Indian firms reach for Chinese models as the AI bill comes due
With inference costs still climbing, Indian enterprises are quietly routing more of their AI workloads through DeepSeek, Alibaba and Moonshot, a shift that puts New Delhi's compute-at-home ambitions on a collision course with its own procurement logic.
Indian companies are quietly rewiring their AI procurement stacks, and the new suppliers do not sit in Bengaluru's whitefield corridor. Reporting from Nikkei Asia on 13 July 2026 details a steady drift among Indian enterprises toward large language models built by DeepSeek, Alibaba and Moonshot AI, the Chinese labs whose open-weight releases have reset the global price of inference. The shift is not ideological. It is arithmetic.
The cost of running frontier models on Western APIs has climbed through 2026 as hyperscalers repriced tokens, throttled free tiers, and pushed customers toward longer commitments. Indian CTOs, facing rupee-denominated budgets and quarterly earnings calls, have started reading the invoice differently. Cheaper models from Chinese labs, even with translation layers and integration overhead, now pencil out for production traffic that would have routed to US providers a year ago.
The price gap that started it
Two structural facts explain the move. First, open-weight Chinese models have closed the capability gap on most enterprise tasks, summarisation, classification, retrieval-augmented generation, customer-support triage, at a fraction of the operating cost. Second, India's domestic compute build-out, the IndiaAI Mission and the planned expansion of sovereign data-centre capacity, is not yet producing the kind of price floor that would let home-grown alternatives compete purely on cost. Until that capacity comes online, procurement teams vote with the spreadsheet.
Nikkei's reporting points to companies in financial services, IT services, and online commerce piloting DeepSeek, Alibaba's Qwen family and Moonshot's Kimi for back-office workloads. The pattern is consistent with what enterprise buyers across Southeast Asia have done since the open-weight wave began: pilot on the cheapest capable model, then expand to customer-facing surfaces once latency and reliability hold up in production.
The sovereignty counter-argument
The Indian state has reasons to be uncomfortable with that arc. New Delhi has invested political capital in a domestic AI stack, in public-sector deployments that route through Indian data centres, and in a narrative that frames compute sovereignty as a strategic asset on par with spectrum or semiconductors. Routing enterprise inference through Chinese labs, even via commercial APIs, sits awkwardly inside that frame. Watchers in the security establishment will note that DeepSeek, Alibaba and Moonshot operate under Chinese jurisdiction and could be subject to intelligence cooperation obligations; Western providers face the same structural exposure under US law, but the geopolitical weighting differs.
The counter-argument from the procurement side is straightforward. Capability and price matter because they decide whether an AI feature ships at all. A model that costs a tenth of its competitor per token is the difference between a product that reaches customers this quarter and one that gets parked in next year's roadmap. Indian enterprise software exports compete on cost with every other outsourcing destination; shaving inference bills is not optional decoration, it is gross margin.
There is also a quieter reading. The Chinese labs have an incentive to underprice Western rivals because the strategic prize is global distribution, the same logic that drove Huawei's early push into African and South Asian telecoms gear. If DeepSeek and Qwen become the default engine inside Indian enterprise stacks, the standards, integrations, and developer habits that follow will travel with them for a decade.
What Beijing wants from this
Beijing's industrial policy has not hidden the goal. Chinese model labs benefit directly from the open-weight strategy. By releasing competitive weights under permissive licences, they commoditise the model layer and push the value chain toward inference infrastructure, applications, and data, areas where Chinese cloud providers can compete on price inside South and Southeast Asia. The same playbook produced the dominance of Chinese battery cells in Indian two-wheeler and storage markets: undercut on price at the unit level, win the design-in, then capture the long-tail services revenue.
Indian policymakers are aware of the pattern. The relevant question is whether New Delhi responds with compute subsidies, local-model procurement preferences, or quiet tolerance of the price-led Chinese imports. Each path carries costs. Subsidies distort capital allocation; preferences slow adoption of genuinely better tools; tolerance hands a strategic dependency to a rival power.
What to watch next
Three indicators will show whether the current drift hardens into a structural dependency. First, the trajectory of the IndiaAI Mission's compute capacity against actual inference demand through year-end 2026, and whether domestic GPU availability narrows the price gap enough to matter for procurement. Second, the composition of large Indian IT services contracts announced in the second half of 2026, where the default model choice is usually disclosed in technical appendices. Third, any movement on India's regulatory treatment of Chinese-origin AI services, currently constrained but not categorically closed, that would either lock the door or leave it ajar.
The Indian AI story is no longer about catching up. It is about which supplier stack gets embedded into the country's enterprise plumbing before the local alternative reaches price parity. That decision is being made in procurement meetings across Mumbai, Hyderabad and Bengaluru, on contracts that rarely make headlines but compound for years.
This article was prepared from wire reporting by Nikkei Asia and reflects Monexus's framing of the cost-versus-sovereignty trade-off in India's enterprise AI procurement. Sources do not specify named CTOs or individual contract values.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://t.me/NikkeiAsia
- https://t.me/nikkeiasia
- https://en.wikipedia.org/wiki/DeepSeek
- https://en.wikipedia.org/wiki/Qwen