For India, AI presents both opportunities and challenges in global race

DeepSeek has raised hopes among countries like India that it is possible to be a key player in global AI race through innovation. But is India on time for the party?

This report has been updated  It was a house divided at the Paris AI Action summit last week. Prime Minister Narendra Modi, co-chair of the summit (French President Emmanuel Macron was the chairman), stressed the need for global pooling of resources and talent to build open source artificial intelligence. He also called for establishing governance standards for such work.

 

That appears tough. The United States (US) and the United Kingdom (UK) did not endorse the joint communique backed by 60 signatories at the summit. US Vice-President J D Vance took a strident stand, saying that his country is winning the race to build AI and would keep it that way. He took potshots at China (he didn’t name the country) by saying that authoritarian regimes are using stolen AI tools and warned allies, especially the European Union, not to tighten regulations on US technology companies.

 

The summit took place amid what is turning out to be an AI war between the US and China. DeepSeek, which is funded by a Chinese hedge fund founder, disrupted the tech world weeks ago by launching a generative AI (GenAI) model that is as good as ChatGPT of America’s OpenAI but at a fraction of the cost and computing power.

 

DeepSeek has raised hopes among countries like India that it is possible to be a key player in the global AI race through innovation that doesn’t require billions of dollars. But is India on time for the party?

 

“The AI race has just begun, so it is too early to predict the winners or who is falling behind. India has the same potential as other geographies like the US or China and incumbency does not matter in this stage. And India has a better vantage point to start with,” said Krishna Rangasayee, founder and chief executive officer (CEO) of Sima.ai, which is based in San Jose, California.

 

Mission AI

 

The cornerstone of India’s strategy is the government’s ₹10,000 crore AI Mission which is building shared compute power, undertaking skill development, financing AI startups, and arranging access to non-personal datasets to develop AI models. Its eventual goal is to build a sovereign AI, a nation’s capabilities to produce artificial intelligence using its own resources.

 

Shared compute is key to train AI models that require extensive processing power. The Ministry of Electronics and Information Technology (Meity) has finalised empaneling private companies to offer Cloud infrastructure and it has collectively offered over 18,000 graphics processing units (GPU, a high-end chip). IT Minister Ashwini Vaishnaw has said that with subsidy support from the mission, it will be possible to offer compute power at $1 per hour on GPUs to users – one of the cheapest in the world (the average is over $2 per hour).

 

Companies say the plan will help them. Akhilesh Sabharwal, founder and CEO of CoRover.ai, said: “We are building a foundational model from scratch which would cost us around ₹4 crore. Now, with the Meity offer our cost of setting it up will fall by a fourth.” CoRover runs Bharat GPT which offers domain-based AI services to clients.

 

Experts said Meity’s GPU numbers are good enough for supporting large language models, or LLMs.

 

Research by Oxford Internet Institute in collaboration with others concluded that there are only 15 countries (leaving out the US and China) that have some quantity of compute relevant to “AI development” (like chipmaker Nvidia’s A100 and H100 GPUs). They are mostly developed countries, like the UK, Australia, France, Japan and Singapore. India is the only developing nation in the list.

 

Another key requirement in the AI race is skill. India is at the top globally in skill quotient, according to ‘Stanford AI Index 2024’. The country’s AI workforce is 2.8 times more skilled in AI competencies than the global average. 

 

India has AI talent

 

To put numbers into perspective the US has more senior AI talent – the key to leading the new technology – than India, the UK, Germany and Canada put together. The US has 330,000 such professionals and India 111, 000, the second largest source, according to research agency Savanah.

 

But can Indian companies harness talent to build AI models trained on domestic datasets? Vaishnaw has said there are already six developers which can come out with models in six to eight months.

 

Indian companies are already on the global AI map. Business analytics firm CB Insights has named Bengaluru-based sarvam.ai as one of the 100 most promising startups in 2024 in the world. Sarvam has raised $53.6 million in two rounds as it develops LLMs.

 

Yet funding is a challenge. Indian AI startups in 2021 and 2022 collectively raised $998 million, but in the next two years ending 2024 funding fell sharply to $261 million, according to Traxcn. Only 40 per cent of the more than 240 plus GenAI startups in India were funded in the first half of 2024 (H1 2024), according to Nasscom. 

 

As a result, average funding size has crawled, growing from $5.3 million in H1 2023 to $5.5 million in H1 2024. Private equity investments in Indian GenAI startups have since 2023 accounted for only 2 per cent of such funding globally.

 

There are two other technology challenges. As more powerful GPUs chips come into the market every 12 to 24 months, Cloud companies will have to invest continuously.  Second, with the US limiting the import of advanced processors (50,000 GPUs a year), India will need indigenous alternatives.

 

Vaishnaw has said the government has held talks with stakeholders in India to build an indigenous GPU and held meetings with Nvidia and AMD (both are American companies) for support. “It will take time but it can happen only with government and private sector partnership,” said Ashok Chandak, president of India Electronics and Semiconductor Association.

 

When a ‘Made in India’ GPU arrives, how will it be monetised? OpenAI earned just $3.4 billion as revenue in 2024 despite the international popularity of its ChatGPT model as it looks towards moving from a non-profit to a profit model and has roped in Microsoft as a key investor. 

 

About AI viability, the senior executive of a global technology company said: “The open source model makes sense for Google or Amazon or Microsoft, which already have an existing underlying business model. Building an independent standalone model is not easy.”

 

One option for monetisation is to offer ‘AI on the edge’ where compute power resides near a customer and not in the Cloud. “We see it as the next big thing: It needs less power, provides more security and safety; it is cheaper and companies providing the service can easily monetise,” said Rangasayee, of Sima.ai, which provides the service to 20 Indian companies.

 

The Stanford Global Vibrancy Index, which measures a country' AI ecosystem based on research papers, private investment and patents, ranked India fourth in the world: After the US, China and UK. That’s an indication that India is a key player in the global AI race. 

What's what  

• A foundation model is a machine learning model trained on datasets that can be used for a range of uses
• Generative AI refers to deep-learning models that can put up text, images and other content based on the data they were trained on
• Large language models (LLM) are AI systems capable of understanding and generating human language by processing vast amounts of text data
• General purpose LLM is trained on text from various domains; domain specific LLM is trained on data from a specific domain like legal, financial
• Open source AI: Source code is public and anyone can modify and distribute it. Closed source AI: Code is confidential for better security