It takes two years, a huge debt, and a lot of faith to go from bootstrapped to unicorn in India’s AI era.

Neysa did it in February when it raised $1.2 billion to build sovereign computeSovereign computeAn initiative under the IndiaAI Mission aimed at reducing reliance on foreign cloud providers for AI processing, and to host, govern, and evolve AI systems under Indian control for India’s AI ecosystem, one of the country’s biggest bets in the space. The AI-cloud startup wants to source around 20,000 graphics processing units (GPUs) from chip giant Nvidia, which are costly and often in short supply, and deploy them in domestic data centres where local companies can easily access them.

Easier said than done. Half of Neysa’s recent fundraise is equity from global investors like Blackstone. The other half is debt, making it a risky bet in a market that hasn’t seen adequate demand yet.

Even with compute priced at roughly Rs 65 an hour—cheap by global standards—more than three-fourths of the IndiaAI Mission’s GPUs are sitting unusedThe Economic TimesUnderused GPUs raise questions about IndiaAI capacity buildout.

This is not a Neysa-specific problem. E2E Networks, India’s first listed cloud provider, is already living through that cycle. The company, backed by infrastructure giant Larsen & Toubro, has also seen its GPU utilisation fall sharply in recent quarters.

The problem runs deeper than the under-utilisation of GPUs. Sarvam AI, selected by the IndiaAI Mission to build a sovereign large language model (LLM), counts Nvidia among its backers. In other words, India is funding both compute and models, but the underlying training stack remains foreign.

Meanwhile, the startups this infrastructure was built for are choosing Microsoft Azure, Amazon Web Services (AWS), or Google Cloud Platform (GCP), thanks to VC funding that includes bundled cloud credits. 

And now, these hyperscalers have a new pitch for India: sovereignty.

While Microsoft is rolling outMicrosoftMicrosoft Sovereign Cloud adds governance, productivity and support for large AI models securely running even when completely disconnected  separate sovereign cloud-computing environments, Google and Amazon are offering local data processing and wider compliance zones, respectively.

© The Ken