For a long while now, India has been the world’s back office for coding and software services. Now, with AI writing code, debugging systems, and automating workflows, the country’s IT industry dreads an existential crisis. Investors are treading cautiously, companies are recalibrating hiring plans, and layoffs are beginning to signal structural shifts. AI is likely to compress lower-value IT tasks. Pressure is building on traditional services models, and demand for repetitive coding roles is set to decline.

At the India AI Impact Summit, Infosys Chairman Nandan Nilekani captured this tension and warned, “The resentment of the blue-collar workers led to the train wreck of globalization. The resentment of the white-collar worker is going to lead to the train wreck of artificial intelligence.”

The India AI Impact Summit unfolded in New Delhi as a policy response to the very disruption reshaping India’s software economy. From sovereign large language models and chip design incentives to subsidized AI compute and telecom modernization, the government signaled that India intends to move up the value chain—from coding services to foundational AI infrastructure.

Sovereign AI infrastructure

If AI is compressing low-value coding work, control over the model layer becomes strategic. Prime Minister Narendra Modi launched the latest model releases of BharatGen, which was described by Union Minister for Science and Technology Jitendra Singh as a government-owned sovereign multilingual and multimodal large language model initiative.

BharatGen isn’t a single model release. It’s a coordinated national AI stack whose architecture spans multiple modalities. The latest releases include Param-2, a 17-billion-parameter text foundation model trained across 22 scheduled Indian languages. The stack also includes Shrutam, a speech-to-text model in 12 Indian languages; Sooktam, a text-to-speech model; and Patramunder the DocBodhframework for multilingual document understanding.

BharatGen operates through a consortium led by the TIH Foundation for IoT and IoE at IIT Bombay, alongside partner institutions including IIIT Hyderabad, IIT Hyderabad, IIT Mandi, IIT Kanpur, IIM Indore, and IIT Madras. The Department of Science and Technology supports it with ?235 crore (~$25 million) under the National Mission on Interdisciplinary Cyber-Physical Systems. It’s further strengthened under MeitY’s ?10.585 crore (~$1.16 million) India AI Mission outlay.

Singh emphasized that BharatGen’s defining feature is sovereignty. The models are government-owned, and the datasets are curated domestically. The initiative has transitioned into a Section-8 company, the BharatGen Technology Foundation, to operate at a national scale. Data and model governance will be anchored by initiatives such as Bharat Data Sagar.

For engineers, the significance lies in stack control. BharatGen supports multimodal pipelines. It enables Indian-language tokenization at scale as well as domain-specific fine-tuning, including Ayur Param, Agri Paramand Legal Param. It also enables deployment-ready platforms designed for governance, healthcare, agriculture, and legal systems.

Design-led semiconductor strategy

Parallel to AI infrastructure, the government is reshaping semiconductor policy through the Design Linked Incentive (DLI) scheme. The focus isn’t fabrication—it’s design. Semiconductor design can contribute up to 50% of value addition and account for 15 to 35% of the bill-of-materials cost. It also anchors architectural control.

India already hosts major global design centers. The DLI strategy aims to convert that service strength into IP ownership.

Vervesemi Microelectronics offers a working example. The fabless startup raised $10 million in a Series A funding round co-led by Ashish Kacholia and Unicorn India Ventures. The company has built a portfolio of more than 140 semiconductor IPs, developed 25 IC variants, and secured 10 patents. It has completed multiple tape-outs at UMC and TSMC.

Its chips include a BLDC controller built on an indigenous RISC-V core, an avionics data acquisition chip, an energy metering chip, and a motor control chip for electric vehicles and drones.

Indian Union Minister for Electronics and IT Ashwini Vaishnaw framed the transition from services to silicon bluntly. “India must now evolve into a product nation,” he said. He called for an incremental development approach spanning the full technology spectrum.

The shift is clear—from outsourcing to silicon ownership, from design centers to chip companies, and from licensing IP to building domestic RISC-V cores.

For engineers, the ripple effects are tangible. RISC-V acceleration gains momentum. Analog design converges with machine learning workloads. Edge AI silicon becomes a domestic opportunity. Defense- and avionics-grade chip design moves into sharper focus.

Subsidizing compute, not data centers

India’s also chosen a different economic lever for AI growth. Earlier last week, EE Times reported that India will expand its AI compute capacity beyond 38,000 GPUs, adding 20,000 more units in the coming weeks, available at a subsidized rate of ?65 per hour (approximately 72 cents per hour).

Rather than subsidizing GPU data centers directly, the Indian government is subsidizing compute access. MeitY Secretary S. Krishnan stated that AI compute in India is available at roughly one-third the cost paid elsewhere in the world.

That pricing strategy reshapes the ecosystem. It underwrites research markets, lowers entry barriers for startups, and avoids infrastructure lock-in. It encourages private data center investments without distorting ownership structures.

At the telecom layer, AI is becoming embedded in network orchestration. Policy focus areas include AI-native network management, Open RAN, 6G, AI-driven cybersecurity, and non-terrestrial networks.

The hardware implication is structural. AI is no longer an overlay. It is becoming part of the network control plane.

Global commitments and deployment signals

Global and domestic AI enterprises, including OpenAI, Google DeepMind, Microsoft, Anthropic, Meta, Cohere, G42, and Sarvam, participated in the summit’s inauguration. Two commitments emerged: research into real-world AI usage, and multilingual, contextual model evaluation.

India is pushing for benchmarks that reflect the Global South. It demands cultural adaptation and multilingual safety standards.

On the exhibition floor, that ambition moved from policy to hardware. Humanoid robots, robotics processors, AI-ready high-performance computing infrastructure, industrial automation systems, smart agriculture robotics, and fintech AI security platforms filled the expo.

AI is no longer confined to cloud software. It’s becoming an industrial layer moving onto manufacturing floors, entering warehouses. It’s even appearing in public distribution systems and healthcare diagnostics in India.

India clearly doesn’t want to remain a peripheral IT services player. It wants to operate as a system-level sovereign technology power. However, to achieve that futuristic goal, speakers at the summit identified the need to establish offline AI capability and independence from internet connectivity. These requirements reflect operational realities in mission-critical systems, such as space systems, defense, satellite imagery, and geospatial analytics.

They also pointed out the need for explainable AI, version control, data lineage tracking, and regulatory-aligned machine learning pipelines in public systems. While government adoption frameworks stress model drift detection, third-party audits, scientific validation, and due diligence before beneficiary targeting, it remains an open question whether this sovereign AI architecture will withstand the pressures of deployment and accountability.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved