On May 15, 2026, during Narendra Modi's state visit to Abu Dhabi, India's Foreign Secretary Vikram Misri and G42 International CEO Mansoor Al Mansoori exchanged a commercial framework for Condor Galaxy India — an 8-exaflop AI supercomputer of 64 Cerebras CS-3 systems, to be installed and operated by G42's unit alongside India's Centre for Development of Advanced Computing (C-DAC). The signing was witnessed by UAE President Sheikh Mohamed bin Zayed Al Nahyan and Modi himself (PR Newswire). The ceremony mattered less than the model it ratifies: a Gulf sovereign-wealth vehicle exporting US-designed AI compute, hosted on a third country's soil, under that country's rules.
A third path, not a bypass
Most coverage framed the deal as India 'bypassing' Amazon, Microsoft, and Google. That overstates it. India already holds roughly $45 billion in cloud commitments from the three US hyperscalers, and its $1.25 billion national AI program runs on Nvidia silicon — about 34,000 GPUs available now, with a target of 100,000 by year-end (Rest of World). Condor Galaxy India does not replace any of that. It adds a second supply line — machines on Indian soil, run by a non-US partner, with data staying under Indian jurisdiction.
This is the right reading of 'sovereignty' in AI infrastructure: not autarky, but optionality. The concentration problem is real. Amazon, Microsoft, and Google together account for nearly two-thirds of global enterprise cloud infrastructure spending (Rest of World). When three vendors control the substrate of national AI capacity, every government rents its strategic compute from companies answerable to a foreign regulator and a foreign export-control regime. A buyer with a credible alternative negotiates better terms even if it never switches. That is the pro-competition case for the deal, and it is a strong one.
Steelman the skeptics
The strongest objection is not protectionism — it is governance and security. A national-scale compute cluster operated by a foreign sovereign's commercial arm concentrates extraordinary capability under an opaque ownership chain. G42 is backed by Abu Dhabi's Mubadala; its prior entanglements drew enough US scrutiny that it unwound Chinese holdings before Microsoft's 2024 investment. Critics reasonably ask who can see the workloads, who controls the off-switch, and whether 'data stays in India' survives contact with the operating partner's privileged access. The UAE's own record on digital rights sharpens the worry: in May 2026, Meta restricted human-rights NGO and researcher accounts from reaching audiences in the UAE and Saudi Arabia at governments' request (Access Now). A state comfortable leaning on platforms to geo-block dissent is a state whose compute exports deserve hard questions, not reflexive welcome.
Those questions argue for disclosure and contestable contracts, not for blocking the deal. The proportionate response is to regulate the governance layer: mandatory transparency on operator access rights, audited logging, a clear sovereign kill-switch vested in C-DAC rather than the foreign operator, and published terms for what 'data sovereignty' contractually means. India's IndiaAI Mission, which Condor Galaxy India slots into, is the natural vehicle for codifying those guarantees. Regulation that forces the contract into daylight protects the public interest without forfeiting the capacity.
Why the export model beats the alternatives
The realistic counterfactuals are worse. India could wait years to build comparable capacity domestically, ceding model-development ground in the interim. Or it could deepen its dependence on the same three hyperscalers it is trying to diversify away from. The G42 route delivers exaflop-scale compute now, on home soil, while domestic capacity matures — and it does so with US chips, since Cerebras designs the CS-3 systems. This is not a China-style decoupling play; it is American hardware reaching an emerging market through a Gulf intermediary. Cerebras's own business underscores how thin the supplier base still is: G42 and MBZUAI accounted for roughly 86% of its 2025 revenue ahead of its May 2026 Nasdaq IPO (Rest of World).
For US policymakers, that fact cuts against the instinct to clamp down. Export controls aimed at keeping frontier chips out of rivals' hands have legitimate security goals. But a blanket suspicion of Gulf-routed compute would push buyers like India toward non-US silicon and hand the sovereign-AI market to suppliers Washington cannot influence at all. A licensing regime that verifies end-use and operator governance — already the direction of recent US-UAE chip frameworks — keeps American hardware in the loop while addressing diversion risk.
The pattern to watch
Condor Galaxy India is a template, not a one-off. G42's Condor Galaxy network already runs in the United States; India is its first major emerging-market extension (Rest of World). Expect Brazil, Indonesia, and African states to be offered similar packages: turnkey sovereign compute, financed and operated by a Gulf partner, running US chips under local data rules. That is, on balance, a healthy diversification of a dangerously concentrated market — provided the governance terms are made public and contestable. The policy failure would be to wave the deals through on national-pride rhetoric, or to ban them on reflex. Insist on transparent contracts and auditable sovereign control, and let the competition run.