Headline & intro
India is no longer asking how to catch up in AI – it’s buying a seat at the table where compute is power. Abu Dhabi’s G42 teaming up with U.S. chipmaker Cerebras to deploy 8 exaflops of AI compute in India is not just another data center announcement; it’s a geopolitical statement. In one move, India strengthens its “sovereign AI” ambitions, the Gulf extends its tech influence into South Asia, and an Nvidia rival lands a flagship win. This piece looks at who really gains, what this means for the global AI arms race, and why Europe should be paying very close attention.
The news in brief
According to TechCrunch’s reporting from the India AI Impact Summit 2026 in New Delhi, Abu Dhabi–based technology group G42 has partnered with U.S. chipmaker Cerebras to deploy a new AI supercomputing system in India delivering up to 8 exaflops of compute.
The system will be hosted on Indian soil and is designed to comply with local data residency, security and regulatory requirements. Its capacity is meant to be shared across universities, government bodies and small and medium‑sized enterprises.
The initiative involves Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in Abu Dhabi and India’s Centre for Development of Advanced Computing (C‑DAC). TechCrunch notes that MBZUAI and G42 previously released Nanda 87B, a Hindi‑English large language model built on top of Meta’s Llama 3.1 70B.
The announcement comes amid a broader wave of AI infrastructure pledges in India: Adani and Reliance are committing over $200 billion combined to data‑center capacity, OpenAI is working with Tata Group on large‑scale compute, and U.S. cloud giants have reportedly pledged around $70 billion for AI and cloud infrastructure in the country.
Why this matters
On paper, “8 exaflops” sounds like a big number in search of a use case. In reality, this deployment is about three things: sovereignty, diversification and leverage.
For India, this is a shortcut to meaningful AI independence. Training state‑of‑the‑art models – especially multilingual ones tailored to India’s dozens of major languages and countless dialects – is compute‑hungry. Renting capacity from U.S. hyperscalers keeps India in a dependency loop and exposes it to foreign policy shocks. Hosting a high‑end system domestically, under Indian rules, lets the government and local ecosystem retain control over sensitive datasets in health, finance, public services and national security.
For G42, it’s the export of a playbook already tested in the UAE: combine capital, access to advanced chips and local partnerships to become a go‑to provider of sovereign AI stacks for countries that don’t want to rely solely on U.S. tech. India gives G42 scale, legitimacy and diversification beyond the Gulf.
For Cerebras, this is a rare, high‑profile win against Nvidia’s near‑monopoly in AI compute. Its wafer‑scale systems are designed for exactly this kind of large, centralized AI cluster. If the project delivers, Cerebras can point to India as proof that alternative architectures can power national‑level AI.
The losers? Smaller cloud providers and local data‑center players without access to comparable hardware will find it even harder to compete. And every such deal deepens the emerging AI world of blocs – U.S.-aligned, China‑aligned and now a more assertive Gulf–India axis.
The bigger picture
This announcement slots neatly into several converging trends.
1. AI compute as industrial policy. Over the past two years, governments have realized that GPUs are the new oil. The U.S. has imposed export controls on advanced chips to China; the EU talks about “technological sovereignty”; Gulf states pour petrodollars into AI funds. India’s response is clear: attract as much compute as possible, as fast as possible, by offering scale, policy support and enormous domestic demand.
The G42–Cerebras project sits alongside massive local commitments by Adani, Reliance and Tata/OpenAI. Together, they signal that India does not intend to be just a market for foreign AI services; it wants to be a production hub for models, infrastructure and, increasingly, standards.
2. The rise of alternative chip players. Nvidia still dominates, but its supply constraints and pricing have created an opening. Cerebras, with its wafer‑scale engines, positions itself as an AI supercomputer in a box, trading ecosystem maturity for density and speed. Deals at this scale matter: they give non‑Nvidia vendors the reference customers they need to be taken seriously by governments.
3. AI nationalism 2.0. The first wave of AI nationalism was about strategies and papers. The second is about hardware on the ground. The UAE built Falcon and invested heavily through G42. Saudi Arabia is funding labs and chip purchases. Now, with this India deployment, we’re seeing cross‑regional alliances where capital from the Gulf, chips and IP from the U.S., and data and market from India fuse into a new power center.
For the global AI industry, this suggests a future less dominated by a simple U.S.–China duopoly and more by regional constellations built around shared infrastructure and aligned interests.
The European angle
From a European perspective, the most uncomfortable part of this story is what it doesn’t contain: Europe.
While the EU debates the AI Act and pushes long‑term programs like EuroHPC and IPCEI initiatives for chips, India is signing concrete deals that bring exaflop‑scale systems online with foreign partners. The contrast between Europe’s regulatory energy and its relatively modest AI compute footprint is growing sharper.
For European companies, India’s strategy cuts both ways. On one hand, a powerful AI infrastructure base in a democratic, data‑protection‑conscious country could become an attractive location for training multilingual models for emerging markets – especially for firms in fields like fintech, education and health that already treat India as a key market.
On the other hand, data sovereignty and GDPR create friction. Routing EU personal data into non‑EU data centers is heavily constrained, regardless of how attractive the compute pricing looks. Even if Indian facilities follow strong security and privacy practices, they are not part of the EU’s legal perimeter.
This means European players face a strategic choice:
- Double down on building their own sovereign AI stacks – likely more slowly and at higher cost.
- Or accept that, for some workloads and non‑personal datasets, India and Gulf‑backed infrastructure will be part of their compute mix.
Either way, Europe needs to think of India less as an outsourcing destination and more as an increasingly equal partner – and competitor – in AI infrastructure.
Looking ahead
Several questions will determine whether this 8‑exaflop system is a historic turning point or just another underutilized supercomputer.
1. Who gets priority access? If government agencies and a handful of big corporations monopolize the capacity, the official narrative of supporting universities and SMEs will ring hollow. A transparent allocation model – including discounted or grant‑based access for researchers and startups – would be a strong signal that this is genuine national infrastructure, not just elite hardware.
2. What gets built on top? The earlier Nanda 87B model shows G42 and MBZUAI are comfortable building on open‑weight foundations like Meta’s Llama. An Indian system at this scale could power:
- Foundation models covering dozens of Indian languages beyond Hindi and English
- Sector‑specific models for agriculture, law, healthcare and public services
- Safety and evaluation infrastructure tailored to local norms
If, instead, the system is mostly used to fine‑tune imported English‑centric models, much of its sovereign potential will be wasted.
3. Can Europe plug in smartly? Expect to see more European universities and startups looking for partnerships with Indian and Gulf institutions to access this kind of compute indirectly – for example through joint research programs where the data stays local but expertise and models are shared.
Over the next 18–36 months, watch for: additional G42‑style sovereign AI deals in Southeast Asia and Africa; whether Nvidia responds with more aggressive national‑scale offerings; and how energy and water constraints shape where the next generation of AI megacenters are built.
The bottom line
G42 and Cerebras bringing 8 exaflops of AI compute to India is less about teraflops and more about who gets to write the rules of the AI era. India gains leverage and a faster route to sovereign AI; the Gulf cements its role as an AI kingmaker; and an Nvidia rival lands a showcase deployment. Europe can either treat this as a wake‑up call to accelerate its own infrastructure – or watch as the map of global AI power redraws itself without much blue on it. Which side of that line do you want to be on?
