VCs are betting billions on AI’s next wave, so why is OpenAI killing Sora?

Dollars are flowing in a vast stream toward generative video and multimodal models, but the foundations of this digital empire are beginning to crack in the face of physical reality. While Venture Capital funds outbid each other for stakes in the next unicorns, OpenAI faces a dilemma that could define the fate of their most high-profile project. The story of an 82-year-old Kentucky resident who rejected 26 million dollars for land intended for a data center is not just a local anecdote — it is a symbol of the material world's resistance to artificial intelligence's infinite appetite for resources.

The tension between virtual progress and physical limitations is becoming a central theme of debates in Silicon Valley. An industry accustomed to scaling software with almost zero marginal costs is suddenly discovering that AI infrastructure requires thousands of acres of land, gigawatts of energy, and the consent of local communities, which is not always for sale. Even as corporations attempt to push through rezoning plans for areas spanning 2,000 acres, individual resistance shows that the path to algorithmic dominance will not be as simple as investors assumed.

Physical resources versus OpenAI's ambitions

When OpenAI first unveiled Sora, the world held its breath. The model, capable of generating photorealistic video, promised a revolution in Hollywood and the creative industry; however, for months, the project has appeared to be at a standstill. The problem lies not only in the quality of the generated frames but in the brutal mathematics of computing costs. Every second of video generated by Sora requires computing power that translates into real electricity and water consumption in data centers — the very same ones whose construction is being blocked by landowners in Kentucky and other regions.

Market analysts are increasingly suggesting that OpenAI may be forced to drastically change its strategy regarding Sora. Maintaining a model that does not pay for itself while consuming resources essential for the development of the flagship GPT-5 is becoming a logistical nightmare. In a world where GPU access is rationed and the construction of new server rooms faces social resistance, priorities must be revised. It is not that the technology doesn't work — it's that maintaining it in its current form could be economic suicide in the face of rising infrastructure costs.

Capital seeks a new wave in the shadow of resistance

Despite these difficulties, VC investors are not slowing down, pumping billions into startups that promise to optimize model training processes. The industry believes that the next wave of AI will bypass infrastructure problems through better compression and more efficient algorithms. However, reality is testing these hopes — every new Claude or Gemini class model needs physical space on servers. It is a paradox: the more advanced artificial intelligence becomes, the more dependent it becomes on old-fashioned resources like land and copper.

We are currently observing a "pushback" from the real world that goes beyond ecology. It is a matter of local sovereignty and the right to decide on the landscape. Attempting to resolve and repurpose 2,000 acres of land without the residents' consent is a risky PR game for tech companies. If OpenAI indeed decides to sunset or drastically limit the Sora project, it will be a signal to the entire market: the era of unlimited expansion "in the cloud" without looking at the ground beneath one's feet has just ended.

• 26 million dollars — the amount rejected by a landowner in Kentucky.
• 2,000 acres — the scale of planned investments in data center infrastructure.
• Sora — the video model that has become a symbol of high AI operating costs.
• VC betting billions — the market still believes in a return on investment despite physical barriers.

Efficiency instead of raw computing power

In the face of infrastructure blockades, the vector of technological development must change. Companies like Anthropic and NVIDIA are increasingly emphasizing not just power, but above all, energy efficiency. If Sora is to survive, it cannot be a model that requires "owning a power plant." The industry must understand that the social resistance encountered by companies in Kentucky will grow with every subsequent attempt to expand into rural and suburban areas.

For OpenAI, the situation with Sora is a lesson in humility. You can create an algorithm capable of generating any image, but you cannot generate additional hectares of land or the goodwill of neighbors for humming server rooms. A strategic withdrawal from the most resource-intensive projects may be the only way to save the pace of work on text and multimodal models, which have a chance for faster monetization. The industry is entering a phase where a "no" from an 82-year-old woman carries more weight than another funding round from a San Francisco fund.

AI dominance in the coming years will not be decided in code, but in negotiations regarding access to power grids and building permits. Models like Sora may become victims of their own success — their potential is so vast that the costs of realizing it exceed the current capacity of Earth's infrastructure. We are headed for an era of optimization, where the winners will not be those with the most parameters, but those who learn to do more with less physical matter. If technology does not begin to respect the boundaries of the real world, the world will simply stop providing it with space to grow.