AI development risks becoming a theatre for great-power rivalry. Critically analyse the implications for the Global South.

Frontier AI capability — chips, compute and model weights — is concentrated in a handful of firms and two states, turning a general-purpose technology into an instrument of statecraft. For the Global South this rivalry is simultaneously a dependency trap and a bargaining opportunity; which one prevails depends on the institutions it backs.

How the rivalry is hardening

• Governance is splintering into competing tracks: the GPAI, launched in 2020 with India a founding member and later integrated with the OECD [1], now sits alongside BRICS-track proposals for bloc-owned open-source stacks, aired around the 18th BRICS Summit hosted by India in New Delhi, 12–13 September 2026 [2].
• The binding constraint is compute, not ideas. India's subsidised common pool crossed 34,333 GPUs — 18,417 empanelled plus 15,916 with industry partners [3] — sized for fine-tuning and inference, not frontier pre-training.

Adverse implications

• Supply insecurity: models and accelerators can be withdrawn or export-controlled by vendor or state decision, stranding deployed public services with no notice obligation.
• Norm-taking, not norm-making: UNESCO's Recommendation on AI Ethics, adopted by 193 countries in 2021 [4], and the Global Digital Compact, adopted by 193 states in 2024 [5], remain recommendatory — no inspection, reporting or enforcement.
• Forum proliferation: each bloc-branded initiative adds a venue rather than a rule, and open weights relocate dependence upstream to whoever funds the training run.

The countervailing opportunity

• Numbers favour universal forums: the Compact's 40-member Independent International Scientific Panel and the Costa Rica–Spain-facilitated Global Dialogue give middle powers a seat blocs do not [5].
• Norm supply: India's AI Governance Guidelines (MeitY, 2025), framed to foster international cooperation [6], can be tabled as negotiating text; shared compute access offers partners a deliverable no communiqué matches.

Non-alignment in AI must therefore be active, not abstentionist: the Global South should invest in universal, evidence-based governance, pool compute regionally, and seek continuity safeguards on vendors. Anchored in the SDGs and human-rights frameworks, such a path converts India's hosting role and adoption scale into rule-making leverage rather than bloc loyalty.