Talk of a ‘compute divide’ won’t close it — time for funds, compute and seats

At the ITU Plenipotentiary Conference 2026 in Doha, Malaysia’s Communications Minister Datuk Seri Fahmi Fadzil warned that artificial intelligence is creating a “compute divide” distinct from the old connectivity gap; our argument: naming the problem is necessary but not sufficient — unless the ITU, wealthy member states, the private sector and regulators translate diagnosis into specific pooled finance, subsidised regional compute capacity and formal implementation roles for developing countries, the gap in practical AI capability will widen and standard‑setting will favour those who already control large compute stacks.

Compute is the new chokepoint

Fahmi’s diagnosis is straightforward and is recorded in Bernama’s reporting: the problem goes beyond internet access to who can train, host and run large models. ITU secretary‑general Doreen Bogdan‑Martin’s reminder that AI can evolve significantly in a year underlines how quickly compute advantages compound. Qatar hosted ITU PP‑26 alongside the Mobile World Congress Doha 2026, events Bernama said enabled engagement between ministers, telecommunications regulators and industry. Convening ministers and regulators is a necessary function of the ITU, but the Bernama report highlights warnings and standards without mentioning transfers of compute capacity, concessional finance for data centres or explicit compute‑sharing commitments in support of the Global South.

Standards without implementation concentrate practical control

The background analysis in unite.ai sets out the mechanism clearly: regulators and standards bodies such as the EU (via the AI Act), U.S. enforcement agencies (the FTC, DOJ, CFPB and EEOC) and technical frameworks like NIST can define outcomes, but companies — and the organisations that operate the compute and deployment stacks — are the ones that translate those outcomes into thresholds, review rules and operational controls. Those operational decisions determine what counts as compliance in practice: when human review is required, what gets logged, and how uncertain outputs are handled. When operationalisation rests with organisations that own the compute and deployment pipelines, authority over how rules are applied accrues to them. Bernama’s coverage flags the risk; it does not report new international mechanisms that would rebalance practical access to compute for less resourced states.

Voice without resources is limited

Malaysia’s bid for re‑election to the ITU Council is sensible — seats and chairs matter for drafting standards, digital trust rules and interoperability norms — and Bernama lists Malaysia’s areas of contribution in those fields. Yet representation on standard‑setting bodies only influences outcomes to the extent that countries can implement them. As unite.ai emphasises, rules change behaviour only when organisations can translate those rules into working systems and then supply evidence that the systems deliver the promised outcomes. That translation requires hardware, power, cooling, secure data centres, trained engineers and often subsidised access to cloud resources or regional compute nodes; the Bernama piece does not record any pooled finance or binding delivery commitments announced at PP‑26 to supply those elements to the Global South.

The strongest counter‑argument — and why it falls short

Advocates of a standards‑first approach will argue that clear rules, technical frameworks and capacity‑building advice are the logical first step: clarify obligations, then investment will follow and domestic capacity can grow. That is partly right — regulation and technical guidance are prerequisites for safe deployment and for attracting responsible investment. But unite.ai highlights a practical flaw: creating a regulatory framework does not itself provide GPUs, subsidise regional training clusters, or staff and power data‑centre operations. When governments and international bodies set obligations but lack mechanisms that shift compute and implementation capacity, countries without local resources will remain dependent on better‑resourced organisations to demonstrate compliance to regulators — and that dependence preserves practical dominance with those who control the compute.

What should happen next is concrete. First, the ITU membership should debate and propose a dedicated “compute equity” mechanism: a pooled fund, financed by wealthier member states with voluntary contributions from private commercial actors, to underwrite regional cloud nodes, subsidised training clusters and energy‑efficiency upgrades for data centres in the Global South. Second, the ITU Council should tie standard‑setting to guaranteed implementation roles for Global South representatives — formal leadership or budget shares on bodies that oversee deployment pilots and technical assistance — so that voice maps to delivery capacity. Third, the EU’s policy‑makers, U.S. regulators and standards bodies such as NIST should make market access and public procurement contingent on demonstrable commitments by deploying organisations to support shared compute initiatives and to ensure data portability with local operators. Finally, hosts and organisers such as Qatar and the Mobile World Congress should use convening leverage to secure memoranda of support from participating firms that commit resources to regional compute pilots.

Fahmi was right to flag the compute divide at PP‑26; the next step is not more statements of concern but binding mechanisms that move money, regional compute capacity and implementation authority towards the Global South so that standards can be turned into on‑the‑ground capability rather than remaining obligations only the well resourced can meet.