
Sarvam Launches Token Factory as India's AI Inference Market Crowds
Sarvam's Token Factory runs Indian and global AI models on in-country GPUs, as Perplexity pulls its IPO forward and Claude downloads surge 15X in India.
- By
- Tom Whitfield
- Filed
- Channel
- AI & Compute
- Read
- 4 min read
Indian AI unicorn Sarvam has switched on Token Factory, an inference service that offers enterprises a single API for Indian and global AI models, with every workload running on GPU clusters located inside India. The launch lands Sarvam in direct competition with Krutrim, OpenAI and Anthropic, all of which already serve inference from Indian servers.
What is Token Factory actually offering?
The platform hosts Sarvam's own 105B-parameter model alongside open-weight models from Google, DeepSeek, GLM and Kimi. Developers can switch between models through one API without provisioning their own compute. Sarvam credits a three-tier caching architecture for latency and throughput figures the company says rival global inference providers.
The commercial core of the pitch is sovereignty. Because every model runs on Sarvam's India-based GPU clusters, the company is targeting enterprises bound by strict data residency mandates. For banking and financial services customers, Sarvam adds dedicated GPU capacity and on-premises inference deployments.
The offering is live now, but the market around it is not empty. Krutrim has hosted DeepSeek models on Indian servers since last year, OpenAI recently launched India-based API inference for GPT-6 Luna workloads, and Anthropic offers in-country inference for select Claude models through Amazon Bedrock.
The takeaway: Sarvam is selling data residency and multi-model access, not a technical first.
Does hosting foreign models undercut the sovereign pitch?
Critics have flagged a tension in the launch. Sarvam positions itself as a champion of sovereign Indian AI, yet Token Factory prominently hosts overseas open-weight models. Sarvam's defense is that hosting global models locally still satisfies data residency requirements.
The question of whether reliance on foreign open-weight models dilutes the push for homegrown AI capability remains open. Sarvam's wider bet rests on Indian enterprises paying a premium for local inference, multi-model flexibility and residency compliance — a bet that demands continuous investment in GPUs, software and enterprise relationships while cloud giants and global inference providers move quickly.
How is the AI app market shifting in India?
New download data shows the competition is already reshaping consumer behaviour. Anthropic's Claude recorded more than a 15X year-on-year increase in downloads in India during the July-September quarter, while ChatGPT downloads fell 21% over the same period. ChatGPT nonetheless kept its lead as India's most-downloaded AI app.
Spending tells a different story. OpenAI's India revenue from in-app purchases rose 70% year on year in Q3 2026 despite the slower download growth, and Claude climbed to third place in India's app-spending rankings. Both OpenAI and Anthropic have opened India offices and are ramping up local investment, with the AI ecosystem projected to add $1.7 trillion to India's GDP by 2035.
The pattern: downloads are fragmenting, but ChatGPT still converts users into payers at the largest scale.
Why is Perplexity speeding up its IPO?
Perplexity now plans to file draft IPO papers by Q1 2027 for a US listing, pulling the plan forward from founder and CEO Aravind Srinivas's earlier position that the company would not go public before 2028.
Srinivas attributed the acceleration to growth outrunning forecasts. The company has already reached $1 billion in annual recurring revenue, surpassing the $750 million ARR projection it had set ahead of the offering. The move comes as OpenAI files confidentially for an IPO it targets next year and Anthropic prepares to list on the Nasdaq this week.
The takeaway: the AI sector's leading names are converging on public markets within the same twelve-month window.
What else moved this week?
Consumer launches and capital raises
- boAt cofounder Aman Gupta has partnered with actor Ranveer Singh to launch H.I.M., a men's grooming brand debuting with eight products priced between ₹299 and ₹899. The venture sits under Gupta's studio OFF/BEAT and raised ₹100 crore in an April seed round led by Bessemer Venture Partners. India's men's grooming market is projected to reach $4.5 billion by 2034.
- Urban Company opened a ₹2,500 crore qualified institutional placement, earmarking ₹600 crore over three years for its 10-minute house help service InstaHelp, ₹540 crore for marketing and ₹200 crore for technology and cloud infrastructure. InstaHelp posted an adjusted EBITDA loss of ₹305.96 crore in H1 FY27 despite delivering more than 1 crore orders.
- FirstCry retained mid-teens revenue growth guidance for FY27 after narrowing its Q2 FY27 net loss by 75% year on year to ₹12.6 crore, with operating revenue up 14.1% to ₹2,395.8 crore. Management expects a stronger Q3 on festive demand and easing diaper-category discounting.
Startup spotlight: Lorazzo
Founded in 2024 and backed by Sauce.vc, Lorazzo sells design-first faucets, sinks, basins, sanitaryware and hygiene products suited to Indian conditions such as hard water. The company sells through its own site and marketplaces, works with more than 100 dealers across Delhi NCR, claims to have tripled its business in three months, and targets a ₹25 crore revenue run rate in FY27. India's bathroom and kitchen fittings market is projected to reach ₹48,300 crore by FY29.
For Sarvam, the decisive test is whether Indian enterprises reward local inference with paid contracts before the global providers deepen their own Indian capacity.
Source: inc42.com
More from Tom Whitfield
Show full bio
Staff writer covering companies, deals and strategy in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,054 articles
Related articles
sarvam-says-token-factory-serves-more-than-450-billion-tokens-a-day
Sarvam Says Token Factory Serves More Than 450 Billion Tokens a Day
india-to-buy-10-000-more-ai-chips-for-national-compute-pool-c2c0695c
India to Buy 10,000 More AI Chips for National Compute Pool
scality-claims-14x-faster-context-restore-in-on-prem-ai-inference-push-94336c78
Scality Claims 14x Faster Context Restore in On-Prem AI Inference Push
nvidia-s-india-footprint-45-bn-in-commitments-face-a-post-huang-test
NVIDIA's India Footprint: $4–5 Bn in Commitments Face a Post-Huang Test


