Sarvam’s Token Factory: 6 of 12 Models Are Chinese-Made

Indian AI startup Sarvam, which pitches itself on sovereign AI, has begun selling access to DeepSeek and other Chinese models, triggering an online backlash, Bloomberg reported (Bloomberg, via Techmeme).

Sarvam’s own Token Factory price list shows 12 model listings. Six come from Chinese developers: two DeepSeek models, two GLM models, Kimi and Qwen, by our count.

Business Pill · Concentration Explained

An evergreen Business Engineer explainer on market concentration. It teaches the general idea only and does not discuss Sarvam.

The key insight: Half of the 12 model listings on Sarvam’s Token Factory price list come from Chinese developers, by our count, while Sarvam says all inference runs on its own GPUs in India.

What Sarvam Sells

The Token Factory runs open-weight models through an OpenAI-compatible API, priced in rupees per million tokens, per Sarvam’s page. A Sarvam post introducing it is dated 16 November.

Sarvam’s own models make up three of the 12 listings: Sarvam 105B, Sarvam 105B Conversations and Sarvam Prism. Google’s Gemma 4 31B and EmbeddingGemma 2 are two more.

Six of the 12 model listings on Sarvam's Token Factory price list come from Chinese developers, by our count o
Six of the 12 model listings on Sarvam’s Token Factory price list come from Chinese developers, by our count of Sarvam’s page.

What Was Posted on X

Bloomberg’s post says Indian startup Sarvam has begun selling access to DeepSeek and other Chinese AI models, triggering an online backlash over its role as one of the main proponents of the country’s AI independence. The piece checks that claim against Sarvam’s own Token Factory price list.

Where The Data Goes

Sarvam says inference “runs on Sarvam-operated GPUs in India”. It adds that “Sarvam hosts the weights; requests are not forwarded to the original model developer’s overseas API”.

That answers where prompts are processed. It does not settle the criticism Bloomberg describes, which is about a sovereign-AI company reselling Chinese-built models.

The Price Gap

Sarvam prices its own 105B model at ₹10 per million input tokens and ₹25 per million output tokens. DeepSeek V4.1 Flash costs ₹14 and ₹55.

So Sarvam’s own model is about 55% cheaper on output than the cheapest DeepSeek listing, by our arithmetic. Kimi K3.1 Pro’s ₹950 output price is 38 times Sarvam 105B’s.

Why It Matters

Other AI-infrastructure deals we covered this week include a US group building Malaysia’s AI cloud. In Washington, a proposal would make companies disclose their use of Chinese AI models.

The Structural Read

As we read it, Sarvam is defining sovereign AI as control of the infrastructure rather than the origin of the weights.

As we read it, its own 105B model’s lower prices suggest the Chinese models are there for choice, not as the cheapest option.

Sarvam, Token Factory FAQ

“Model inference runs on Sarvam-operated GPUs in India.”

Three Implications

FOR INDIAN BUYERS Sarvam says requests are not forwarded to the original developers’ overseas APIs.

FOR SARVAM Bloomberg reports an online backlash over selling Chinese models while pursuing sovereign AI.

FOR POLICYMAKERS Six of 12 listings are Chinese-built, by our count of Sarvam’s page.

What Is Not Established

We read only Bloomberg’s headline and Techmeme’s summary, not the full article. The page we read does not respond to the criticism, and we have not seen usage figures for any model.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

The Map of AI

A Business Engineer map of 200+ AI companies across 9 layers.

Read the Map of AI →

The Bottom Line

Half of the 12 listings on Sarvam’s Token Factory price list are Chinese-built models, by our count. Sarvam says it runs them on its own GPUs in India.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Model listings, prices and the data-location statement are from Sarvam’s Token Factory page, read 19 November. The backlash is as described in Bloomberg’s headline and Techmeme’s summary; we did not read the full article. Model developers are from the models’ own public cards. The 6-of-12 count and the price comparisons are our arithmetic.

Sources: Bloomberg (headline) and Techmeme summary, 19 Nov 2026 · Sarvam Token Factory page: pricing (INR per 1M tokens, ex-GST) and FAQ, read 19 Nov 2026

Scroll to Top
FourWeekMBA