risks to competition in digital markets and to support the AdC’s proactive and informed enforcement strategy.