India needs a National Inference Mission to capture its share of the emerging $1 trillion AI economy

‍As global AI revenue grows five times in two years to reach $435 billion in Q1 2026, India must build a National Inference Mission to keep more of that economy onshore. 

India's external account survives because services exports cover for goods imports. In FY26, that surplus reached $213.9 billion, the highest in India's history, up 13.3% year-on-year. It paid for the $130 billion net crude oil bill, the $72 billion gold import bill, and most of India's import dependence.

The Prime Minister's appeal in May, asking citizens to work from home, avoid foreign travel and reduce buying gold, is the surface signal that this balance is under live pressure from rising commodity prices. The Reserve Bank's foreign exchange reserves have fallen $37.8 billion in two months, from $728.5 billion on 27 February to $690.7 billion on 1 May. Standard projections of the current account deficit have widened in recent months as commodity prices have risen.

Every conversation about India's external account today is about oil, gold and remittances. A new variable is forming that will, over the next decade, also become material: the rising USD bill that India is paying for AI inference.
‍

The Services Surplus Is the Reason India's External Account Holds

India's external account stability is built on top of services exports, and this matters more today than ever before. FY26 services exports crossed $400 billion for the first time, reaching $418.3 billion, against services imports of $204.4 billion. The Finance Ministry's April Monthly Economic Review confirmed that this net surplus accounts for 64.2% of India's merchandise trade deficit. The share of services in total exports has risen to 48.6%, up from 47% in FY25. India is consistently approaching service-goods parity in the composition of its export base, a fundamental reordering that will have significant second and third order benefits to the economy.
‍

We argued in our earlier piece, The Great Rebalancing, that this services surplus is the engine carrying India towards an overall net export surplus within two to three years. That would be a milestone the country has never reached in its modern economic history. It would change the current account, the sovereign rating trajectory, and ultimately the cost of capital itself.
‍

The surplus is not abstract macroeconomics. It is Bengaluru engineers, Pune analysts, and Hyderabad operations staff selling time to foreign clients at a rupee cost base and a dollar billing rate. Alongside remittances, this has been the structural foundation of India's external account for two decades. No other line of the trade balance has compounded this consistently. 
‍

The AI Token Economy Could Become a New Dollar Cost on India's Economic Engine

A new dollar-denominated input cost is forming inside the same services engine that balances India's external account. The velocity of that cost is exponential.
‍

Altimeter's analysis of the global AI economy shows revenue across the stack, from semiconductors to infrastructure to models + applications, has grown roughly fivefold in two years, from $90 billion in Q1 2024 to $435 billion in Q1 2026. The semiconductor layer moved from $75 billion to $300 billion, the cloud and data centre infrastructure layer from $10 billion to $75 billion, and the apps and models layer from $5 billion to $60 billion. The apps and models segment is the smallest of the three in absolute terms, but it has grown twelve times in two years, the steepest growth rate of any layer. This is the layer that charges tokens, that scales costs with usage, and that lands directly on the software bills of customers, Indian enterprises and SMEs included.
‍

India's position in the global demand picture is now disproportionate. The Stanford HAI AI Index Report finds over 80% of Indian employees surveyed use AI at work on a regular basis. India tops LinkedIn's global relative AI skill penetration index at 3.0, three times the global average. 85 to 90% of Indian respondents report their organisation actively supports AI strategy and literacy, the highest in any country surveyed. India is a leader in AI skills penetration and diffusion.
‍

Commercial scale follows the workforce signal. India is OpenAI's second-largest market globally, with 100 million weekly active ChatGPT users disclosed by Sam Altman ahead of the February 2026 AI Impact Summit. India is Anthropic's second-largest market for Claude, with India run-rate doubling between October 2025 and February 2026. ElevenLabs has identified India as its second-largest enterprise market globally by revenue, and industry events have reported it targeting $100 million from India alone.

The trend is straightforward. Global AI capital expenditure is estimated to exceed $700 to $900 billion in 2026, with cumulative multi-year infrastructure spending projections reaching as high as $7-$8.9 trillion by the end of the decade. The expected revenue yield needs to surpass $1 trillion annually for the investments to pay off. As the global AI economy builds towards the multi-trillion-dollar projections being published for 2030, India could pay an increasing share of its services growth back in dollars to foreign model providers. 
‍

This is not a crisis today. It is an emerging dependence, and India has a short window in which to convert that dependence into domestic value capture.
‍

India Has Started Building the Infrastructure, the Inference Layer Is the Urgent Next Step

India has executed reasonably quickly at the chip and infrastructure layers. 
‍

Semiconductor expansion is underway. Thirteen semiconductor projects have been approved across seven states under the India Semiconductor Mission and SPECS. Budget 2026-27 allocated Rs 8,000 crore to the mission, its largest single-year outlay. Tata-PSMC's Dholera fab targets first silicon by December 2026. Micron's Sanand assembly and test facility was inaugurated by the Prime Minister on 28 February 2026. Kaynes Semicon's Sanand facility was inaugurated on 31 March 2026. CG Power, the HCL-Foxconn joint venture at Jewar, and Tata's TSAT facility in Assam are all under construction. Close to ten semiconductor facilities will be operational or near-operational within the next year. While this doesn’t yet target the advanced compute and GPUs driving AI infrastructure, this will be the foundation for more onshoring over the next decade.
‍

