India’s GenAI Boom — What Are VCs Looking For?

With generative AI (GenAI) making waves across the world since the end of 2022, epitomised by the rise of OpenAI, it has given new wings to the early stage Indian tech ecosystem, growth startups as well as enterprise giants. 

From open-source large language models (LLMs) for horizontal applications to specialised models tailored for specific industries and domains, we are witnessing a proliferation of startups in GenAI, with well over 70 GenAI startups actively propelling this growth in India.

Of course, this has also drawn interest from the venture capital (VC) ecosystem. While Inc42’s interactions with several leading VCs revealed that investors are still cautious about investing amid the boom, many are exploring ways to understand which GenAI models offer the best potential outcomes.

For instance, there is a slightly more conservative approach to investing in startups building the foundational models because of the capital-intensive nature of this space. Most VCs are therefore looking into startups building GenAI-based applications and models for specific sectors such as healthcare, fintech, and enterprise tech. 

Besides this, startups in the middle enablement layers for GenAI are also gaining attention, particularly those enabling LLM ops and on-premise deployment. It is pertinent to note that as per Inc42’s analysis, the GenAI startups in the country raised over $580 Mn between 2019 and 2023, with around $50 Mn going to seed-stage startups.

This indicates that Indian startups are still exploring the right engineering models, and not many have achieved the product-market fit, which means the US-based tech giants continue to garner the most attention. 

For instance, Dev Khare, partner at Lightspeed, told Inc42 that India has lagged in catching up with global GenAI investment and development trends.

“If you think about investments in GenAI, I would say India has seen a one or two-year lag to what’s been going on elsewhere in the world. The amount of capital going into and companies being formed in the US and Europe is just a whole different level in comparison to India,” said Khare, adding that the VC firm is now ensuring that it invests early in some of the best GenAI startups of tomorrow.

In a recent research report, brokerage JM Financial said that despite the interest and media excitement, the number of companies that have incorporated GenAI tools remains a small subset of the ecosystem. 

“Investors have so far experienced dismal results in the last two disruptive technologies (cryptocurrency and metaverse) though we do not believe the story is yet over for these technologies. We do, however, believe that investor patience is warranted and enough due diligence should be conducted as the hysteria might result in several pretenders emerging,” the brokerage added.

GenAI Application Layer Has The Largest VC Interest

In India, GenAI adoption is expected to be rapid among businesses across various applications, including content generation, generative coding and more. These are essentially applications that leverage existing LLMs such as ChatGPT, Gemini, and Llama, among others. 

Speaking to Inc42, Arjun Rao, partner at Speciale Invest, said that the application layer will see the highest number of startups, becoming an attractive area for VCs to invest in.

“We are very interested and exploring more in the application layer, which uses GenAI and other machine learning technologies for various use cases, helping to enhance productivity and efficiency of various functional units in businesses. The gamut of application layer AI companies that we are looking at seems fairly competitive. Therefore, we will invest in some of them if not a whole lot of them,” Rao added

Echoing a similar sentiment, Ivy Chin, partner at Inflection Point Ventures said that the VC firm is more keen on finding the best applications of GenAI rather than the startups building foundational models itself.

“At one point in time, it was a question of who would develop AI-based technology. Right now, in that sense, that is cornered by the large names that are out there or the would-be newcomers who could secure large funding up front,” Chin explained. 

She added that more funding opportunities will be in the middle layer or the top layer that builds on the models further in specific areas of expertise to then deliver the outcome or benefits of GenAI. 

Bengaluru-based Inflection Point Ventures has so far invested in GenAI startups such as NeuroPixel.AI, while Speciale Invest’s GenAI investments include Protecto AI and Tune AI.

What About LLMs?

While the applications layer takes most of the attention, we also need to ask what will happen to companies creating foundational layers in India. The likes of Krutrim and Sarvam are building Indian LLMs after raising big rounds to start with, but will there be a proliferation of such startups in India? 

Investor sentiment is dictated by two aspects — the first being the cost involved in building the foundational models and the second in the probability of the number of startups entering the space.

IPV’s Chin believes the likes of Google, Meta, and OpenAI have already made their positions strong in building the foundational models in GenAI and the business globally would utilise those models for further value addition in specific industries. Hence, this is a space that would anyway see fewer companies entering.

However, if we look at it from an Indian context, the two major names building the open models include Ola Krutrim and Sarvam AI. While Krutrim is led by seasoned entrepreneur Bhavish Aggarwal and it managed to secure $50 Mn in a single investment round, Sarvam AI raised $41 Mn in its Series A funding led by Lightspeed.

