If you thought it’s raining money in generative AI, you might be in for disappointment. 

Despite the hype around generative AI funding, reports show that startup funding remains bleak, especially for newcomers who have no moat. In June, Venture Intelligence reported that under ChatGPT’s shadow, funding received by domestic AI startups witnessed a significant decline this year, totalling $510 million from January to May. This amount is merely half of the five-month average of $1.02 billion recorded in the previous year. 

Securing funding from investors for generative AI startups is not a cake walk anymore. Investors have realised that these companies need to find their niche before starting out. Convincing VCs with an idea isn’t the recipe to snag their money. Nowadays, most of the generative AI startups are just trying to cash the generative AI wave to make some money without adding any substantial application for their customers. 

There are going to be selective applications from startups that will work well and can build businesses, but maybe 70-80% of them are going to die in the end. Recently, ZuAI, a generative AI Edtech startup raised $484K (Rs 4 crore) in a seed round led by the early stage investor Prime Venture Partners. However, there is one problem that ZuAI’s GPT-centric platform cannot solve. It cannot generate content for all the subjects yet as mathematical solutions are not the strong suit of AI models currently. 

If not in an idea, VCs tend to invest in founders of AI startups that are from premier institutions like Stanford, Harvard and others, alongside having prior experience at big techs. Based in Paris Mistral AI, which raised $133 million in seed funding, is co-founded by alums from Google’s DeepMind and Meta to compete against OpenAI in the building, training and application of large language models and generative AI. The case is the same with Inflection.AI by Mustafa Suleyman, who has worked with DeepMind before. Anthropic AI and former OpenAI employee Dario Amodei have the same story.

So, what can be the moat then?

Focus on Data 

One of the most widely discussed AI moats in venture capital circles is the concept of proprietary data. The reason behind its prominence is evident: data serves as the lifeblood of AI and acts as the crucial driver for foundation models. Despite the substantial funding pouring into generative AI startups, the risk of failure looms if they are unable to access the appropriate data—a challenging task in itself. A lot of AI startups fail because they don’t have the essential data to train their model. 

If the companies are pursuing a brilliant application of AI, but don’t have the access to data that will give them the ability to build a powerful application, the future would be bleak. OpenAI recently partnered with AP and will have access to AP news stories going back to 1985. It is going to be tough for AI startups to challenge Microsoft-backed OpenAI.  

Lately, companies are looking at creating specialised datasets from ChatGPT (based on GPT-4), and others to further train open-source, smaller models like LLaMA, to fit into the required use cases and usability. Focusing on building quality datasets for training your foundational models can be rewarding. 

UI/UX is all that matters 

Jasper AI, which is based on GPT 4, is trying to differentiate itself from ChatGPT. They even posted a blog saying ‘Jasper vs ChatGPT: Why Jasper Wins’. The question here is, is it really winning. Why would a company buy a Jasper subscription when they can directly use ChatGPT or build their own model based on their API. 

Amidst the hype around ChatGPT, we are increasingly seeing startups that are either using GPT in their name, or building technologies using the APIs provided by OpenAI. If you are building on other’s APIs what would be your moat? Your UI/UX should be so good that the user doesn’t want to use the original model. The importance of creating a unique experience for users solely based on UI/UX shouldn’t be discarded. UI/UX can help you turn the tide. A Twitter discussion (below) about how UI/UX can or cannot be a moat has been heating up. 

Know Your Customers 

Creating generative AI solutions or products for B2B customers is a tough ask. Executives at large companies have shown great interest in AI since the beginning, leading many startup founders and venture capitalists to believe that these companies would be ideal early customers. However, the startups building AI solutions for these companies underestimated the executives’ and engineers’ ability to quickly implement AI using open-source tools. 

An engineering leader would rather spin up their own infrastructure for free and build tech themselves than buy something from a new startup. Zoho, a prominent Indian software-as-a-service (SaaS) company, has embraced the challenge of developing its own large language model (LLM) akin to OpenAI’s GPT and Google’s PaLM 2 models. The project is under the supervision of Zoho’s founder and CEO, Sridhar Vembu.

Similarly, TCS is attempting to build its own Github Copilot alternative, which is touted to be used for enterprise code generation.  In April, Tech Mahindra became the first IT giant to launch a Generative AI Studio.

Know Your Investors 

If we look at the popular AI startups like Inflection, Cohere and Anthropic, most of them are backed by Microsoft and Google. It takes ton of money and exceptional minds to build a successful AI company. And to build a successful generative AI model, you need huge amounts of data, infrastructure and money, which VCs are not going to shell out easily. 

Once you have found an idea, and worked on the MVP, it is wise to join the accelerator or incubation program offered by large tech giants. Following are the programs offered by them — Google for Startups Accelerator, Microsoft for Startups. Also recently, Neo announced  a new AI track for Neo Accelerator with help from Microsoft and OpenAI to empower startup founders with the best of generative AI. 

In Conclusion

If we look at startups that are successful in generative AI, they are finding their own niche application and not copying OpenAI or using its API, examples being Midjourney and Stability AI. We need to see from the customer perspective and understand business use cases. Generative AI isn’t valuable by itself. We are just seeing the shedding of the companies that didn’t offer officiant business value and differentiation to keep their foothold.