Can India Capitalise On The AI Homecoming?

extract

For decades, Indian engineers and researchers moved to Silicon Valley, joining global technology companies and contributing to products and research that shaped the industry. Now, the tides are turning. Indian talent with experience at some of the world’s most influential AI labs is making its way back home to build startups and research organisations of its own. This emerging trend could signal a new wave of reverse brain drain. But what has changed to make India a more credible place to build AI companies for those who have worked at the world’s leading research labs? And why is this happening now? Let’s ponder in this edition of The AI Shift… A Land Of Opportunities The shift is being driven by a combination of entrepreneurial ambition and a changing Indian AI ecosystem. Experienced researchers are beginning to see an opportunity to build companies around their own research agendas, while India's growing startup landscape and emerging AI infrastructure are making that ambition more achievable. Shyamal Hitesh Anadkat, a former OpenAI researcher who returned to India to build Empirical Machines, a startup focused on accelerating scientific discovery, pointed to this change in an interview with Financial Express. He cited growing momentum in the startup ecosystem, the emergence of companies such as Sarvam AI, and a stronger appetite for risk among researchers and founders. The sense that more talent was beginning to rally around India’s AI opportunity had also become an important factor. Parth Sarthi, cofounder of Chariot, offers another perspective. After completing his masters from Stanford University, Sarthi joined Contextual AI and later Google DeepMind as a research engineer. His work at DeepMind included multimodal reasoning for Gemini Deep Thinking, involving systems that reason across text, numbers and images. Sarthi had long felt that India needed to develop similar frontier research capabilities. Limited GPU access had previously constrained the research that could be conducted locally. The IndiaAI Mission’s call for proposals to develop indigenous foundation models made the prospect of building in India more tangible. He saw an opportunity to apply his experience to developing advanced AI capabilities in India rather than continuing to build similar systems for Western markets. Government support is also helping make the proposition more credible. The IndiaAI Mission is expanding access to compute, supporting indigenous foundation models and providing assistance for AI startups. These initiatives do not eliminate the gap with global AI labs, but they can lower some barriers to experimentation and company formation. Not to forget, AI funding in India grew 235% YoY in Q3 2026, as per Inc42’s latest report. From Global AI Labs To Indian Startups Anadkat is among the clearest examples of this shift. After nearly four years at OpenAI, where he led the Applied Evals team, he returned to India in 2026 to build here. The startup aims to use AI to accelerate research involving molecules and materials. According to Moneycontrol, Empirical Machines is in talks to raise around $15 Mn from Peak XV Partners and Bessemer Venture Partners at a reported valuation of approximately $80 Mn. Sarthi's Chariot is another example of experienced researchers attempting to build frontier AI capabilities from India. The company is developing speech-native foundation models and was among the teams selected under the IndiaAI Mission to develop foundational models. His earlier research included RAPTOR, a retrieval technique for language models published at ICLR 2024 and subsequently incorporated into frameworks such as LangChain and LlamaIndex. The movement also extends to specialised AI infrastructure. Anish Das Sarma, a former Google engineering director, founded Reinforce Labs to build tools that evaluate, red-team and remediate enterprise AI systems. The startup was featured in Peak XV's Surge cohort, reflecting demand for infrastructure that helps businesses deploy AI more reliably. Meanwhile, Sarvam AI has drawn experienced researchers into its effort to build Indian foundation models. These examples represent different forms of talent movement: researchers returning to start companies, experienced technical leaders launching specialised ventures, and global researchers advising Indian startups. They point to a growing willingness to build AI capabilities outside traditional technology hubs. Are Investors Following The Trend? For venture capitalists, experience at OpenAI or Google DeepMind can signal familiarity with advanced research, model development and the challenges of taking AI systems into production. It may also help founders recruit specialist talent and establish credibility with customers and investors. But pedigree alone does not establish whether a company can succeed. Sayantan Sarkar, principal at Stellaris Venture Partners, believes hands-on experience building models at frontier AI labs offers three distinct advantages. Researchers who have worked closely with advanced models can anticipate how the technology is likely to evolve, identify gaps in existing capabilities and spot opportunities to build products for emerging needs. Their understanding of post-training techniques can also help improve models and develop better products. In addition, their credibility within the AI ecosystem can help attract high-quality researchers and engineers, including those from frontier labs. Ultimately, Sarkar argues, the advantage is not simply having worked at a leading AI lab, but being able to translate that experience into a differentiated company. Rahul Mathur, the vice president at DeVC and a pre-seed investor, approaches the question from the founder’s perspective. Mathur believes investors should start by understanding the founder: why they left their previous role, what motivates them to pursue the opportunity, and whether they have the resources and commercial judgement to build a sustainable business. He also examines the team they have assembled and why its members are willing to leave comfortable or well-paying jobs to join the startup. For AI research ventures, Mathur asks a more fundamental question: what customer-backed research is being