Despite hosting a global summit, it is a bystander in the race to develop frontier models
Listen to this story
Your browser does not support the <audio> element.
S MALL SLIGHTS matter when you want to be a big power. Take a recent exchange on a panel at Davos, an elite salon, in which the boss of the IMF appeared to dismiss India as a second-rate AI power. A few seats along Ashwini Vaishnaw, the country’s IT minister, bristled. Then he launched a salvo citing Indian prowess, from chips to apps. In patriotic, tech-mad India, the clip went viral.
There was lots of such talk this week, as world leaders and AI barons gathered in Delhi. India is the first developing country to host the annual summit, following Britain, South Korea and France. What began as an earnest discussion of risks has morphed into something like a giant trade show; “safety” has been replaced with “impact” in its title. The backdrop this time is buzz about the new models released by Open AI and Anthropic, two American companies. Even so, India is using the event to hype itself as an emerging AI superpower, alongside America and China.
In many ways that is fanciful. India has little aptitude in advanced manufacturing and cannot produce the chips on which AI depends. Although a government-backed semiconductor mission has made some progress, success in the coming years would at the most mean India being able to produce some lower-grade chips, while continuing to import whizzier ones. On compute it lags, too, though that is starting to change thanks to a data-centre boom. Like the rest of the world, it is a bystander in the two-horse race to develop frontier models.
Chart: The Economist
Mr Vaishnaw cited a Stanford study that puts India third in the imaginative category of “ AI vibrancy”—behind America and China but ahead of the likes of Britain and Japan. That ranking gives weight to India’s large tech workforce, growing number of startups and digital infrastructure. Yet other measures do not even put it in the top ten. In private, insiders admit that India’s AI ecosystem remains small. A partner at a venture-capital firm says the country has lots of engineers but fewer than 300 skilled AI researchers, far fewer than Britain or France. The government spends a paltry 0.7% of GDP on research and development. Risk capital remains scarce.
The summit aimed to show off India’s abilities as a global convener. Things got off to a bad start, with delegates complaining about long queues and poor Wi-Fi. But India did draw in the industry’s biggest names. On February 19th Narendra Modi, the prime minister, stood on stage with Sam Altman and Dario Amodei, the bosses of Open AI and Anthropic (who declined to link arms). The summit declaration is likely to focus on ensuring affordable access to AI, a priority for developing countries.
Despite its weaknesses, India does have attributes that will give it an important role in shaping AI adoption. For starters it has 900m internet users, who spend on average seven hours a day online. That makes it an unignorable market for AI companies looking to hoover up data on which to train models. Open AI, Anthropic and Google are all competing for Indian custom. On February 16th Anthropic, whose second-biggest market is India, opened an office in Bangalore. And when India played Pakistan a day earlier in the T 20 cricket world cup—South Asia’s nearest equivalent to the Super Bowl, with an audience of 450m—the prime advertising slots were taken by Chat GPT.
Indian companies and consumers are also embracing the technology enthusiastically. One survey found that almost 90% of Indian firms are using AI, versus 62% globally. The country is genuinely world-leading in voice-driven AI, which is how most Indians use the technology. One company that delegates will hear much about is Sarvam AI, which has adapted open-source models by training them on local voice and language data. There had been scepticism about whether it would outperform general-purpose models. But recent benchmark data shows it now does so for certain (albeit niche) tasks, such as reading documents in Indian languages.
At the summit, India’s political and tech leaders will promote this distinctive approach. “We are not trying to burn millions of GPU s building artificial general intelligence,” says Abhishek Singh, the boss of India AI, a government agency. To become the world’s “adoption capital” is India’s goal. “We think it’s not just about LLM s and four companies running the world,” says Rudra Chaudhuri of ORF, a think-tank, which is a “very expensive” approach to building AI. “India’s approach is bottom up. It’s not the model, it’s the use case that you have to build around.”
That leads towards a focus on applications. Several Indian startups have attracted global interest and investment, in areas such as cloud computing and customer service. Others focus on applying AI to problems that are urgent in the developing world. Supernova, which aims to make English tuition affordable to all Indians, many of whom lack access to decent schools, has a million monthly users and is growing at a clip. Telemedicine and AI -triage chatbots are expanding in a country in which health care remains limited.
