G42 and Cerebras bring 8 exaflops to India

India’s latest bid for “sovereign AI” looks less like a clean national stack and more like a three-way joint venture: Gulf capital, American chips, and Indian compliance paperwork.

TechCrunch reports that Abu Dhabi-based G42 has partnered with US AI-chip firm Cerebras to deploy “8 exaflops” of compute in India, announced on the sidelines of the India AI Impact Summit in New Delhi. The system is to be hosted in-country and operated under India’s data residency and security rules, with access aimed at educational institutions, government entities, and small and medium enterprises.

The marketing language is familiar: G42’s India CEO Manu Jain calls sovereign AI infrastructure “essential for national competitiveness,” while Cerebras frames the project as a leap in India’s ability to train and run large-scale models. The stack, however, is inherently multinational. The compute is delivered by a US hardware vendor; the financing and corporate control sit in the UAE; and the “sovereignty” is largely defined by Indian regulators and procurement channels.

At the same summit, TechCrunch notes a broader rush to build data centers and AI capacity. Adani pledged $100 billion to build up to 5 gigawatts of data-center capacity by 2035, while Reliance said it would invest $110 billion over seven years for gigawatt-scale facilities. OpenAI, via its Stargate project with Tata, is seeking 100 megawatts of compute initially, scaling to 1 gigawatt. US hyperscalers—Amazon, Google, Microsoft—have already committed roughly $70 billion to expand cloud and AI infrastructure in India.

Capital is also piling into the startup layer. TechCrunch reports that Peak XV (the former Sequoia India) has raised $1.3 billion for India and broader Asia-focused funds, with most earmarked for India over the next two to three years. Managing director Shailendra Singh emphasized returns over asset-gathering, while noting the firm has already made more than 80 AI investments.

France 24, covering the summit’s “AI governance” discussions, captures the other half of the Indian state’s approach: committees, frameworks, and the perennial question of “who owns AI.” The governance conversation is not irrelevant—rules shape markets—but it also offers political cover when the hard dependencies remain external.

“sovereign AI” is becoming a branding exercise for states that can’t (or won’t) let markets simply buy compute and compete. India wants local data control, but the hardware, capital, and model ecosystems are globally entangled. The likely outcome is not sovereignty, but a new class of gatekeepers: regulators, subsidized infrastructure champions, and politically connected allocators of scarce compute.

When three power centers—Washington’s chip supply chain, Abu Dhabi’s balance sheet, and New Delhi’s compliance regime—each hold a veto, the only truly sovereign actor is the one who can walk away. That actor is rarely the state.