When Giants Dispute Over Resources: What's Really Going On?
According to reports published by EU Today, tensions arose between Google and Meta in March 2026. Google reportedly restricted Meta's access to its advanced Gemini series models. The reason was not an attempt to directly stop software competition, but capacity limitations. Meta was trying to obtain a larger volume of computational power than Google was able (or willing) to provide.
This incident is fascinating because it points to a new market dynamic. In the past, competition was defined by who had a better algorithm or a larger dataset. Today, competition is shifting below the level of the models themselves – into the realm of physical infrastructure. Even though Meta invests billions of dollars in its own data centers and chips, it still needs access to competitors' models for benchmarking, testing, and diversifying its services.
For comparison: While models like GPT-4o from OpenAI or Claude 3.5 Sonnet from Anthropic strive for maximum intelligence in every prompt, their real availability is directly conditioned by how many processors (GPUs) these models are currently able to "feed".
Hardware as the Newest AI Currency
The development of artificial intelligence is not just a matter of mathematics; it is a matter of physics. Training cutting-edge models requires enormous clusters of processors running continuously for months. Serving these models to millions of users (so-called inference) creates constant, massive pressure on infrastructure.
In this context, we see several key bottlenecks:
- Chips: The dominance of companies like NVIDIA in accelerators (H100, B200) creates extreme dependence on supply chains.
- Energy: Data centers are extremely energy-intensive. The fight for grid capacity is becoming a political issue – data centers compete for electricity with industrial plants and households.
- Cooling and Water: Keeping these systems running requires massive amounts of water and sophisticated cooling technologies, creating local ecological tension.
For companies using AI, this means that the price per token or API availability will not be determined solely by market demand, but primarily by the stability of energy and chip supplies.
European Dilemmas: Regulation vs. Reality
This dispute between Google and Meta has a much deeper impact on Europe than just a commercial dimension. While Brussels focuses on the EU AI Act, which defines rules for safety and ethics, it forgets that rules do not solve the hardware shortage.
European startups and state institutions are in a situation where they have no idea whether, by the time their systems are ready for deployment, they will have available computing capacity at a reasonable price. Most European AI projects today run on the infrastructure of American hyperscalers (Google Cloud, AWS, Microsoft Azure). If even the two largest players in the USA cannot agree on capacity, what chance does a smaller European developer have?
Sovereignty in the digital age no longer just means having your own data protected by GDPR. It means having the ability to operate critical systems on your own hardware and with your own energy. Without this, Europe remains merely a "regulatory sovereign" that sets the rules of the game but must play on a field built by someone else.
What Does This Mean for the Czech Market and Ordinary Users?
For Czech companies and developers, this trend brings several practical consequences:
- Multi-cloud strategy: Relying solely on one provider (e.g., only Google or only Microsoft) is now risky. Companies must build architectures that allow for easy migration between models (e.g., from Gemini to Llama or GPT) if one of them experiences a capacity outage.
- AI Costs: With increasing compute demands, prices for API services will likely increase. For Czech companies integrating AI into their products, it is necessary to account for higher margins to cover these costs.
- Localization and Czech language: Models like Gemini have excellent Czech language support, which is crucial for the local market. However, if access to the best models is restricted due to capacity, Czech companies may be forced to use weaker models (e.g., smaller Llama versions) that may not achieve the same quality in Czech.
Example of affordability: While for a regular user, Google Gemini (within Workspace) or ChatGPT Plus is available for approximately 20 USD (approx. 460 CZK) per month, for businesses, the price ranges in thousands of Czech crowns for millions of tokens, with this price being extremely sensitive to the current "compute bottleneck".
Conclusion
The Google-Meta dispute is a symbol of a new era. The AI race has shifted from pure research to industrial capacity. For Europe and the Czech Republic, this means that investments in digital infrastructure and energy security are just as important as investments in the algorithms themselves. If we want to be more than just passive customers in AI, we must start building our own "computational base".
Why would Meta want to use Gemini when it has its own Llama models?
Meta needs models from competitors for benchmarking (performance comparison) and as a safeguard. If their own Llama models were to encounter a technical trap or if they required a specific type of task, it is strategically advantageous to have access to the best available models on the market, even if they are from a rival.
Can a lack of computational power affect the speed of AI responses in Czech?
Yes. If data center capacity is limited, providers may implement "throttling" (speed limitation) or prioritize paying business customers over free users. This can result in longer latency when generating text in any language, including Czech.
What are the alternatives for Czech companies that fear dependence on the USA?
An alternative is to use open-source models (like Llama from Meta or Mistral) running on local servers or within European cloud services. However, this requires own investments in hardware (GPUs) and experts to manage these systems.
