Before AI Summit: Know who actually controls India’s Artificial Intelligence, not India

India’s artificial intelligence ambitions rest on a foundation the government doesn’t own, built with hardware it can’t manufacture, controlled by private companies operating foreign technology. As the AI Impact Summit convenes at Bharat Mandapam this week, this uncomfortable reality goes beneath the celebratory announcements: Who actually controls the infrastructure powering India’s AI future? The answer is corporate, concentrated, and dependent.

The sovereignty illusion

The Ministry of Electronics and IT announced in May 2025 that India's “national compute capacity has crossed 34,000 GPUs”- 34,333 units to be precise, more than triple the IndiaAI Mission’s original 10,000- GPU target.

Impressive on paper. Problematic in practice: Every single GPU in that pool is foreign silicon. NVIDIA H100 and L40S accelerators. AMD chips. Google TPUs. All imported. All subject to foreign export controls and geopolitical calculations that could shift overnight. And over half- more than 50%- of this “national” capacity is controlled by a single private company: Yotta Data Services. Yotta has committed 9,216 advanced GPUs to the IndiaAI Mission through its Shakti Cloud platform. The company explicitly states it will provide “more than 50%” of the Mission’s GPU capacity, delivered in phases under commercial contracts the public cannot read.

Also Read: Higher education at risk? Ex-Google AI leader warns law and medical degrees may lose value as AI moves fast

Empanelled, not owned

The government calls these “empanelled GPUs”- a term that obscures ownership. These aren’t state assets. They’re private sector capacity approved for IndiaAI workloads under commercial arrangements. The IndiaAI compute portal functions as a marketplace. Researchers and startups access these GPUs through the platform, but the hardware belongs to cloud providers who bill for usage.

This model allowed rapid scaling without the government building data centers itself. But it also means India’s AI capability depends entirely on commercial relationships with firms whose primary obligation is to shareholders, not national strategy.

If Yotta faces technical failure, financial trouble, or supply disruptions, more than half of India’s AI compute goes dark. If U.S. export policy shifts- as it has dramatically against China- India’s GPU pipeline could be cut at the source.

The silicon gap

India has launched a RS76,000 crore semiconductor manufacturing program. The 2026-27 budget includes RS1,000 crore for India Semiconductor Mission 2.0, targeting equipment production and supply-chain development. But budget documents reveal execution lag. The semiconductor program spent RS638.07 crore in 2024-25 against a budget estimate of RS6,903 crore- a 9.2% utilization rate.

The IndiaAI Mission itself shows even bigger gaps. Budget estimate for 2024-25: RS551.75 crore. Actual expenditure: RS19.24 crore- just 3.5%.

No operational semiconductor fab capable of producing AI-class chips exists in India. The government’s own documents show these programs remain in the incentive and project-approval phase, not production. Even optimistically, domestic AI-grade chip manufacturing is years, possibly decades, away from cutting-edge nodes.

Also Read: ‘AI will replace most white-collar jobs soon,’ says Microsoft AI head Mustafa Suleyman

Big Tech at the table

The summit’s institutional architecture deepens corporate influence. Participant lists include Google, Microsoft, Meta, IBM, OpenAI, and DeepMind alongside Indian majors like Infosys, TCS, Reliance, and Bharti. These companies sit on multi-stakeholder working groups shaping AI policy, standards, and deployment. They provide the cloud platforms, foundation models, and developer tools most Indian AI practitioners use.

The Ministry of Agriculture’s Digital Agriculture Mission includes MoUs with Cisco, Jio Platforms, ITC, and NCDEX for AI deployment- valuable partnerships that also ensure agricultural AI flows through corporate channels.

Start-ups as renters, not owners

Four Indian start-ups- Sarvam AI, Soket AI, Gnani AI, Gan AI- were selected to build indigenous foundation models under IndiaAI. But they access compute through the same marketplace dominated by large providers, train on foreign GPUs, and build on frameworks (PyTorch, TensorFlow) created by Meta and Google. The architecture gives start-ups visibility but keeps them structurally dependent on infrastructure they do not control.

The control question

So who controls India’s AI?

Hardware layer:Foreign manufacturers (NVIDIA, AMD, Google) subject to their home governments’ export rules.

Infrastructure layer:Private cloud providers, with Yotta holding a majority position.

Platform layer:Global Big Tech and Indian conglomerates providing cloud services and foundation models.

Application layer:Indian start-ups and researchers building on infrastructure they don’t own, using tools they didn’t create, running on chips they can’t manufacture.

Policy layer:Government sets broad direction but lacks technical capacity and budget execution speed to build alternatives.

The government frequently cites BHASHINI- India’s national language AI platform- as proof of sovereignty. The framing is accurate but incomplete: BHASHINI is Indian-designed software serving Indian goals, running on Indian-operated infrastructure. But that infrastructure executes on NVIDIA H100 GPUs manufactured in Taiwan and subject to U.S. licensing. Data sovereignty and access sovereignty exist. Hardware sovereignty does not.

Unanswered questions

As ministers will be cutting ribbons at Bharat Mandapam this week, critical governance questions remain: How will India prevent Yotta- or any dominant provider- from using market position to set pricing or restrict access? What happens if GPU export restrictions hit India, as they have China? Will semiconductor programs deliver operational fabs, and when? Should the government mandate ownership diversification to reduce concentration risk?

The summit will showcase achievements- and they are real. India has mobilized resources and created frameworks that didn’t exist five years ago.

But the fundamental architecture is one of dependency: sovereign in aspiration, corporate in execution, foreign in silicon. India’s AI runs on NVIDIA’s chips, in Yotta’s data centers, shaped by Google’s frameworks, governed by contracts the public cannot read. The sovereignty is rhetorical. The control is corporate. The silicon is foreign. That’s the infrastructure reality behind “Make AI in India.”