Why Are Investors Betting on AI Skyscrapers, Not Foundational Models?

As the adoption of GenAI grows in the country, varied applications, draped as horizontal and vertical solutions, are increasingly finding their use cases. Whether conversational tools, speech-to-text/speech-to-speech bots, text summarisation, or video generation, Indian startups seem to be taking charge of paving the future of this ingenious tech in the country

In 2024 alone, several new-age tech startups, including the likes of Highperformr.AI, Ayna, Gnani.ai, Clodura.AI and Vitra.ai, raised funding from investors such as Inflexor Ventures, Info Edge, Venture Highway, and Bharat Innovation Fund, just to name a few. 

However, worth noting is the fact that 91% of the money invested in Indian native GenAI startups to date has gone to companies that are building business or consumer applications around GenAI technology, according to Inc42’s latest report — The Rise Of India’s GenAI Brigade. 

Consequently, foundational solutions like LLMs and cloud infrastructure are seemingly less of a priority for the VC ecosystem. Substantiating this, a recent survey by Inc42 of over 50 VCs has revealed that 62% of the VCs prioritised startups building tools and applications on existing LLMs. Meanwhile, merely 11% see scope for startups in developing domestic open-source foundation models (LLM) and 19% in closed-source foundation models.

This could be because building foundational models is expensive due to the high cost of graphics processing units (GPUs) and other expenses related to managing the infrastructure.  

In fact, earlier this year, many investors highlighted the capex-heavy nature of building a foundational model, making vertical applications a more attractive investment choice.

For instance, Speciale Invest and 100X.VC, two of the leading GenAI investors in the country, stated that their relatively small fund and ticket sizes prevent them from making large investments in startups developing foundational models, which require billions of dollars in funding.

Access Free Report

But Is Money The Only Concern?

According to the VP of growth and marketing at Inflection Point Ventures, Sahil Chopra, along with the capital-intensive nature of foundational GenAI models, there are several other reasons why VCs choose investing in the application layer. 

“Applications frequently meet immediate market demands, resulting in quicker ROIs and fewer technical development hurdles. They improve operational efficiency by integrating with current corporate processes and scaling more readily,” Chopra said.

In addition, he said that the startups building GenAI applications have a greater emphasis on user experience and the possibility of strategic alliances, which also increases the appeal of application-layer investments. 

“It also enables entrepreneurs to demonstrate the viability of their GenAI solutions rapidly and more efficiently, he said.

Murali Krishna Gunturu, principal at Inflexor Ventures, too, believes that foundational models face monetisation challenges. According to him, in the presence of platforms like ChatGPT, Gemini, and Claude bringing out a significant differentiating factor that users would opt for is challenging.

Offering a slightly different perspective, Sonal Saldanha, VP at 3one4 Capital, pointed out that model development faces the challenge of rapid depreciation. She explained that the market is very competitive, and margins are extremely fine, so it is tough to assume one will make money back on the cost of R&D without sustained investment and long-term strategy.

According to Saldanha, like the earlier waves of the Internet and mobile, “infrastructure investments” will be fewer than “application investments.” 

“In this context, applications also have better margin profiles and can be more economical to develop, distribute and establish. Overall the surface area for applications is very large, and this is now turning into the surface area for services as well, which expands the addressable market meaningfully.”

Access Free Report

What Lies Ahead?

As of now, while GenAI infrastructure tools such as GPUs, LLMs, and model fine-tuning capabilities are crucial for practical AI applications, less than 5% of funded Indian startups are focussed on developing these foundational tools.

Meanwhile, besides building the foundational model and working with the application layer, there are a few other niche categories that are also getting increased investor focus. 

Interestingly, Inc42’s survey also suggests that 51% of the respondent VCs want Indian entrepreneurs to build products for model fine-tuning. On the other hand, 46% believe there is scope for products in Retrieval Augmented Generation (RAG).

“We are seeing domestic companies do interesting R&D work through distillation and fine-tuning, building tooling for ops and observability, building proprietary models for specific applications,” Saldanha said.  

He gave the example of one of its portfolio startups Smallest.ai, which has built a text-to-speech model that generates 10 seconds of audio in 100 milliseconds and uses only 1GB of RAM.

In fact, the small language models (SLMs) are increasingly finding more traction in the market, given they work with smaller datasets and can perform and excel in specific tasks while also keeping the costs lower. 

Recently, Sarvam AI also doubled down on SLMs with the launch of its full-stack GenAI platform. Hemant Mohapatra, partner at Lightspeed, said the VC firm’s rationale behind investing in Sarvam AI stems from its belief that AI in India must be uniquely tailored to the scale and diversity of its population.  

He added that Lightspeed remains open to evaluating other companies building foundational models from any region and investing in the best teams. 

Overall, the investors’ sentiment is clear — while foundational model startups will continue to receive funding, the pace is expected to remain slow. Existing players will remain relevant only if they can consistently innovate and outperform competitors in global markets.

Further, niche areas like SLMs and model fine-tuning are emerging as promising opportunities, which suggests that innovation tailored to India’s unique ecosystem will be key to shaping the next phase of GenAI’s evolution in the country.

[Edited By Shishir Parasher]

Access Free Report