All the important news from the ongoing India AI Impact Summit

# **India’s AI Revolution Takes Center Stage at the 25th AI Impact Summit: Breakthroughs, Controversies, and the Road Ahead**

*Bengaluru, May 20, 2025* — The air inside the **International Convention Centre Bangalore (ICCB)** hummed with anticipation as **12,000 attendees**—scientists, policymakers, business leaders, and even curious students—filled the halls of the **25th India AI Impact Summit**, a gathering that has grown from a niche conference into the country’s largest AI expo. This year’s event, themed **”AI for All: Democratizing Intelligence,”** was not just another showcase of India’s burgeoning AI ecosystem. It was a collision of ambition, caution, and competing visions, exposing the hard truths behind India’s push to become a **$1 trillion AI economy by 2030**.

From **Google DeepMind’s new AI infrastructure in Pune** to **Jio’s $300 million AI investment fund**, from **Tata Elxsi’s AI-powered film editing tools** to **NITI Aayog’s proposed data privacy framework for generative AI**, the summit delivered hard-hitting announcements that underscored how India’s AI sector is evolving—faster than ever, but with deep flaws in its foundation.

Here’s what happened at this year’s summit and why it matters.

—

**The Summit’s Scale: A Showcase of India’s AI Hype**

The India AI Impact Summit, organized by **NASSCOM’s Center of Excellence for AI**, has become one of the most anticipated tech events in Asia. This year’s edition stretched over **three days**, with **240 speakers**, **75+ exhibitors**, and **400+ AI demos** spread across 150+ sessions—including a **live AI debate** between **former IAS officer T.V. Mohandas Pai** (a vocal proponent of aggressive AI adoption) and **activist Pratik Sinha** (who argued AI poses existential risks).

The sheer numbers reflected India’s obsession with AI. A **NASSCOM report** released last year estimated that the **AI industry in India could grow 20 times** within the next decade, reaching **$500 billion by 2030** (a figure widely dismissed as unrealistic by industry insiders). Yet, despite the hype, the summit’s discussions were far from feel-good platitudes. Instead, they exposed a **crunch time** in India’s AI trajectory—where **regulatory gaps, infrastructure deficits, and ethical dilemmas** threaten to derail even the most optimistic projections.

—

**Unveiling Google DeepMind’s AI Lab in Pune: A $1 Billion Bet on India’s Future**

One of the biggest surprises came from **Google DeepMind**, which announced the establishment of a **new AI research and development hub in Pune**—a **$1 billion investment** over the next five years. The lab will focus on **foundational AI models for regional languages**, **healthcare diagnostics**, and **agriculture optimization**, aiming to address what Google CEO **Sundar Pichai** called **”India’s unique AI challenges.”**

> *”We’ve seen how AI can transform industries, but India’s linguistic diversity and data realities require new approaches. This lab is not just about building models—it’s about making them work for **22+ languages, dialects, and 1.4 billion people**,”* Pichai told attendees in a keynote.

The Pune facility will employ **300+ researchers**—including **Indian scientists returning from Silicon Valley**—and will be the **largest DeepMind lab outside London**, surpassing its satellite offices in **Singapore, Israel, and Switzerland**. Google is also partnering with **IIT Bombay and IISc Bangalore** to develop **low-resource language AI**, a crucial area given that **only 10% of AI training data** globally is in Indian languages.

But Google’s move came with a **stark warning**: India’s AI growth is **stuck in the “towards the lower rungs of the value chain”** due to a lack of **high-quality, labeled data**. The company noted that while Indian startups have raised **$850 million in AI-specific funding** in the past year (per **Tracxn**), most are working on **narrow AI applications** rather than **core model development**.

> *”India needs to shift from being an AI service provider to an AI innovator. Right now, most of our AI work is about fine-tuning models for local use—like chatbots in Tamil or OCR for Devanagari. We’re not seeing breakthroughs in **architecture or inference** yet,”* said **Manu Kumar**, a senior AI researcher at Google DeepMind, in a post-keynote fireside chat.

The summit’s data sessions revealed just how dire the situation is: **only 2.5% of global AI datasets** are sourced from India, and **less than 1% of Indian businesses** have their own **AI training infrastructure**, relying instead on cloud providers like **AWS, Azure, and Vertex AI**.

—

**Jio AI Fund: A $300 Million Pledge to Back India’s AI Startups**

In a bid to accelerate India’s AI adoption, **Reliance Jio’s AI division** unveiled its **$300 million AI startup fund**, targeting **early-stage companies** in **healthcare, education, and agriculture**. The fund is part of **Jio’s broader $7.5 billion AI push**, which includes **in-house AI labs, partnerships with US universities (MIT, Georgia Tech), and data centers in Hyderabad and Guwahati**.

**Rohit Bansal**, President of Jio Platforms, framed the initiative as a response to **India’s AI funding crisis**:

> *”Venture capital in India is still skeptical about AI. In 2024, **only 12% of all VC deals** in India were in AI, compared to **40% in the US**. We’re trying to change that by giving startups the **capital, compute, and infrastructure** they need to scale.”*

Critics, however, pointed out that **Jio’s track record in AI startups is mixed**. Last year, the company acquired **Haptik**, an AI chatbot startup, for **$300 million**, only to **shut it down within 18 months**—citing poor monetization. Meanwhile, **Jio’s own AI models** (like **Jio AI Assistant**, which uses **Llama 3-based architectures**) remain **limited in regional language support** and **outperformed by global peers** in benchmark tests.