Hyperscale cloud and AI infrastructure has followed. Microsoft committed $17.5 billion to India in December 2025, on top of the $3 billion pledged in January 2025. AWS has committed $12.7 billion by 2030, on top of $3.7 billion already invested. Google announced a $15 billion AI hub in Visakhapatnam with Adani, which has separately committed $100 billion to AI data centre buildout by 2035. The Summit also produced the right kind of business-to-business announcements. TCS HyperVault will host OpenAI's Stargate workloads in India, starting at 100 MW and scaling to 1 GW. Infosys will deploy Claude through its Topaz AI platform. India's installed data centre capacity is set to grow from 1.3 GW to 9 GW by 2030.
‍

The model and inference layer is the gap. The IndiaAI Mission has done the basic groundwork at the compute layer. More than 38,000 GPUs have been deployed to a common compute facility, with another 20,000 committed at the February 2026 AI Impact Summit. Subsidised access is available at Rs 65 per hour. An LLM startup has received the largest single allotment of 4,096 H100 GPUs to build a sovereign foundation model.
‍

The inference layer that runs on top of this must now be built in parallel and urgently, not after.
‍

A Multi-Pronged National Inference Mission Is the Answer

India needs a coordinated industrial policy on inference, executed with the urgency historically reserved for energy security. The app and model layer is the urgent next priority, and the IndiaAI Mission must expand into a National Inference Mission. Three priorities deserve immediate attention.
‍

First, designate national champions and direct incentives at the three layers, not just at chip procurement. Targeted incentives are needed for hyperscale data centre buildout, GPU import duty rationalisation, accelerated power and water capacity provisioning, and land allocation for inference clusters. The benchmark must shift. It is no longer how many GPUs the government has bought. It is whether Indian inference can be priced in rupees and delivered at production-grade latency. An Indian developer should be able to migrate from a foreign inference API to a domestic endpoint without rewriting the application or waiting six months for model parity. Until that is true, India has compute but not inference capacity.
‍

Second, enhance IndiaAI grant transmission beyond general models and into vertical AI models and production use cases. The current pace of deployment is too slow, and too oriented towards logo collection rather than workflow validation. Grants must reach Indian startups building vertical AI capabilities in pharmaceutical research and drug discovery, semiconductor design, accounting and financial planning, vision, physical AI, and world models, and other domain-specific workflows. These companies have no clear priority into the IndiaAI grant system today, which remains oriented towards GPU imports and general model support. Private corporations and the venture ecosystem should be partnered with directly to channel capital into vertical AI builders with real customer commitments. Government procurement must direct projects to these companies to provide revenue support and prioritise indigenisation.
‍

Third, IndiaAI must establish a $5 billion AI Fund to back indigenous LLMs, small language models, AI applications, and local inference development, with co-investment from India's largest IT services companies. DeepSeek demonstrated that frontier-grade model capability can be built at a fraction of the cost incurred by the leading Western labs. China did not arrive at that position by accident. It has pioneered low-cost open-weight model development through deliberate state coordination and ecosystem subsidy. India must replicate that pattern. The fund can back dozens of indigenous model teams to develop LLMs and SLMs at meaningfully lower cost for specialised applications, including legal, healthcare, financial, educational, regional language, software development, and government workflows. TCS, Infosys, Wipro, HCL, and other Indian IT services leaders should be incentivised to commit a minimum of $1 billion to the fund, through favourable participation terms in IndiaAI-supported government procurement and vertical AI partnerships. PSUs and Government Agencies must be directed to pilot and procure Ai capabilities from these funded companies. The goal is not to produce one sovereign model. It is to produce several, each at meaningfully lower cost than global alternatives, that can onshore inference across multiple sectors.
‍

Fourth, use India's demand gravity to pull global model providers' inference infrastructure into the country. India is already the second-largest market for OpenAI and Anthropic. That is leverage. The TCS HyperVault arrangement and the Infosys-Topaz partnership show model providers will localise when commercial logic and policy invitation align. The government should formalise this as the default expectation, negotiating data residency and inference localisation commitments as a condition of preferred market access, the way every other major economy has done with cloud and data infrastructure.
‍

For three decades India imported energy and exported software. That bargain held the external account together. India is now entering a comparable arrangement at a different layer of the technology stack. The AI token economy is a real and growing claim on the dollar earnings that pay for our oil and gold.
‍

The window to act is open. India is building a chip and infrastructure foundation through the Semiconductor Mission and the hyperscaler investment cycle. The IndiaAI Mission has installed a foundational compute layer. A National Inference Mission is the floor that must be built on top of it. 
‍

By 2029, India should be a net exporter of AI value to the world, not a net importer of inference from it. The services surplus that has carried India's external account for two decades must be enhanced by this next move.
‍

DISCLAIMER

_{The views expressed herein are those of the author as of the publication date and are subject to change without notice. Neither the author nor any of the entities under the 3one4 Capital Group have any obligation to update the content. This publications are for informational and educational purposes only and should not be construed as providing any advisory service (including financial, regulatory, or legal). It does not constitute an offer to sell or a solicitation to buy any securities or related financial instruments in any jurisdiction. Readers should perform their own due diligence and consult with relevant advisors before taking any decisions. Any reliance on the information herein is at the reader's own risk, and 3one4 Capital Group assumes no liability for any such reliance.Certain information is based on third-party sources believed to be reliable, but neither the author nor 3one4 Capital Group guarantees its accuracy, recency or completeness. There has been no independent verification of such information or the assumptions on which such information is based, unless expressly mentioned otherwise. References to specific companies, securities, or investment strategies are not endorsements. Unauthorized reproduction, distribution, or use of this document, in whole or in part, is prohibited without prior written consent from the author and/or the 3one4 Capital Group.}

‍