Generative AI’s Competition Challenges: Market Power, Data Control, and Regulatory Gaps

Decoding Competition Concerns in Generative AI

Generative AI is poised to add trillions to the global economy in the coming decade. This massive growth potential has caught the attention of competition authorities worldwide, who are keen to prevent market imbalances early on. Key issues include high entry barriers, network effects leading to market tipping, the integration of foundation models within dominant digital ecosystems, and the dependency on major players for essential infrastructure.

The Upstream Infrastructure Layer

The foundation of generative AI lies in three critical inputs: computing power, data, and skilled talent. Additionally, access to venture capital is vital to fuel innovation, especially given the high sunk costs involved in AI development. Silicon Valley’s thriving AI scene owes much to abundant VC funding, which quadrupled for AI startups between 2022 and 2024, reaching $25.2 billion.

Computing Power

Traditional CPUs struggle with the demands of training large language models, which led to the rise of GPUs, originally designed for gaming. Nvidia, a key player, has become central to this hardware shift. Its GPUs powered ChatGPT’s training, and its valuation soared from $300 billion in 2022 to $2.3 trillion in 2024.

While Nvidia has driven innovation, it faces scrutiny for exclusionary practices, including its acquisition of AI optimization software Run:ai. The company outsources manufacturing to Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung Electronics, with TSMC relying on ASML, the only supplier of critical EUV lithography machines needed to manufacture cutting-edge GPUs.

Nvidia has also built an ecosystem around its GPUs with CUDA software, enabling developers to harness processing power efficiently. Despite holding over 90% market share, Nvidia faces emerging competition from Google, Amazon, Microsoft, Meta, and IBM, all investing in developing their own AI chips.

Cloud computing, another essential infrastructure, is dominated by Amazon Web Services, Microsoft Azure, and Google Cloud Platform. High costs, technical barriers, and restrictive practices contribute to this concentration. The increased demand for cloud services due to generative AI only strengthens their market positions.

For instance, OpenAI’s GPT models are exclusively trained on Microsoft’s Azure cloud, highlighting close collaboration that extends influence across the generative AI value chain. This raises the question of whether competition rules like the EU’s Digital Markets Act (DMA) should be expanded to better regulate cloud services as essential infrastructure.

Data

Data is the lifeblood of generative AI. Major firms like Google, Amazon, Microsoft, Meta, and Apple (GAMMA) control vast and valuable data sources spanning search, social media, e-commerce, productivity tools, and more. This gives them a feedback loop advantage: more users generate more data, improving models and attracting even more users.

Generative AI models often train on massive datasets scraped from the web, sometimes without explicit consent, sparking copyright and ethical debates. The EU’s 2019 Copyright in the Digital Single Market Directive provides a licensing framework enabling lawful access to copyrighted data for AI training, which is crucial for fostering competition.

The DMA also requires gatekeepers to grant fair, reasonable, and non-discriminatory access to aggregated data, including personal data, under GDPR compliance. This can level the playing field, allowing smaller AI firms to compete despite the data advantages of incumbents. Advances in algorithm efficiency further widen the gap, as newer models need far less data to perform well.

Talent Acquisition (Acqui-Hires)

Technical talent is critical for AI innovation. Many researchers are hired by industry, and sometimes companies acquire startups mainly to gain their teams. The Microsoft acquisition of Inflection’s key personnel raised concerns about effective market control through talent acquisition.

While some such deals pass regulatory review, they expose gaps in current merger control frameworks, especially when acquisitions fall below turnover thresholds but still impact competition by controlling unique expertise.

The AI Modelling and Deployment Layers

Once infrastructure inputs are secured, the focus shifts to training foundation models. These can be general-purpose or specialized for specific industries. Specialized vertical models often outperform general ones in their niche. Building a general-purpose AI model from scratch is extremely costly—estimates suggest up to $8.8 trillion.

Most foundation models today are open source, but large firms tend to shift towards closed models over time. The deployment layer involves integrating AI into user-facing products, such as Microsoft’s Co-Pilot or Meta’s AI-powered features on social platforms.

Regulatory intervention through frameworks like the DMA could encourage competition here. One approach is to amend the DMA to explicitly cover generative AI platforms. Another, more practical path is to apply existing DMA rules to generative AI features embedded within already designated gatekeeper platforms. This approach could enable timely and effective oversight without complicating enforcement.

Conclusion

Generative AI is reshaping markets with significant competition challenges at every layer—from infrastructure and data to talent and deployment. Regulatory bodies are exploring ways to prevent dominant players from locking in unfair advantages. Ensuring open access to essential inputs like cloud services and data, coupled with vigilant merger control, will be key to fostering a competitive and innovative AI ecosystem.

For professionals looking to deepen their understanding of AI and its market dynamics, exploring dedicated courses can offer valuable insights and skills. Check out Complete AI Training’s latest AI courses for practical knowledge in this rapidly evolving field.