sb.scorecardresearch

Updated 15:43 IST, January 28th 2025

Can DeepSeek R1 Be a Key to Unlocking India’s AI Revolution?

DeepSeek’s achievement is an example of optimal use of restricted resources and a lesson for India in plain sight, especially with the curbs on AI chips.

Reported by: Shubham Verma
Follow: Google News Icon
  • share
deepseek
DeepSeek surged in popularity to reach the top spot on the App Store in the US earlier this week. | Image: Reuters

DeepSeek, a Chinese startup that has upended the artificial intelligence industry with its eponymous AI assistant after it dethroned Microsoft-backed OpenAI’s ChatGPT as the most downloaded app on Apple’s App Store in the US, the UK, and China, has brought billions of AI investments by big tech players under scrutiny. However, it could also be crucial to the sustainable AI development puzzle.

Based on the company’s V3 model, which was developed with less than $6 million in investment, DeepSeek R1 is claimed to be on par with ChatGPT, the development of which has already cost OpenAI over $100 million. That claim has caused a stock selloff. Nvidia, the biggest manufacturer of chips essential for training foundational models, lost over 17 per cent of its stock price, wiping $593 billion from its market value. DeepSeek’s researchers said they used Nvidia’s less capable H800 chips, while its more expensive counterparts power likes of ChatGPT, Llama 3, and Gemini AI assistants.

DeepSeek’s achievement is an example of optimal use of restricted resources and a lesson for India in plain sight, especially with the upcoming curbs on AI chip exports by the US.

Cost-effectiveness

DeepSeek’s success story can inspire multiple Indian AI startups with constrained budgets to build their technologies around a large language model (LLM). A mere cost of $6 million is what DeepSeek invested to train its LLM, resulting in a cost of just $0.14 per million input tokens, compared to an expenditure of $7.50 for OpenAI. Bootstrapped startups can learn a thing or two from DeepSeek’s spending, but they also have the option to integrate R1 into their products without incurring large bills. This would allow startups to minimise their reliance on more expensive LLMs like OpenAI’s GPT.

Indian startups can also use DeepSeek’s AI assistant and V3  model to run and quickly check if their ideas are commercially viable before they pitch their business to potential investors. R1 will cost significantly less than leading LLMs, resulting in cost efficiency for startups. DeepSeek’s R1 can also be customised for specific functions, supporting several applications in different industries, such as medical care and hospitality.

Even individuals can use DeepSeek R1 to economically build their prototypes or applications that would otherwise cost a bomb with the likes of ChatGPT’s paid tiers.

Circumvention of curbs

Former US President Joe Biden ’s executive order, passed earlier this month, to impose blanket restrictions on the exports of AI chips to other countries raised major concerns about universal tech collaborations and the exchange of technologies between major hubs worldwide. Although the directive particularly targeted the “adversaries” China and Russia, it prohibits countries such as India from buying advanced AI chips from Nvidia and other firms without a license. However, a small quantity of AI chips is allowed, requiring just a government notification instead of a license.

While the future of AI chip trade between the US and India hangs in the balance after Donald Trump assumed the office, the availability of a handful of AI chips may pose challenges to AI startups — unless they do what DeepSeek has done: putting minimum resources to their best use. Relying on open-source AI can offer solutions to navigate the constraints while providing startups the necessary resources to build their AI technology.

This could also increase competition in the AI space, forcing leading players to innovate further and develop cheaper alternatives to their premium technologies. While developers from around the world are rushing to study and analyse DeepSeek’s V3 model, they are seeking to develop similar LLMs that require minimal resources, including Nvidia’s top-tier chips responsible for a big chunk of the operating costs.

“More competition among models is really good for the industry,” said Adam Nathan, the CEO and co-founder of Blaze, a marketing AI tool. “In most mature tech markets, components within the stack tend to get commoditised over time (i.e. processors, chips, batteries, etc.), and the layer at the top where there's the most differentiation for the user tends to win. In other words, you know Apple's name, not the names of most of its suppliers; you know YouTube but you don't know the server companies where all the videos are hosted, etc. And with more large models in the world, the more specialised models with bespoke data and outputs will become relatively more valuable.”

Concerns

Although DeepSeek’s prominence in the AI industry has rattled the legacy firms, which experts believe would drive innovation and help AI products become more mainstream, they have expressed concerns about DeepSeek’s Chinese affiliation and how that could raise a question on user privacy. Amid increased geopolitical tensions between India and China, regulators are analysing the impact of a technology such as DeepSeek being considered an undisputed alternative to more expensive AI technologies. In the past, India has banned apps, such as TikTok, from operating in the country due to national security concerns, so whether DeepSeek will find a free way into India’s AI startup ecosystem is a big question.

Published 15:39 IST, January 28th 2025