conducted, and who will pay for it? He is looking for a “resourceful & commercially sound founder”. The two perspectives highlight the distinction between technical advantage and investment merit. A researcher may understand how to build an advanced model, anticipate the next wave of capabilities and recruit exceptional talent. But those strengths must translate into a product that solves a real problem and creates a defensible business. That challenge is particularly acute in AI, where research can take years to translate into a commercial product, while compute, experimentation and specialist talent demand substantial investment. Investors must establish whether a founder can identify a market, prioritise the right research and build a sustainable business around the technology. A strong research pedigree may open the door, but customer demand, execution and commercial judgement determine whether a company can endure. For India, the more consequential shift would be one in which experienced researchers build organisations that generate proprietary research, train new talent and sell globally competitive products. The measure of positive reverse brain drain will not be how many researchers leave global labs but how many enduring AI businesses they build from here. Top Stories From India & Around The World RapidCanvas Nets $29 Mn: The startup raised $29 Mn in a Series B round led by Vertex Ventures SEA & India to expand go-to-market operations, grow its AI engineering team and strengthen enterprise AI deployment. Confido Health Raises $12.5 Mn: The healthtech startup raised $12.5 Mn in a Series A extension led by Z47 to expand its AI-powered CareOS platform and automate more administrative workflows across healthcare providers. Anthropic Opens Its Models To Safety Evaluators: Anthropic has given researchers at nonprofit METR employee-like access to assess its models, with the independent evaluator expected to publish risk assessment findings as the lab expands external scrutiny of AI safety. Meta’s Algo Under Review: The Ministry of Electronics and Information Technology has asked Meta to review its content recommendation algorithms, renewing scrutiny of how AI-driven systems rank and distribute content. SaaS Models Under Pressure: Zoho cofounder Sridhar Vembu has sparked a debate on AI economics, warning that rising server prices and token expenditure are pushing up operating costs as software companies grapple with shrinking margins and uncertain returns on AI-driven productivity gains. The Weekly Buzz: Meta’s Muse Wonder Claims The Frontier Recently, Meta launched Muse Wonder and Muse Wonder Full, the first frontier-class models from its Superintelligence Labs. The company says the new models reach parity with OpenAI’s GPT-6 Astra and Anthropic’s top Claude variants on multistep reasoning, coding, and long-horizon agentic tasks, priced at $4 per million input tokens and $20 per million output tokens. Wonder is positioned as Meta’s largest and most capable model yet, built specifically for agent workflows such as processing insurance claims or complex lease negotiations. It introduces Active Memory and Active Focus features that let the model carry context across conversations and persist through difficult multi-step work. A companion open-weight release, Muse Spark Open (816 Bn parameters, nearly 42 Bn active), was optimised with Apple to run efficiently on the Mac Studio. The models power the paid tier of Meta’s Muse personal agent and mark the clearest sign yet that Meta intends to compete head-to-head with the closed frontier labs rather than remaining primarily an open-weights player. Early independent benchmarks appear supportive of the parity claims, though full third-party verification is still underway. With Muse already a consumer hit, the Wonder family gives Meta a credible path to monetise advanced agent capabilities at scale while keeping a strong open-source presence. Startup In The Spotlight: Coreshield Founded in 2023 by Chandan Singh Rathore, a former Indian Air Force squadron leader, Bhavya K, an ex-Palo Alto Networks executive, and Rana Amit K N, CoreShield Technologies is building AI-driven security systems to protect critical infrastructure and national assets. India’s growing geopolitical complexity and rising threats to defence, energy and public infrastructure have made real-time security intelligence increasingly important. However, existing defence and civil protection systems often operate in silos, with cameras, sensors and communication networks functioning independently. This fragmentation can delay threat detection and response during emergencies. Operating from IIT Delhi’s Research & Innovation Park, CoreShield develops indigenous platforms that integrate video and acoustic analytics, AI-based decision orchestration and secure data-flow monitoring. Its systems bring together intelligence from multiple sources to help defence and infrastructure agencies detect, analyse and respond to threats in seconds. Designed for both military and civilian applications, the startup aims to improve situational awareness and enable faster, data-driven responses. Its focus on coordinating intelligence across disparate systems differentiates it from solutions that address individual security components in isolation. With India’s defence tech market projected to grow from $7.6 Bn in 2025 to $19 Bn by 2030, rising defence modernisation and AI adoption could create opportunities for indigenous security platforms such as CoreShield. Prompt Of The Week What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work? Here’s Priti Sawant, founder & CEO of JoulesToWatts Business Solutions, with a prompt she uses to turn a messy business situation into a clear decision. “Here is the situation as I see it: . These are the outcomes that matter: . These are the constraints and trade-offs: . My current hypothesis is: . First, reflect the problem back in my logic. Then challenge my assumptions, draw relevant parallels across industries and markets, identify what I may be missing, and recommend the best path forward. Finally, capture the reasoning in a reusable playbook.” Editor’s Note: Some prompts may need to be adjusted by users for best results or may not work as intended for certain users.