India AI: 3 Startups to Build Indigenous Foundation Models

India’s AI landscape is buzzing with fresh momentum as the government pushes hard to establish a robust indigenous foundation model ecosystem. As of May 2025, three more startups have been selected to build homegrown foundation models, joining Sarvam AI, the pioneer startup that kicked off this ambitious journey earlier this year. Alongside this, India’s common compute capacity has skyrocketed past 34,000 GPUs, signaling a major infrastructural leap to support cutting-edge AI research and development. But what does all this mean for India’s AI future? Let’s dive deep into this unfolding story that’s reshaping the nation’s digital ambitions.

The Rise of India’s Indigenous AI Foundation Models

Foundation models—large-scale AI systems trained on massive datasets—are the backbone of today’s AI revolution, powering everything from chatbots to complex decision-making systems. India’s approach to building these models is nuanced and strategic. Instead of relying solely on global AI giants, the government has launched the IndiaAI Mission’s Foundation Model pillar to foster domestic innovation that reflects India’s unique linguistic, cultural, and business nuances.

Since the launch, over 500 proposals poured in by April 30, 2025, reflecting immense interest from startups and research groups eager to contribute to this national effort[1][2]. The government’s multi-stage expert evaluation process has now shortlisted three new startups to join Sarvam AI in this mission.

• Sarvam AI: The trailblazer chosen in April 2025 to build India’s first domestic foundational AI model. Sarvam AI has already raised $41 million in Series A funding, one of the largest for an Indian AI startup, backed by Lightspeed, Peak XV Partners, and Khosla Ventures[4]. Their model promises to incorporate India-specific data, aiming for applications ranging from healthcare to finance.

• Soket AI: Selected to build India’s first open-source 120-billion parameter foundation model. This is particularly exciting given India’s linguistic diversity. Soket AI’s model will be optimized for multilingual applications with practical use cases in defense, education, and healthcare[3]. Open-sourcing this model could accelerate innovation and democratize access.

• Gnani AI: Tasked with developing a 14-billion parameter multilingual Voice AI foundation model capable of real-time speech processing and advanced reasoning. This aligns well with India’s need for robust voice interfaces, especially in vernacular languages[3].

• Third Startup (unnamed): Alongside these, a third startup has been selected, details of which are expected soon, further expanding the ecosystem.

This multi-pronged startup selection highlights a deliberate strategy to cover foundational AI across modalities—text, speech, and open collaboration—tailored for India’s complex, multilingual environment.

Scaling Compute: From Vision to Reality

None of this would be possible without powerful computing infrastructure. The IndiaAI Mission has massively expanded the nation’s common compute capacity to over 34,000 GPUs as of May 2025, a quantum leap from previous years[1][2]. This pool of computational power is available for startups and researchers working on AI foundational models, breaking down one of the biggest barriers to AI innovation: access to high-end hardware.

To put this in perspective, 34,000 GPUs represent one of the largest aggregated AI compute capacities in emerging markets, enabling training of models with billions or even hundreds of billions of parameters. This infrastructure boost dovetails with the government’s broader Electronics Component Manufacturing Scheme (ECMS), which, while primarily focused on electronics manufacturing, also supports AI hardware ecosystems through enhanced design innovation and quality standards[4].

Government’s Holistic AI Ecosystem Vision

Information Technology Minister Ashwini Vaishnaw has repeatedly emphasized the government’s holistic approach. Drawing parallels with India’s semiconductor ambitions—where efforts span chip fabrication, OSAT (Outsourced Semiconductor Assembly and Test), design, and critical materials—the AI ecosystem is being nurtured from multiple angles: hardware, software, talent, and research[3].

Notably, AI-focused skilling initiatives are underway in over 240 universities nationwide, aimed at building a deep talent pool. The government recognizes that talent is the lifeblood of innovation. With AI experts in high global demand, India is ramping up educational programs and research incentives to retain and grow homegrown expertise.

Why Indigenous Foundation Models Matter

At first glance, the idea of building “indigenous” AI models may sound like a purely nationalistic endeavor. But it’s really about creating AI systems that understand and serve India’s unique context better than off-the-shelf models from Silicon Valley.

Consider India’s vast linguistic diversity—over 20 major languages and hundreds of dialects. A foundation model trained primarily on English or Mandarin will struggle to capture the nuances of Hindi, Tamil, Bengali, or Telugu. This linguistic gap impacts everything from customer service chatbots to government digital interfaces.

Moreover, India-specific data encapsulates unique socio-economic patterns, cultural norms, and regulatory frameworks. Indigenous models can be designed to respect privacy norms, local laws, and ethical considerations in ways global models may overlook.

Real-World Applications and Impact

The startups selected are focusing on domains critical to India’s growth:

• Healthcare: AI models can assist in diagnostics, patient monitoring, and personalized medicine, especially in rural areas with limited access to specialists.

• Education: Multilingual AI tutors and content generators can help bridge educational divides, enhancing learning for students in regional languages.

• Defense and Security: Advanced AI models optimized for local languages and contexts can improve surveillance, threat detection, and strategic decision-making.

• Finance: AI-powered credit scoring, fraud detection, and customer support tailored for diverse Indian users can drive financial inclusion.

Soket AI’s open-source approach, in particular, could catalyze a wave of innovation by enabling startups and researchers across the country to build on a common foundation.

Challenges and Looking Ahead

Building foundation models is not without challenges. Training models with tens or hundreds of billions of parameters requires massive datasets, computational resources, and deep expertise. Ensuring data privacy, avoiding bias, and creating explainable AI are ongoing concerns.

However, India’s strategy of combining government support, academic partnerships, startup innovation, and infrastructure expansion is a promising formula. The influx of over 500 proposals in just a few months shows a vibrant ecosystem ready to leap forward.

We can expect more startups to join this fold, pushing the boundaries of AI applications tailored for India’s unique needs. The government’s commitment to expanding compute resources and talent development will be crucial in sustaining this momentum.

Comparison Table: India’s Foundation Model Startups Selected in 2025

Final Thoughts

India’s AI ambitions in 2025 are more than just a tech push—they’re a strategic move to claim a seat among global AI leaders by building systems that understand and empower its own people. With startups like Sarvam AI, Soket AI, and Gnani AI leading the charge, backed by unprecedented compute capacity and government commitment, India is carving out a distinctive AI identity.

As someone who’s been tracking AI’s global evolution for years, I find India’s approach refreshingly comprehensive. It’s not just about importing technology but cultivating an ecosystem that respects and harnesses local realities. The coming years will be fascinating as these foundation models start powering real-world applications, transforming industries, and maybe even rewriting the rules of AI development on the world stage.

**