The global battle for AI sovereignty: The race to build ever-larger AI models, control the compute infrastructure to support them, and — above all — hold the dominant “model weights” is fast becoming a key battleground of strategic competition. Just as control over ports, railways or pipelines once shaped geopolitical influence, control over AI’s deep memory circuits and “mental infrastructure” is now emerging as a site of sovereignty. Nations can no longer treat AI as merely a commercial or technological domain — it is central to power, identity and influence in the 21st century.
Without deliberate strategy, citizens’ cognitive life will come to be dominated by a handful of foreign providers. The levers of export control, alliance blocs, and selective partnerships will determine who leads, who follows, and who is merely a consumer. In this competition, India must chart a balanced path: resist technological hegemony, preserve autonomy, and yet not be isolated.
READ I Europe’s far-right find happy hunting grounds in social media
Compute, models and mental infrastructure
Historically, digital competition emphasised data, algorithms, networks. But the frontier has shifted: model weights — the numerical parameters of huge neural nets — have become the locus of value. Export rules in 2025 explicitly treat model weights as controlled technology. Under the new US Diffusion Framework, closed-weight models trained above a threshold (on the order of 10^26 FLOPs) now require licenses for export. In effect, the weights are the crown jewels of AI—the memory, the capacity, the ideological and cognitive power.
Control over compute infrastructure — data centres, high-performance GPUs, interconnects, power supply — is the backbone that anchors those weights. A nation that lacks sovereign compute becomes dependent on foreign estates and rules. Further, the model providers’ ability to embed memory, fine tuning, and personalisation creates cognitive lock-in: users become entwined with a provider’s memory graph, hard to migrate away, and gradually shaped by its biases and incentives.
Thus, AI is no longer simply a competitive technology sector; it is mental infrastructure — the substrate of future thought, belief, persuasion and societal calibration. Whoever holds the weights and infrastructure commands the narrative, shapes judgment, and in time may influence choice.
Export controls as power levers
In this new contest, export control regimes are no longer niche trade tools but instruments of strategic deterrence. The United States has rolled out tiered restrictions combining high-end chip limits and licensing requirements for advanced model weights. Only a select group of 18 allied countries gain privileged access; others find themselves subject to compute caps or blocked transfers.
These rules also incorporate a “foreign direct product” extension: model weights generated outside the US but using US technology may still be counted as controlled exports. The logic is clear: the US seeks diffusion control — to decide who inherits advanced AI capacity.
Critics warn this can backfire. Overly aggressive export restrictions may provoke nations to develop decoupled ecosystems, eroding US leadership in the long run. Indeed, Chinese researchers claim to have released advanced weights (DeepSeek) using pre-existing resources, partly circumventing constraints. The balances are perilous: too lax, and control slips away; too rigid, and innovation migrates.
Complementary to export control is emerging compute oversight. Proposals for “Know-Your-Customer” rules for compute providers would require identity vetting of buyers of large cloud workloads. That creates a governance choke point over who may access large compute quotas in practice, beyond just hardware shipments.
Strategic blocs: US-EU vs China
In this new environment, alliance blocs are paramount. The US and its allies aim to cluster friendly nations into “compute trust zones,” and define standards, licensing exceptions, and trusted supply chains. The EU, in turn, pursues digital sovereignty and variant regulatory instruments like the AI Act, but remains structurally dependent on US foundational models and hardware. Tensions already surface in regulatory style and industrial policy.
On the other side, China is backing a competing governance model: closed standards, domestic compute, firewalled data, and the extension of influence via its Digital Silk Road. China is massively investing in chips, memory, cloud platforms, and AI full stack. The result is an AI Cold War: blocs aligning around compute, model weights, and standards. Smaller powers may be forced to choose sides or remain squeezed in the middle.
AI sovereignty: Why domestic control matters
It is not enough to build data centres or host clouds. The question is: who controls the memory — the model weights, the fine-tuned preferences, the bias and pedagogical choices? If citizens depend on foreign AI providers, they surrender de facto influence over their own thought environment.
This is analogous to how foreign media, platform monopolies or social networks can skew public discourse. But AI is more intimate: over time, the assistant may know a user better than they know themselves, influencing suggestions, nudges, worldview. National memory lock-in may lead to algorithmic soft power—cultural influence in code, not flags.
Hence, cognitive sovereignty demands three layers of protection:
Sovereign compute & memory — domestic infrastructure and legal guarantee of residency and control.
Portability and auditability — mandates that memory graphs, model weights, and personal embeddings be portable and auditable.
Strategic guardrails and alliances — not total isolation, but trusted partnerships.
India seeks to balance autonomy and partnerships
India occupies a distinctive position: data rich but infrastructure poor. While generating about one-fifth of global data, India holds only about 3 per cent of global data centre capacity. It cannot afford isolation, but cannot complacently cede cognitive space either.
First, India should invest aggressively in sovereign compute clusters, edge memory vaults, and trusted AI stacks. Domestic compute cannot guarantee sovereignty if built and operated by foreign entities — indeed, 48 per cent of non-US data centre investment lies under US companies. India must insist on trusted ownership, control, and auditing.
Second, it should negotiate bilateral and multilateral memory alliances with like-minded nations — for instance, bundling with the European Union, Japan or Australia to gain reciprocal access, joint laboratories, and license waivers. Given US export tiers, India must wisely seek inclusion in “trusted” circles or negotiate carve-outs.
Third, it should champion open-source & federated AI models as a hedge against capture. Open weights, federated learning, and user-hosted memory would help break cognitive monopoly. India must insist on portability, audit trails, and compliance transparency in any outsourced system.
Fourth, the country should engage in export control architecture design, rather than being passive. For instance, in forums such as the Quad, G20, or in UN AI governance, India can push for rules that preserve the rights of developing nations to host, share, and regulate cognitive infrastructure. The United States is positioning an “Export Promotion Strategy for the Global South” to tie AI diffusion to diplomacy. India must avoid being a mere recipient; it must be a rule-maker.
Finally, it should encourage domestic AI start-ups to open-license foundational models, espouse ethical guarantees, and collaborate in federated networks across the Global South. In doing so, India can build membership in a cognitive commons rather than a captive consumer market.
Risks, trade-offs and the road ahead
The path is fraught. Excessive rigid control may stifle domestic innovation or provoke decoupling. Overreliance on foreign partners even within alliances may expose vulnerabilities. The choices around compute scale, model licensing, audit regimes, and partner trust must be calibrated.
Yet delay is not an option. The weight of a model is not merely a technical file — it is a strategic proposition. Nations that fail to reckon with the cognitive dimension will find themselves as vassals in a new intellectual order. India must act with foresight and boldness, anchoring its own cognitive sovereignty while participating in responsible alliances that guard against both hegemony and fragmentation.