Original text published by No es época para tontos.
In 2025, the artificial intelligence race looked deceptively simple: OpenAI, Google, Anthropic, Meta and xAI were fighting to build the smartest model on Earth. By 2026, that race had fractured into something far bigger. The real battle is no longer only about code. It is about inference costs, semiconductors, cloud infrastructure, massive capital expenditure, electricity grids, data centers, regulation, national sovereignty and the physical territory required to keep AI running.
This investigation examines the moment intelligence itself begins to look like a commodity. For many business tasks, companies no longer need the single most powerful model in existence. They need a model that is good enough, fast enough, reliable enough and, above all, cheap enough. That shift changes the economics of the entire industry.
China is central to that transformation. DeepSeek, Qwen, Kimi, MiniMax and other open or open-weight models are pushing prices down and spreading through enterprise applications around the world. Platforms such as OpenRouter and Fireworks AI make it increasingly easy to switch between models in real time, routing each task to whichever system offers the best combination of price, speed and capability. If intelligence becomes interchangeable, the power may move away from the model maker and toward whoever controls the routing, infrastructure and distribution layer.
That is why Amazon and Microsoft occupy such powerful positions. AWS and Azure can make money regardless of which laboratory builds the leading model. Google and Meta play a different game, using AI to strengthen ecosystems they already dominate. Meta can spend enormous sums on infrastructure because AI can improve advertising, recommendations, content and messaging without being sold as a standalone product.
At the center of the story is the financial tension facing frontier labs such as OpenAI and Anthropic. Anthropic has grown rapidly through coding tools and enterprise agents, where customers are willing to pay and results can be tested. OpenAI shows the other side of the equation: explosive revenue growth does not eliminate the extraordinary cost of training, serving and scaling advanced AI. Revenue is not profit, and inference is not free. Every prompt, image, video and autonomous agent action consumes real compute.
That leads directly to NVIDIA. The company is no longer just a chip supplier. It sits at the heart of an infrastructure loop in which capital finances data centers, data centers buy massive computing systems, those systems rely heavily on NVIDIA hardware, and rising demand helps justify even more investment. The same loop can look like a virtuous cycle if demand keeps growing, or a serious financial risk if future revenues disappoint.
Then comes the physical bottleneck. ASML makes the extreme-ultraviolet lithography machines needed to manufacture the most advanced chips. Those chips go into servers. Those servers go into data centers. And those data centers need land, substations, transmission lines, cooling systems, water rights, backup power and enormous amounts of electricity. The “cloud” is not weightless. It is concrete, steel, cables and megawatts.
The investigation also moves into geopolitics. Advanced models are increasingly treated as strategic assets. Disputes involving Anthropic, the U.S. Department of Defense, cybersecurity, biology and export restrictions raise a fundamental question: who decides how powerful AI systems may be used? The laboratory? The client? The military? The government? The courts?
And then there is the bubble debate. Hundreds of billions are being committed to AI infrastructure. The risks are obvious: long-term contracts, circular financing and expectations of extraordinary future demand. But unlike many companies from the dot-com era, today’s largest technology investors generate enormous real cash flows from cloud, advertising and software.
The final paradox is OpenAI itself. A company could theoretically win the technological race and still become increasingly dependent on the cloud providers, chipmakers, financiers and power grids that make that victory possible.
The throne of artificial intelligence has not disappeared. What has changed is what it is made of.
It is no longer built only from code.
It is built from capital, silicon, electricity, territory and geopolitical power.
And that leaves one final question:
If intelligence becomes abundant, commoditized and cheap, what becomes scarce enough to control the future?