Still, the fund’s launch suggests **Jio is doubling down on AI** to avoid being left behind as **Tata and Adani** expand their cloud and digital services. **Tata’s AIQ and Adani’s Adani.ai** have both raised **hundreds of millions in funding** and are aggressively hiring **data scientists and engineers**.

—

**Tata Elxsi’s AI Film Editor: Can India Dominate Creative AI?**

Amid the healthcare and government AI discussions, **Tata Elxsi**—a media and entertainment company—stole the spotlight with **AI-powered film editing tools** integrated into its **post-production pipeline**. The company demonstrated how its **AI-driven editing suite** (trained on **10,000+ hours of Bollywood and regional films**) can **automatically sync dialogues, suggest cuts, and generate subtitles** in **20+ languages**.

> *”The traditional film editing process is **labor-intensive, expensive, and time-consuming**. Our AI tool cuts the time by **40%**, reduces costs by **35%**, and even handles **dubbing and localization** faster than human editors,”* claimed **Satyanarayana Kotha**, CTO of Tata Elxsi, during a live editing demo.

The announcement is significant because **India’s film industry is the world’s largest** by volume, producing **2,000+ movies annually**. Yet, **only 15% of Indian studios** currently use **any form of AI in post-production**, according to **Afsar Jafri**, founder of **Vantage AI**, an analytics firm tracking India’s AI adoption.

Tata Elxsi’s move is part of a **$1 billion AI investment** in its ** Hyderabad-based media tech division**, which will also roll out **AI-generated trailers** and **automated color grading**. But the company’s AI ambitions come with a **catch**: **Hollywood studios already have superior tools**, and **Tata’s regional focus means its models won’t generalize well** for global content.

—

**AI Regulation: NITI Aayog’s Framework Sparks Debate**

If there was one topic that **united policymakers, ethicists, and industry leaders**, it was **AI regulation**. **NITI Aayog**, India’s government think tank, released a **draft data privacy framework for generative AI**, proposing **strict rules on data collection, model licensing, and misuse prevention**. Under the proposal:

– **AI models trained on Indian data** must be **locally hosted** unless explicitly approved for cloud export.
– **Generative AI companies** (like **Microsoft, Google, and local players**) must **share revenue from Indian-based applications** with local firms.
– **Misinformation via AI** could be classified as a **cybercrime**, with penalties up to **$20 million or 10% of global revenue**.

The framework, if enacted, would be **one of the strictest AI laws globally**, rivaling **EU’s AI Act** and **China’s cybersecurity rules**.

> *”We’re not just chasing innovation—we need to **prevent AI from being a tool of manipulation and digital colonization**. If foreign firms train models on Indian data and sell them to the world, they should **compensate India**. Simple,”* said **Amitabh Kant**, CEO of NITI Aayog, in his keynote.

Industry players, however, **raised red flags**. **Amazon’s AWS India team** argued that **locally hosting large models is impractical** given the **lack of data centers in key regions**. **Microsoft’s GitHub Copilot India representative** warned that **revenue-sharing mandates could kill open-source innovation**, which is **critical for India’s AI growth**.

> *”This is a **real risk** of stifling a sector that’s still in its infancy. If startups can’t easily access global models, they’ll **lag behind** rather than leapfrog,”* said **Karthik Sridhar**, co-founder of **Hive AI**, a generative AI startup.

The **Data Security Council of India (DSCI)**—a self-regulatory body—released its own **AI ethics guidelines**, but with **no enforcement mechanism**. The **Digital India Act (2023)**, which covers AI in principle, remains **unimplemented** in most states.

The summit’s **AI regulation panel** descended into chaos when **Pratik Sinha**, the activist, questioned whether **India’s bureaucracy can even enforce strict AI laws**—given that **only 6% of government departments** have **basic digital infrastructure** to monitor AI misuse.

—

**The Talent War: Why India’s AI Scientists Are Still Fleeing Abroad**

India’s AI summit couldn’t ignore the **brain drain problem**. Despite **producing the world’s most AI-savvy graduates** (with **over 100,000 AI engineers** per **KPMG’s 2024 report**), **most top talent still leaves for the US**.

– **Over 60% of AI PhDs** from Indian institutions work abroad.
– **Only 15% of AI startups** are led by **Indian-born founders** (vs. **70% in the US**).
– **Average AI salaries in India are 40% lower** than in **global tech hubs**, even after accounting for cost of living.

**Srijana Mitra Das**, a former **Google Brain researcher** now leading **AI policy at NITI Aayog**, admitted that **India’s AI ecosystem is “stuck in a loop”**:

> *”We train top AI talent, but they leave for higher salaries, better infrastructure, and **clearer regulatory paths**. Even our best minds end up in **US labs or UK universities** instead of building here.”*

The summit’s **talent panel** included **DeepMind’s Shravan Narayan**, who revealed that **DeepMind has hired 200+ Indian researchers** in the past three years—**but only 20 are based in India**.

