When asked if G42 truly holds the answer to the universe, Manu Jain, the chief of G42 India, said, “I hope we can be a small catalyst in that.” Jain was speaking at Cypher 2024, India’s biggest AI conference organised by AIM Media House. G42, the Abu Dhabi-based AI company, was inspired by The Hitchhiker’s Guide to the Galaxy.
At Cypher, G42 unveiled its plans to build the largest data centre in India, besides launching a local ‘Hinglish’ LLM, called NANDA (named after the second-highest peak in India). This move mirrors their JAIS Arabic LLM, inspired by the UAE’s highest peak, as G42 positions itself to ignite a generative AI revolution across the subcontinent.
With this, the company is looking to spearhead AI in India, with other players such as Krutrim AI and Sarvam AI already having made their presence felt in the country by working towards building an ‘AI Full Stack’.
What G42’s Nanda Means for India
“After Kanchenjunga, the second-highest peak in India is Nanda Devi. Kanchenjunga is longer and more difficult to pronounce, so we decided to go with Nanda. It’s a short and sweet name, which is why we chose it,” said Jain, sharing the story behind the name NANDA.
NANDA is a 13-billion parameter model trained on approximately 2.13 trillion tokens of language datasets, including Hindi. The LLM was an outcome of a collaboration between Inception (a G42 company), Mohamed bin Zayed University of Artificial Intelligence, and Cerebras Systems.
Jain, who previously spearheaded the smartphone adoption in the country during his tenure at Xiaomi, is now following a similar strategy with their largest Hinglish AI model.
“With NANDA, we are creating a model that speaks to India–its languages, its people. This is not just about AI for a few; it’s AI for a billion,” said Jain, adding that the adoption of AI in real-life cases is sometimes higher than smartphones. “In the coming years, we cannot imagine our lives without AI, and I believe its adoption curve will be even higher than the smartphones,” he added.
“I think the one big difference between smartphones and AI models is that the former is very incremental, but what is happening with AI is that the next feature is very exponential—in terms of capabilities and adoption. We have seen this with our own models—the capabilities of the newer ones are much better than the previous ones,” said Jain.
He further added that, unfortunately, for most Indian languages, the digitised data is significantly limited. Hindi is the fourth most-spoken language, with approximately 600 million speakers. However, the amount of digital data available is extremely low.
“As a country, we need to make an initiative, and a lot of people are doing it, like Nandan Nilekani’s organisation, many of the IITs, and AI4Bharat. They’re taking the initiative. We are also trying to digitise, but I think, as a country, we need to do a lot more. Then we will be able to build really good quality models,” said Jain.
Jain said NANDA is going to be pretty useful for day-to-day tasks and the model has all the information required. “You can ask about, like, lessons on swimming. You can ask about translating from Hindi to English. You can actually ask it to make a presentation or help you prepare a presentation. It’s your digital assistant,” he said.
Building Data Centres in India
G42 is planning to build large data centers in India focused on AI workloads and develop one of the country’s largest supercomputers. This initiative is part of a pact signed between the UAE and Indian governments aimed at co-developing sovereign AI.
Jain explained that G42 is a holding company with several operating companies under it, structured around three fundamental building blocks of AI, data centres, supercomputers (or compute clusters), and AI models.
Under G42, there is Khazna, one of the largest data centre providers in the region. The second company, Core 42, is responsible for building supercomputers and AI cloud infrastructure. Lastly, Inception focuses on developing AI models. Together, these companies form the core components of G42’s operations and expertise in the AI and tech sectors.
Moreover, the company is working closely with Cerebras. Together, the two companies recently built the supercomputer Condor Galaxy 1 (CG-1), which is already operational in Santa Clara, California. It offers 4 exaFLOPs of AI computing power and features 54 million cores.
The company also partnered with NVIDIA to work on climate tech and develop solutions to boost weather forecasting accurately.
Microsoft recently invested $1.5 billion in the company, securing a minority stake. “This was part of a larger partnership between Microsoft and G42 to advance AI globally, starting with the UAE. In fact, the vice chairman of Microsoft sits on our board. We have a great relationship with OpenAI, AWS, and several other technology companies worldwide,” said Jain.
Not only that, G42 encompasses several companies, including Space 42 (formerly Bayanat), which specialises in geospatial intelligence, and M42, a healthcare subsidiary focused on using AI to improve patient care and operational efficiency in the medical field. Earlier this year, M42 launched two new open-access versions of the Med42 LLM.
