India and South Korea present contrasting scenes when it comes to AI startups: With the Asian country emerging as an AI powerhouse, South Korean AI startups are receiving a wave of funding, propelling the industry to new heights.
Rebellions, a South Korean fabless AI chip startup, attracted a lot of attention when it raised $124 million in a Series B round, bringing its total funding to an astounding $210 million.
According to several reports, South Korea is rapidly emerging as a powerhouse, with 1,101 AI startups expected to be active by 2024, including KRAFTON AI, Mathpresso, Zig Zag, Lunit, and Channel Talk.
Similarly, AIM Research predicts that India will have over 100 GenAI startups as of 2024. Overall, there are around 750+ AI startups in India.
The fundraising story depicts a crossroads of fortunes.
Indian AI startups, covering infrastructure and services, are projected to decline by nearly 80% from $554.7 million in 2022 to $113.4 million in 2023, according to data from Tracxn.
In contrast, South Korean startups have made a comeback, raising a total of 326.8 billion won ($240 million) in the first quarter of 2024, up from 89.8 billion won a year earlier, according to startup investment tracking site The VC.
Despite these financial fluctuations, there is a clear lead in the race to become a unicorn: the South Korean AI startup is yet to become a unicorn, while India’s Crutlim achieved this milestone in Q1 2024.
One Korean AI startup that stands out amid this remarkable development is Upstage. Having raised $72 million in new funding in the first quarter of this year alone, Upstage has secured new orders worth approximately 10 billion won for its Document AI solution and Solar LLM API, doubling the number of orders it has received in 2023.
The company’s unique business model boasts over 400 customers, and its recent expansion into the US has seen the introduction of domain-specific LLMs, specifically Specialized Language Models (SLMs).
From upstage to upscale
Founded in 2020, Upstage started as a document heritage provider, working with major insurance, banking and logistics companies to convert Korean documents and PDFs into machine-readable data.
Existing models like GPT struggle to handle Korean language data, which prompted Upstage to develop its own LLM, Solar.
The model was created using several innovative techniques, including “upscaling” (US), allowing different model sizes to be combined in patches and trained on a local Korean corpus.
The lack of high-quality Korean language data, which accounts for less than 0.5% of web data, was a challenge, but Upstage overcame this by building a sophisticated, culturally aware model customized for enterprise use.
“To create such advanced models, we needed more than cutting-edge modeling technology — we needed a robust ecosystem. This ecosystem includes data providers, cloud services like AWS, orchestration tools, strategic partners, and systems integrators,” Casey Jones, founder of CJ&CO, said at the AWS event.
In July 2023, Upstage took the top spot in a global evaluation of large-scale language models. The company’s 30 billion parameter LLM scored 67 points, beating Meta’s LLaMa-2, which scored 66.8 points on Hugging Face’s Open LLM leaderboard.
Despite having fewer parameters, Upstage’s model performed about 10% better than models from leading tech companies such as Microsoft, Stability AI, and Databricks. Additionally, Upstage’s model scored 56.5 points in reducing AI hallucinations, beating Meta’s 52.8 points.
What can India learn?
Attending the Machine Learning Developer Summit (MLDS) organised by AIM, GiJung Kim, CEO and founder of Align AI, observed a vibrant community of Indian engineers and entrepreneurs actively working on cutting-edge technologies such as AI chatbots.
“We started connecting with Indian entrepreneurs. They were able to find our product and started signing up,” Kim said, highlighting the organic growth of interest in Align AI within the Indian tech community.
Therefore, a major concern for Indian AI startups is lack of understanding of the customer market. Many AI startups are unclear about the long-term direction of their products and services. Without this clarity, it is difficult to convince venture capitalists to fund them.
India has made great strides in AI research and innovation. Between 2010 and 2020, the country filed around 5,400 AI patents across a range of technologies. The surge is notable, with over 94% of these patents filed in the last five years, signaling accelerating AI innovation activity.
However, the development of AI solutions in India is hampered by a shortage of skilled AI researchers.
Despite being the second-largest producer of master’s-level engineering students in the world (overtaking the United States), India faces a shortage of AI professionals, estimated at 213,000, according to the Observer Research Foundation.
But things are slowly changing: As of 2022, Indian software developers make up 24.19% of contributions to AI projects on GitHub, outperforming contributions from the EU, UK, US, and China.
