“We are seeing a lot more application-layer pitches from founders. Our investments will likely be weighed more heavily there,” said Hemant Mohapatra, a partner at global VC firm Lightspeed Venture Partners. “We’re very active in the market.”
It’s a sentiment that’s echoed by the largest VC firms in the Indian market including Lightspeed, Accel India and Peak XV Partners (formerly Sequoia Capital India and Southeast Asia).
Lightspeed said its cheque sizes for AI startups would vary, depending on a company’s technology. Its investments have ranged from $3.5 million with a recent Pre-Series A round in Stimuler, a voice-first AI tutor for English-as-a-second language, to leading foundation model company Anthropic’s $3.5 billion Series-E round earlier this year.
“We look for extraordinary founders who are going after large markets. We’re just seeing more AI companies,” said Rajan Anandan, managing director at Peak XV. The firm said it doesn’t reveal the size of investments it makes in startups.
“It’s not an increase as a strategy but more because for every one foundational company there are 10,000 application-layer companies,” said Prayank Swaroop, a partner at Accel, a global VC firm. “By the law of large numbers, application-layer investments will be a lot more not just for us, but for other Indian VCs, even for VCs all over the world.”
A part of the reason for the shift towards the application layer is the significant improvement in middleware, the software that connects operating systems to applications, data and users.
It’s the ‘glue’ that links various moving parts, allowing them to function seamlessly. As a result, AI startups aren’t building horizontal platforms anymore. They’re focusing on vertical solutions. Many AI startups from India are building vertical AI agents – targeting the US as well as Indian markets.
“Startups are building for banks, insurance, retail, healthcare, construction, even real estate. Vertical AI is a mega theme, followed by building for the tooling and infrastructure layers,” said Anandan.
According to Accel, a global VC firm, three trends have emerged over the past few months in India: The shift to agentic applications, and the ease and convenience of building AI applications have contributed to boost in the ecosystem. Companies such as Cursor, Windsurf and other ‘vibe-coding’ platforms have made the barrier to build applications very low.
Due to these preceding trends, applications with better solutions are being built in India’s AI ecosystem.
“Founders will be building software for testing of applications, data management and cybersecurity,” said Swaroop. “When people make so much software, they realise there are problems and start solving for them as well.”
Startup founders have noticed the improvements in infrastructure as well, given that algorithms have advanced significantly over the past 18-24 months.
“The quality of applications, the complexity which we’re able to build for enterprise applications is superior and at the same time, at much lower cost than what was available two years back,” said Saurabh Mishra, co-founder and chief executive officer of OrbitShift.
Founded in 2022, OrbitShift was part of the most recent cohort of Surge, Peak XV’s seed platform for startup founders, and raised $7 million in a Series-A round with participation from Peak XV and Stellaris Venture Partners in 2024.
CARPL.ai’s founder Vidur Mahajan said the ease of building an AI application and the amount of accessible data have played a role.
“Given that reality, naturally there is a huge draw towards building application layer products,” Mahajan said.
The startup is building an AI-based image analysis platform for hospitals and imaging centres. CARPL.ai raised $6 million in seed funding last year in a round led by Stellaris. Novo Holdings, LeapFrog Investments, Bain & Co, Boston Consulting Group and UnitedHealth Group were also part of the round.
Regardless of the ease of building applications, it’s still a very competitive market. The cost of starting a company may fall, but customer value proposition will be key, with larger cheques being cut only when that happens.
“I think it’ll be small raises up to $3 million in the beginning and once customer value is proven, a very large next round between $10 to $15 million,” said Swaroop. “For proven teams, the initial round can be larger, from $8 to $15 million.”
Founder behaviour
Given the shift in technology, there’s also been a change in founder behaviour, including many more pitches for the AI application layer and a clarity of vision that wasn’t present last year.
“Founders are able to think more clearly about how they’re trying to differentiate themselves in the market,” said Swaroop of Accel.
What’s more, traditional software-as-a-service business models are being viewed as unsustainable. Founders realise they must build AI solutions to stay competitive as customers expect complex applications that can offer more.
“There’s more clarity in Indian founders’ minds about why they’re different. They are a lot clearer on what their role is and what their company’s role is in the global landscape,” said Lightspeed’s Mohapatra. “Founders have more clarity in going after the application layer and they are able to articulate those advantages to us.”
As a result, the firm has seen an uptick in founders recognising new use-cases for AI and building to solve specific problem statements.
Peak XV clocked the trend earlier in its latest cohort of Surge, with eight of the 14 companies in AI – six developing AI applications and two building for AI infrastructure and tooling.
“We are seeing more companies building across the AI stack and you can expect to see that in our next cohort of Surge and the companies our venture team is partnering with,” Anandan said, adding that there’s also a boost in the infrastructure and tooling layers.
Companies such as Auquan and Wobot were part of the most recent Surge cohort and are building vertical AI agents in industries ranging from information technology-enabled services and banking to food services and retail.
The India advantage
While the US and China are widely considered to be ahead when it comes to building foundation models in AI, India’s advantage lies in data – something that neither of the two nations has.
“Data is going to be critical – that’s where India’s moat is in terms of both specific data like healthcare information and general data like enterprise workflows,” Mohapatra said.
The advantage that startups in India have is when they launch a product, the sheer number of potential users in the country gives them a lot of data to iterate their product on. Not just that, it gives them a chance to collect feedback and go to the market quicker.
Indian foundation model companies already have access to huge repositories of data or, in some cases, are building it themselves. Attentive.ai, which uses AI to map out construction sites and help with facilities maintenance, has been building its own AI models because there are no datasets for their product. Peak XV was part of the company’s $12 million Series-A funding round led by Tenacity Ventures.
Netradyne, the first unicorn – a startup valued at $1 billion or more – of 2025, built all its datasets related to driver and road safety from scratch by putting its devices on vehicles in the US and in India when it launched.
“The best vertical AI companies will have access to proprietary data and one of the things that Indian startups can do – because of our cost of talent – is we can collect a lot more data and train domain specific models. It’s a huge, huge moat,” said Peak XV’s Anandan.
What’s more, given India’s long history as a business process outsourcing destination, there is a lot of experience, knowledge networks and demand for application layer products, according to Mohapatra.
“Now is our moment. It is where the world is headed, it’s where the capital is going and it’s what we’re already really good at. Frankly, it is our game to lose,” he said.
But where does India stand when compared with AI giants in the US and China?
“We’re ahead of China on the application layer. They don’t have a global supply chain of applications,” said Mohapatra, adding that nobody really uses a Chinese frontier model in an enterprise setting in India or the US. That is corroborated by the fact that soon after the release of DeepSeek earlier this year, several countries banned its use on government devices and for official work.
But given the pace of innovation, improvements and ease of building applications in AI, will older startups get pushed out by newer companies with better technology? It’s not that simple.
“It requires lot more domain knowledge, spending time with the customers to understand their business, create something which meets their requirements, so they don’t feel the friction,” said OrbitShift’s Mishra.
