Abu Dhabi's G42 will run 64 Cerebras systems under Indian data law, building a parallel sovereign data layer alongside the $45 billion India has already committed to Microsoft, Google, and AWS.
On May 15, 2026, Abu Dhabi's G42 signed an agreement to install 64 Cerebras systems in India, the first country to sign onto what the company calls its Intelligence Grid, a network of AI facilities the UAE firm builds, owns, and operates for governments under their own data rules (Rest of World, June 1, 2026).
G42 is not a startup. Backed by Abu Dhabi's Mubadala sovereign wealth fund, it has become the UAE's national AI champion, and its subsidiary Core42 will handle installation, operations, and maintenance of the India systems. The hardware partner is Cerebras, a U.S. chipmaker whose wafer-scale silicon is roughly dinner-plate sized and optimized for running AI applications rather than training them, a fit for India's stated focus on healthcare, agriculture, and public services. On the Indian side, the Centre for Development of Advanced Computing (C-DAC), an autonomous scientific body under the Ministry of Electronics and Information Technology, will work alongside Core42. All data generated will fall under India's Digital Personal Data Protection Act, 2023.
The deal reads at first glance as India breaking with the U.S. cloud giants. The numbers say otherwise. Microsoft has committed $17.5 billion over four years to AI and cloud in India. Google has pledged a $15 billion AI hub. AWS has earmarked $12.7 billion for cloud infrastructure. That is more than $45 billion in U.S. hyperscaler commitments, all built on Nvidia processors and the companies' own cloud platforms. India is not walking away. It is adding a second layer.
G42's own framing matters here. The Intelligence Grid is pitched as a sovereign alternative, a way for governments to rent AI capacity from a non-U.S. operator, under their own jurisdiction, without surrendering data to American or Chinese clouds. Cameron Kerry, a former U.S. acting Commerce secretary now at the Brookings Institution, calls this "pragmatic AI sovereignty." In a February 2026 report, Kerry and his co-authors argued that no country can control the full AI stack of chips, models, cloud, and applications, and that the realistic path is to assemble capacity from multiple partners. India's deal with G42 is the first national-scale test of that thesis.
The data-governance piece is the part India can actually claim. The Cerebras chips are still American. The software stack running on them is still largely American. What changes is who operates the facility and which law governs the data. G42 declined to disclose the financial terms of the deal or whether G42 or its Indian counterpart will own the hardware after installation, leaving a material gap in any sovereignty claim (Rest of World).
The exposure runs both ways. Cerebras IPO'd on Nasdaq on May 14, 2026, raising $5.55 billion in the largest U.S. tech offering since Uber in 2019. The same SEC filing showed that G42 and the UAE's MBZUAI research institute accounted for 86 percent of Cerebras' 2025 revenue. G42 and Cerebras have already co-built three Condor Galaxy supercomputers in California, Texas, and Minnesota, a track record G42 cites for multi-jurisdiction operations. The UAE has, in turn, become a hyperscaler market in its own right. Microsoft has committed $15.2 billion to UAE data centers through 2029, working through G42's subsidiary Khazna. The same UAE firm that offers an alternative to the U.S. hyperscalers abroad is a partner to them at home.
The open question is whether the software follows the hardware. Chris Miller, a Tufts Fletcher School historian of semiconductor competition, notes that matching the integrated packages of Amazon, Microsoft, and Google, the developer tools, enterprise support, and model libraries, is the harder lift. Cerebras is fast at running trained AI models, but Indian researchers and businesses will still need the surrounding software ecosystem. India's own national AI mission is a $1.25 billion program running 34,000 Nvidia processors today, with a target of 100,000 by the end of 2026. The G42 deal does not replace that. It adds another on-ramp.
What India is buying is not chip independence. It is a second operating layer, run by a non-U.S. firm, under Indian law, available alongside the U.S. cloud giants it has no plans to leave. Whether other middle-power governments copy the template, and whether the software stack on top of Cerebras hardware matures fast enough to make the second layer useful at scale, are the next two questions worth watching.