Artificial intelligence has ceased to be a technological race and become a battle for control. Beyond the competition between blocs (USA, EU, China) we discussed yesterday, governments and major corporations are building what Accenture calls "Sovereign AI": the national capacity to retain data, train models locally, and build proprietary infrastructure.
Control, Rules, Value
Sixty-one per cent of business leaders already consider it critical. Europe leads the charge: sovereign clouds (AWS, Microsoft with EU options), national champions like STACKIT, consortia such as "AI Gigafactory" spanning 16 countries. Spain is investing €2.1 billion (2024–2025) focused on ALIA—models in Castilian and co-official languages—, reinforcing supercomputing capacity (MareNostrum 5) and leading regulatory efforts through AESIA. Saudi Arabia converts oil into compute; Canada launches its own "AI Factory". What is at stake: who controls data, who writes the rules, who captures value.
Social AI
On the other side, an alternative is emerging from the ground up. Plantix (India) already operates in 16 local languages with 87 per cent of users improving agricultural decisions. Teacher.AI—a WhatsApp chatbot co-designed with teachers in Sierra Leone—accumulated 40,350 queries in 17 months. BabyChecker (Guatemala) supports midwives in offline fetal monitoring with 285 women screened. These are not "communitarian decentralised systems" but rather global solutions adapted locally. What is novel is that anyone can now do this: LLaMA and Mistral (open-source under Apache 2.0) allow developers to download models, adapt them to local data in specific languages, and deploy them without relying on OpenAI's proprietary APIs.
The barriers to entry have collapsed. DAOs (Decentralized Autonomous Organizations)—structures governed by smart contracts on blockchain—enable communities to make decisions collectively through tokenised voting. Platforms such as SingularityNET and Gaia use DAOs to allow independent developers to build AI collaboratively, distributing value through tokens.
Breaking Moulds
What is striking is that both strategies respond to the same problem: 70 per cent of leading models originate from the USA, 25 per cent from China. Real concentration, real risk. They differ only in who should resolve it.
Dario Amodei (Anthropic) champions in his essay "Machines of Loving Grace" a highly controlled and centralised Artificial General Intelligence (AGI), arguing that responsible development requires rigorous oversight. Ben Goertzel (SingularityNET) warns the opposite: that centralisation perpetuates the biases of its creators, that intelligence must emerge from distributed collaboration. One points to uniformity, the other to plurality.
Multipolarity
From this struggle between state sovereignty and distributed governance emerges not a single victor, but multipolar AI in which multiple actors—states, enterprises, communities, consortia—negotiate who controls what. Europe plays for ethical regulation. India for domestic semiconductors. Saudi Arabia for infrastructure. Communities for open models. No one wins everything; all negotiate fragments of the future. Whoever waits indefinitely inherits the rules others have written.
This article is republished from Futuribles. The original article in English and Spanish is here.
