印度战略：为什么印度需要自己的DeepSeek

India Strategy: Why India needs its own DeepSeek India can’t build its AI future on borrowed models. While renting compute to power foreignLLMs may look like progress, it leaves the country dependent. The recent US restrictions onaccess to the latest AI model for non-citizens drive home the risk that India could find itselflocked out some day, with its applications trailing those in the US and China. We had raisedthis topic even earlier but were muffled with noises of “LLM does not matter”. But the recentdevelopment compels us to have a relook. Venugopal Garre+65 6326 7643venugopal.garre@bernsteinsg.com Nikhil Arela+91 226 842 1482nikhil.arela@bernsteinsg.com AI is the next ‘fighter jet’: Technology access has been rationed and controlled in the past,and a trend emerging in last few years is shattering the “Global” image of AI. It started withcritical minerals & semiconductors equipment, moved to GPUs and now exhibits itself inthe form of restricted frontier models. The recent ban on Anthropic’s latest models for non-US citizens confirms that it’s no longer an anomaly. As AI moves from a mere commodityto a tool of strategic importance, one thing is clear: foundational models will no longerbe SaaS products, they would rather be critical resources, becoming the “fighter jets” ofan era where bleeding-edge models will be guardrailed. This will only aggravate further,with deep implications for India currently trying to convince itself that applications builton foreign LLMs and earning rent on datacentres is a sound AI strategy. To be clear, weare not dismissing this approach. There is value in participating in the ecosystem throughinfrastructure build-out and application layers. But at the very least, India needs to be awareabout the risks embedded in effectively outsourcing the core AI models. Why doesn’t India have its own large-scale LLM?It may be counterintuitive that despitebeing the source of vast data, which is virtually the fuel to train global models, India itselfhasn’t harnessed this advantage to build competitive GenAI systems at global scale. Thisabsence of an India “DeepSeek moment” is structural rather than a thought out strategy.India’s tech ecosystem has long been services-led, with sparse consumer-facing platformsin areas like search, social media or messaging—domains that typically generate the rich,organized datasets critical for training advanced AI. The lack of such data ecosystemshas never necessitated the development of talent pipelines and academic depth to buildfoundational models. The IT services model, where low cost labor effectively fine-tunessoftware built by global giants, further adds to the baggage of sticking to application layer.Heads of many leading Institutions have argued that India does not need its LLMs, but canfocus on AI applications. These views are more reflective of the path India has taken tostand where it is today, rather than a deliberate strategic choice. Can India afford its AI stack at the mercy of someone else?Let’s picture this: India’score intelligence layer, from enterprise software to defence and space could be poweredby foreign LLMs. Enter a geopolitical disruption, and that access could be curtailedovernight, bringing critical applications to a halt. A more likely outcome is subtler butequally concerning: India operates one or two generations behind. Applications built onolder models struggle to compete with global offerings. An Indian IT giant with talentat its disposal could lose out to a young US SaaS firm bootstrapped by amateur codershaving access to cutting-edge models. There is, however, a path that looks more resilient—building domain-specific LLMs on proprietary data and layering AI solutions on top.The real vulnerability lies with large-scale, horizontal GenAI applications built on externalmodels....continued on page 2 DETAILS ...continued from page 1.Framing India’s AI Policy Choices:The solution set is not easy for policymakers. Ideas such asringfencing AI for domestic firms (which we have given in the past) may lack practicality in a globally integrated technologylandscape. That said, two strategic levers stand out. One approach is to restrict or pace access to global AI models, whilechanneling significant capital and talent into building India-native LLM capabilities. The other is to mandate or incentivizelocalization—requiring foreign firms to build and operate India-based AI stacks that are insulated from geopolitical controls.Both options are imperfect, but they frame the core policy trade-off between access and autonomy. INDIA’S TECH DEPENDENCE For years, India’s relationship with technology has followed a familiar script. Build locally where possible, but rely on globalgiants where it matters. The internet era was perhaps the clearest example. While China built walls and created large scaleplatforms that have provided it cash flows, talent pools and data to build AI models, India stayed open, and global platformsquietly became