The Week India Went Full-Stack AI

India’s AI ecosystem has had big moments, but the past week felt different. From Sarvam AI launching multilingual agents and Krutrim debuting India’s first agentic assistant, to CoRover releasing a multilingual AI model that works offline, something finally clicked.

These aren’t one-off announcements. This is India building its own AI stack, block by block. Before we dive deeper, let’s look at some of the top stories that made headlines this week: 

• Snowflake and Databricks just took their rivalry to the next level—acquiring Crunchy Data and Neon to bet big on PostgreSQL, the new backbone for AI-native applications and agentic speed. Why is everyone rushing to Postgres? Read to find out. 
• From MIT dropout to building a $14 billion AI powerhouse, Alexander Wang’s Scale AI just became Meta’s biggest bet in the AGI race with a $14.3 billion stake and a move to Meta’s SuperIntelligence Lab. What does this mean for the future of AI? Read more here. 
• AMD is done playing second fiddle. Its MI350 GPUs, Helios racks, and open AI stack signal a bold challenge to NVIDIA’s dominance and a $500 billion bet on the future of agentic AI. Read more here. 

Now, let’s explore some exciting collaborations and exclusive insights from the AIM ecosystem, presented with a unique twist beyond our niche editorial content.

• Still running your AI on batch pipelines? EPAM says it’s time to rethink your entire data stack, because in the age of agentic AI, legacy systems won’t cut it. Read the full story to see what the future of intelligent data platforms looks like. 
• In the age of AI-powered operations, your biggest competitive edge isn’t just technology—it’s talent. Accenture reveals why a ‘reinvention-ready’ workforce is key to scaling GenAI and unlocking growth. Read the full story to see what skills the future demands.

If you are looking to network, learn, and stay ahead in the AI game, here are some exciting upcoming events from AIM and its partners: India’s most powerful GCC conference is happening in Goa this weekend. MachineCon 2025 is where the world’s top 100+ GCC leaders will converge to shape the future of innovation, transformation, and strategy. Know more.

Right before that, India’s biggest cloud and AI powerhouses will take the stage at AWS Summit Mumbai 2025 on June 19. It’s going to be an unmissable day of bold ideas and tech breakthroughs with visionary leaders. Grab your free pass now!

Next, we have ABBYY’s AI Pulse Developer Conference, which is coming to Bengaluru on July 9-10, 2025. Join top automation experts, devs, and innovators for hands-on sessions, live demos, and exclusive rewards. Register now to save your seat! Now, shifting gears to all the exciting developments lighting up the Indian AI ecosystem…

What’s Behind The AI Buzz This Week?

Let’s start with Sarvam AI. The startup rolled out Samvaad, a conversational platform that supports 11 Indian languages and works across phones, WhatsApp, mobile apps, and the web. In addition, it offers real-time insights and can go from PoC to production in days.

That’s not all. The company also launched Sarvam-Translate for long-form Indic translation (think LaTeX, HTML, scanned documents), and released Sarvam-M, its 24B parameter language model. It supports 10 Indian languages and is fully open weights, a key milestone under the IndiaAI Mission.

Just days later, Krutrim launched Kruti, India’s first agentic AI assistant—not a chatbot, but an actual doer.

You can ask Kruti to book your cab, pay your bills, research a topic, or generate an image. It talks back in 13 Indian languages and learns your preferences over time. It’s built on Krutrim V2, their 12B parameter LLM, trained from scratch.

With Kruti, Krutrim isn’t building an assistant alone; it’s offering SDKs, APIs, image/voice models, and platform-level orchestration. It’s quietly becoming a developer platform. CoRover, Soket, BharatGen: The Build-from-Scratch Club

Meanwhile, CoRover dropped BharatGPT Mini, a 534M parameter model that works offline, supports 14 Indian languages, and is already being used in sectors like healthcare and banking. With rising cloud costs, this small model may have a big future.

Soket AI Labs took it a step further.

Led by Abhishek Upperwal, the team is building a 120B parameter Indic model from scratch and open-sourcing it. Their 7B version is expected in six months. They are digitising archives, applying OCR to Indic texts, and even fixing grammatical errors in Hindi.

Then there’s BharatGen, backed by the government of India, that quietly released Param-1, a 2.9B bilingual LLM built entirely from scratch, with 25% Indic data (compared to Llama’s 0.01%).

Param-1 is now on AIKosh, India’s open-source AI repo, alongside 20 high-quality Indic speech models. This is a goldmine for developers looking to build Indic copilots.

And Then There’s TWO AI

While others build token-based LLMs, Pranav Mistry’s TWO AI is building models like SUTRA for real-world data—weather, traffic, and transactions.

A few days ago, TWO AI just partnered with Coxwave to scale conversational AI across APAC. Backed by Reliance Jio and Naver, the company is already at $10 million ARR and aims to be profitable by 2026.

Does It All Add Up?

For years, India has spoken about ‘sovereign AI’. This week, it finally felt like the pieces were in place with multilingual deployment tools (Samvaad, Kruti, BharatGPT), base models built from scratch (Param-1, Pragna-1B), developer ecosystems (AIKosh and Krutrim SDK), open weights and transparent roadmaps. 

Most importantly, these models aren’t just English-first chatbots. They are built for how India speaks, thinks, and works.

The best part? This is just the beginning.

[Must Watch] If you want to understand how we got here, from Aadhaar to DPI to AI, don’t miss the latest episode of AIM TV’s What's the Point with Shankar Maruwada, co-founder of EkStep and one of the key minds behind Aadhaar.