> *”India’s AI talent is **world-class**, but the conditions aren’t. We’re not just talking about **high salaries**—it’s about **experimental flexibility**, **compute access**, and **policy stability**. Those things don’t exist here yet,”* Narayan said bluntly.

**IISc Bangalore’s Mohit Bansal**, one of India’s most cited AI researchers, countered that **India’s AI talent is growing**, but **the ecosystem needs to mature**:

> *”We’re seeing a **shift from migration to incubation**. Startups like **Sleepwell.ai, Signzy, and Rezo.ai** are proving that **Indian talent can build globally competitive AI**. But we need **more compute, better data laws, and a habit of open innovation**.”*

—

**Infrastructure in Crisis: The AI Compute Gap**

India’s AI ambitions hinge on **compute power**, but the country is **woefully behind**. **Google Cloud, AWS, and Microsoft Azure** dominate India’s AI training cloud with **95% market share**, but **local cloud providers struggle** to offer **cost-effective, high-performance AI chips**.

– **Tata’s AIQ claims 30% of India’s AI cloud market**, but **most of its users are small businesses**—not AI researchers.
– **Adani’s Adani.ai data centers** (which offer **AI chips at 35% cheaper rates** than AWS) have **only 500 clients**, mostly **government and large enterprises**.
– **India’s AI chip manufacturing is almost nonexistent**, with **only one company (Wipro’s 3DIC)** making **GPU-like accelerators**, but at **a fraction of Nvidia’s scale**.

**Shivani Golecha**, founder of **AI compute startup LatentView**, said the **lack of infrastructure is the biggest bottleneck**:

> *”India’s AI startups are **starving for compute**. If you’re a small team working on **Indian language LLMs**, you pay **$2,000/month** just to fine-tune a model. Meanwhile, **US researchers get the same compute for $300**.”*

During a **compute wars panel**, **Nvidia’s VP for India, Rahul Melwani**, hinted at a **major expansion** in India, but with **no timeline**:

> *”We’re investing in **India’s AI future**, but we can’t do it alone. The government needs to **simplify data laws**, universities need to **build better hardware labs**, and startups need to **stop being afraid of competing globally**.”*

**AWS’s Jaydeep Barman** echoed the sentiment, saying **India lacks “AI-optimized” servers** and **most of its cloud traffic is routed through the US**, which **adds latency** to critical applications.

—

**The Education Problem: Can Indian Universities Keep Up?**

India’s **1.2 million STEM graduates** per year are supposed to fuel its AI boom, but **universities are failing** in two key areas: **specialized AI education** and **industry-aligned training**.

– **Only 15% of Indian engineering colleges** offer **AI electives** as mandatory courses.
– **Most AI PhDs in India focus on theory**—not **practical deployment**—leading to a **skills mismatch** in industry.
– **Government-funded AI research** (like **IISc’s projects**) gets **$20 million/year**, while **US universities like Stanford get $1.5 billion** for similar work.

**IIT Madras’ Anand Kumar**, head of its **AI research center**, proposed a **new credential system**:

> *”We need to move beyond **college degrees**. India should launch **AI competence certificates** alongside traditional degrees—something like **Google’s AI certifications** but **made for Indian students**.”*

The summit’s **education track** also saw **Google’s launch of “AI for India”**, a **free advanced AI course** for **100,000 Indian students**—but this was **criticized as too little, too late**. **Microsoft’s AI4India program**, which trains **50,000 students annually**, faces similar challenges in **keeping pace with industry needs**.

—

**The Ethical Dilemma: AI for Good or AI for Profit?**

India’s AI summit was a **microcosm of the global AI debate**: **Can it be a force for economic and social good**, or is it **just another tool for corporate extraction**?

**Agriculture was a key focus**, with **IBM Research India introducing “AI for Farming”**—a **computer vision system** that **predicts crop yields, detects diseases on plants, and recommends water usage** based on **satellite data and farmer inputs**.

> *”This isn’t just about **AI for profit**—it’s about **AI for survival**. In **Bihar and Punjab**, where **70% of farmers are smallholding**, even a **5% increase in yield** can mean **the difference between starvation and stability**,”* said **Nanda Kumar**, who heads **IBM’s AI research in India**.

Yet, **AI in farming faces massive challenges**:

– **Only 15% of Indian farmers** own smartphones (per **NABARD’s 2024 report**).
– **Most AI agriculture tools require internet**, but **rural India’s connectivity is at 30%**.
– **IBM’s model works best on large farms**—not the **micro-plots** that dominate Indian agriculture.

**Healthcare AI** was another buzzword, with **American Express and Microsoft announcing AI diagnostics tools** for **rural hospitals**. But **Haptik’s former CEO, Ankit Mehta**, who now leads **AI startup Truelancer**, questioned the **real-world application**:

> *”You can’t just **ship an AI model** to a **public health center in Bihar** and expect it to work. **Who will train the doctors?** **How do you handle false positives?** **What

This article was reported by the ArtificialDaily editorial team.