In the case of open-source LLMs, there are massive costs involved in graphics processing unit (GPUs) usage. Besides, there are additional costs related to managing the infrastructure, optimisation, and scaling, which require specialised and expensive resources.

However, for most early-stage VCs, such a cost-intensive sector is not the ideal choice for investment.

For instance, Vatsal Kanakiya, CTO, 100X.VC said it’s tough for early stage VCs to invest in foundation models because these companies need a lot of capital to buy GPUs to gather data and train the model. It is very tough for a $100 Mn fund to sign $20 Mn or even $10 Mn as a seed cheque in one of these startups, he added.

“Those businesses generally are open to funds who can write the large cheques unless there is some innovation, which allows the companies to do the same thing but significantly at lower costs.”

Echoing this, Speciale Invest’s Rao added that the startups building foundational LLMs require tens of millions of dollars, at least at today’s cost structure.

With Speciale Invest’s ticket size being about $1 Mn, big size rounds don’t make sense for the VC firm and changing this is an existential change to the thesis of the firm. “That is the challenge. Now, if the right company comes along and there is amazing innovation in it and somebody says we can start with a small seed round, that fits our mandate,” Rao said.

The sentiment is similar on a global level.

In a recent research report, brokerage Morgan Stanley said that reducing training and inference costs through software and hardware innovation can be a key differentiator for the foundational model companies as they look to compete on price.

As a result, a new trend is emerging – the developers of small language models (SLMs). To that end, some developers are chaining together SLMs to drive costs lower without sacrificing performance. Lightspeed-backed French startup Mistral AI is one such example.

Rao said that this is slowly becoming an interesting area for VCs to look into.

The Emergence Of Startups Building SLMs & Vertical GenAI

As per Rao, the emergence of startups developing small language models (SLMs) or tiny language models (TLMs) could change the game in India. But this is still at a nascent stage. “We are just starting to see in the last three or four months companies emerging in this area and we are positive that this momentum will continue,” he said, adding that Speciale Invest is eyeing an investment in at least one such startup in 2024.

Given the massive use cases these models can find in several domains and verticals, it would witness significant growth in the coming days.

In fact, Inc42 reported earlier that a rise in enterprise adoption of GenAI in 2024 would create a new generation of entrepreneurs who can address vertical problems in India, rather than the global horizontal approach.

Most investors are expecting a breakthrough in GenAI usage in enterprise SaaS and fintech, along with a few other consumer service domains. While there are still some inhibitions around the extent to which it can be leveraged by the healthtech sector, many investors are seeing a massive opportunity for GenAI models developed specifically for the healthcare segment.

Meanwhile, some investors are also approaching it from a slightly different angle.

For instance, Inflection Point Ventures’ Chin said that industry-specific and enterprise-specific models could drive interest in the next few months. At the moment, the VC firm is more keenly looking at horizontal models that can cut across industries and domains. 

Meanwhile, Lightspeed’s Khare said that when it comes to text generation, it is seeing more use cases in legal, marketing, advertising, and anything around content production. The utilisation of GenAI in audio and video is still in its nascent stages, but the VC firm is closely monitoring the development of emerging models within this space.

The Enablers Of GenAI Applications

To some investors, the startups enabling the thriving of the GenAI applications and the building of models is another interesting segment to look into. Lightspeed-backed Portkey.AI is one such example, while GigaML is another middle-layer startup that has gained funding at an early stage.

“There’s another layer in the middle that is seeing a lot more companies emerging, this can be called the infrastructure layer… there are security, gateways, model training companies,” Khare said. 

Portkey.AI is an AI gateway that enables integrations with LLMs such as OpenAI, Anthropic, and others through a unified API, hence simplifying interactions and minimising integration hassles.

Khare said that more new investments are expected in this segment in India and globally. But most importantly, even the India-based startup will have to compete with global players in this space, he added.

Speciale Invest is also exploring more investment opportunities in this area. Rao explained that large to medium companies are building an AI strategy to understand their use cases to make various aspects of businesses more efficient. 

“They will all have to build either their own applications or we will have to have partners through which they will adopt applications. But for all of these to thrive, there is a need for enablement, tooling, and infrastructure that help these companies use the fundamental AI tech.”

For instance, Protecto.ai, backed by Speciale Invest, has built a data security platform, safeguarding data and making it secure in an enterprise setting before that data is used for any AI model training.

While certain challenges remain to all of these theses that the investors have built for GenAI investments, and with time, they will evolve, it goes without saying that India is looking at a massive adoption of this new technology across areas and investors are all set to take opportunities of this boom.

[Edited by Shishir Parasher]