One question—given the way in which frontier models are galloping along—is whether such indigenous apps will prove most effective in the long run. Another is cost. Many uses will not require expensive “bleeding-edge models”, argues Mr Chaudhuri. A positive spin on limited capital and revenue streams is that Indian companies will be forced to work out how to apply AI frugally. Rajan Anandan of Peak XV, a venture capital firm, thinks the country will be a pioneer of “ultra-affordability” (he predicts a mini “DeepSeek moment” at the summit, with an Indian company set to launch a video-generation model that costs a fifth as much to build as its nearest Chinese competitor).
The biggest challenge for India may be holding on to talent. It has a huge tech workforce and engineers already skilled in the messy plumbing involved in helping companies adopt new technology, notes Shruti Rajagopalan of the Mercatus Centre, a think-tank. But it struggles to offer the salaries demanded by AI stars. And many founders feel compelled to leave in search of funding; it is increasingly common to see companies with engineers in India and management in California. The government is set to announce investment in AI labs and P h D programmes, designed to make staying put more attractive. Perhaps a bit of hype won’t hurt, either. ■
논증 분석
유형: diagnosis
핵심 주장
India는 AI 글로벌 서밋을 개최하며 AI 강국을 자처하지만, 실제로는 프런티어 모델 개발 경쟁의 방관자에 불과하며 ‘채택 수도’로서의 틈새 역할에 만족해야 한다.
논리구조
- 전제: India는 Delhi AI 서밋을 개최하며 미국·중국과 어깨를 나란히 하는 AI 강국으로 자국을 홍보하고 있으며, IT 장관 Ashwini Vaishnaw는 국제 무대에서 인도의 AI 역량을 적극 선전하고 있다.
- 진단: India는 AI에 필수적인 첨단 반도체를 생산하지 못하고, 컴퓨팅 인프라도 뒤처져 있으며, AI 연구인력은 300명 미만으로 Britain·France에도 크게 못 미치고, R&D 지출은 GDP의 0.7%에 불과해 AI 강국 주장은 과장에 가깝다.
- 진단: Stanford 연구의 ‘AI 활력’ 지표에서 India가 3위를 기록했지만, 다른 지표에서는 10위권에도 들지 못하며, 벤처캐피털 자금도 부족하고 AI 생태계 자체가 작다는 것이 업계 내부자들의 솔직한 평가다.
- 논거: 그럼에도 India는 9억 명의 인터넷 이용자라는 거대 시장을 보유하고 있어, OpenAI·Anthropic·Google 등 글로벌 AI 기업들이 데이터 확보와 시장 점유를 위해 경쟁적으로 진출하고 있다.
- 논거: India는 음성 기반 AI에서 세계적 수준이며, Sarvam AI 같은 기업이 현지 언어·음성 데이터로 오픈소스 모델을 파인튜닝해 특정 틈새 과제에서 범용 모델을 능가하는 성과를 보이고 있다.
- 처방: India AI의 Abhishek Singh와 ORF의 Rudra Chaudhuri 등 인도 정·재계 리더들은 거액의 GPU 비용을 들여 AGI를 개발하는 대신, 현실적 사용 사례 중심의 ‘AI 채택 수도’ 전략을 추구해야 한다고 주장한다.
- 논거: 교육 스타트업 Supernova, 원격의료, AI 분류 챗봇 등 개발도상국의 현실적 문제를 해결하는 응용 분야에서 India의 AI 스타트업들이 성장하고 있으며, 자본 제약이 오히려 ‘초저비용(ultra-affordability)’ 혁신을 촉진할 수 있다.
- 반론: 프런티어 모델이 급속도로 발전하는 상황에서 인도산 응용 앱이 장기적으로 경쟁력을 유지할 수 있는지, 그리고 저비용 전략이 지속 가능한지에 대한 의문이 남는다.
- 진단: India의 가장 큰 도전은 인재 유출로, 높은 연봉을 제공하지 못해 AI 스타의 이탈이 이어지고, 많은 창업자들이 자금 조달을 위해 California로 떠나는 ‘엔지니어는 인도에, 경영진은 미국에’ 구조가 고착화되고 있다.
결론
India는 프런티어 AI 개발에서 방관자 신세를 벗어나기 어렵지만, 거대한 시장과 저비용 응용 혁신, 음성 AI 강점을 활용한 ‘AI 채택 수도’ 전략으로 독자적인 역할을 모색할 수 있으며, 이를 위해서는 인재 유출 방지가 핵심 과제다.
Stay on top of our India coverage by signing up to Essential India, our free weekly newsletter.