Can India develop a self-sustainable AI ecosystem?

As artificial intelligence (AI) enters banking, healthcare, coding, education and enterprise workflows, India risks building its next digital revolution on foreign models, cloud infrastructure, and compute.

While India already has an oil and gold import bill burden, experts caution that it might get compounded by AI. This would be driven not by cargo shipments, but by subscriptions, cloud services, inference charges, graphics processing units (GPUs) access and AI infrastructure controlled by global technology companies.

All our AI assistance is imported. Gemini, ChatGPT, Claude, Perplexity, Copilot and others are foreign AI models powering our synthetic intelligence. Mostly free for end users at present, with paid versions for users and enterprises offering a better experience, these could eventually burn a hole in India’s pocket, potentially creating a new category of import, i.e., intelligence. The risks are not about something being done in labs to train models. Inference is the use of AI models for practical applications like fraud detection, medical diagnosis, language translation, research, and driverless cars, among others. In FY26, India spent over $206 billion on crude oil and gold imports. According to the Ministry of Commerce and Industry data, crude oil imports were $134.7 billion, and gold imports stood at $71.98 billion. AI subscriptions are nowhere near this yet. AI, however, is increasingly becoming the cognitive layer sitting atop economies, governments, enterprises and digital public infrastructure.

“The dominant models powering Indian enterprises and developer workflows, including OpenAI’s GPT series, Google Gemini, Anthropic Claude, etc., are built, governed and monetised entirely outside India,” says Amit Chand, founder of BYT Capital (which has backed AI ventures). Even access to newer models like Anthropic-owned Mythos, which can completely change the cybersecurity landscape, is tightly controlled.

According to Medha Kannapally, Associate at Endiya Partners, India is heavily dependent on foreign AI models and currently imports intelligence in much the same way it imports energy, electronics or advanced semiconductor technologies. Endiya Partners, a venture investor, has backed AI companies, including SigTuple, ThirdAI, Perceptyne and others.

The dependence extends beyond models. It includes GPUs, cloud infrastructure (data centres), model-training platforms, research ecosystems and inference services — the engines that make AI useful in real-world applications.

India is primarily a consumer rather than a producer of foundational AI technologies. Global market intelligence firm IDC projects India’s AI spending to touch $6 billion by 2027. Much of that spending will flow toward subscriptions, APIs, cloud services and inference charges controlled by foreign technology companies.

According to Gaurav Vasu, CEO, UnearthInsight, a Bengaluru-based consultancy, India, which accounts for 10-12% of the user base of global AI models, spent $2.5 billion in FY25, and it is expected to be more than double by FY28. “As users lean on AI for their work, there will be more users and enterprises paying out for using AI; hence the payouts will increase significantly by FY28 and beyond.”

Not everyone sees India’s dependence on foreign AI models as necessarily negative. One view is that AI is increasingly becoming a “must-have” tool for work, making adoption inevitable as more users and businesses begin to rely on it. In this perspective, the market is likely to evolve through a mix of free and paid offerings, with users paying according to their budgets and needs.

Dollar Flight

The debate around AI dependence often focuses on large language models. Yet, the real economic battleground may be inference. Inference occurs every time an AI model processes new information and generates an answer, recommendation, prediction or decision. Every ChatGPT query, fraud alert, medical diagnosis, customer support interaction or coding suggestion involves inference. Inference is where the bulk of future AI consumption will occur.

For example, in healthcare, inference helps analyse medical scans and detect diseases. In banking, it identifies suspicious transactions and flags fraud within milliseconds. In retail, it powers recommendation engines. In manufacturing, it predicts factory equipment failures. Autonomous vehicles, voice assistants and AI agents all rely on inference running continuously. Much of this inference today runs on infrastructure owned by foreign providers and is billed in dollars.

Dependence on foreign intelligence layers raises questions of sovereignty as well. Kannapally says, “The biggest risk is not technological dependence, but economic dependence. If every major enterprise workflow, government service and digital product is powered by imported intelligence, India could end up exporting capital while importing AI capabilities at scale.”

Lack of control or local options for AI for governments and users could compromise user data as data flows to a foreign company.

Today, AI providers are competing aggressively, offering free or heavily subsidised services to build user habits. But that may not last forever. Ganesh Gopalan, co-founder and CEO of Gnani.ai, argues that free AI should raise questions. “The first thing we should ask is: why are these companies offering their services for free? They want to make users stick to the platform. Once users become dependent, they will pay up if alternatives are not available,” says Gopalan.

The more users rely on AI inference systems for daily work and decision-making, the more difficult it becomes to resist using them. That creates both economic and strategic lock-in.

According to the Economic Survey 2025-26, India generates roughly 20% of the world’s data but stores only around 3% domestically. Every interaction with foreign AI systems potentially contributes to improving those systems.

“If they are charging, Indian companies will eventually compete. The larger concern is when AI systems are trained and improved using Indian data. And this is different from the internet. The internet democratised information. AI is democratising intelligence. That makes it much more fundamental,” says Gopalan. Though some of the start-ups are using open-source AI models, the user base is small.

Local Efforts

Under the India AI Mission, backed by an allocation exceeding Rs 10,000 crore, several indigenous efforts are underway. Among the most prominent are Sarvam AI, BharatGen and Gnani.ai. Sarvam recently unveiled its Vikram 30-billion-parameter model and a larger 105-billion-parameter reasoning model trained on sovereign Indian compute infrastructure. BharatGen has launched Param2, while Gnani.ai has developed multilingual voice models tailored for Indian languages. Yet, the gap with frontier global models remains significant.

“When compared with OpenAI, Anthropic or Google Gemini, there is a substantial gap in reasoning capabilities, multimodal performance, research depth and training infrastructure,” says Kannapally.

Experts argue that India’s goal should not be to replicate GPT-5 or compete head-on with Silicon Valley. Instead, India’s opportunity lies elsewhere, in niche small language models and localisation.

Global models often struggle with Indian languages, dialects and cultural nuances. That’s where Indian companies can play a role. Gopalan believes voice AI could become India’s strongest competitive advantage. “Bharat (the hinterlands) will never use AI through chat interfaces alone. They will use voice,” says Gopalan.

India must also invest in research talent, datasets, semiconductor capabilities and open-source AI ecosystems. “The answer is not necessarily replacing foreign models. It is building domestic alternatives and increasing competition,” Kannapally says.

Local innovators are already working on small language models to address unique local requirements. India has the tech prowess; now the AI models must show reliability at scale, and this could arrest some of the forex flight.