FULL STORY
Sarvam's Token Factory Launch Sparks Sovereignty Backlash
India's Sarvam launched Token Factory, a GA inference platform on in-country GPUs, only to face immediate criticism for serving open-source models of Chinese origin, questioning its sovereign AI credentials.
2026-11-16 ~ 2026-11-17 · 2 episodes · 17 posts
Episode 1 · India's Sarvam Launches Token Factory, Hosting Open-Weight Models Domestically (2026-11-16, 15 posts)
On November 16, Indian AI company Sarvam officially launched its inference platform Sarvam Token Factory, now generally available. It hosts Sarvam's in-house models alongside several leading open-weight models on self-built GPU clusters located in India, serving all users.
Confirmed
- The platform offers 8 open-source models through a single OpenAI-compatible API — one integration gives access to open-source, reasoning, and decision models
- Official performance figures: Sarvam 105B achieves a time-to-first-token (TTFT) of 0.2 seconds and throughput of 330 tokens/s; specific data for the Gemma family and other models is also mentioned in official materials
- Covered use cases include enterprise workflow automation, software development, automated research pipelines, plus chat, code, and collaboration — with cybersecurity scenarios coming soon
- Documentation and a technical blog were released alongside the launch
Not Yet Confirmed
- itschefai, a member of the official marketing team, claims internal benchmarks show its TPS outperforming several inference providers — this is a vendor statement with no independent verification yet
Why It Matters
Data residency requirements and the lack of local inference capacity had previously limited Indian enterprises' adoption of frontier models. Token Factory delivers a domestically hosted open-source model stack, giving enterprises more model choices without vendor lock-in — a landmark step in building out India's sovereign AI infrastructure capabilities.
- Sarvam AI Launches Token Factory: India-Hosted Inference for Open Models — itsOmSarraf_ · 2026-11-16
- India's Sarvam launches Token Factory, hosting top open models on local infra — HarveenChadha · 2026-11-16
- Sarvam Token Factory goes GA, offering open models for enterprise workflows — itsOmSarraf_ · 2026-11-16
- India's Sarvam launches Token Factory: 8 open models, 0.2s TTFT on a B200 cluster — itsOmSarraf_ · 2026-11-16
- Sarvam Token Factory goes live, internal TPS benchmarks claim parity with top providers — itsOmSarraf_ · 2026-11-16
- Sarvam launches Token Factory inference stack serving open-weight models from India — itsOmSarraf_ · 2026-11-16
- India's Sarvam launches Token Factory hosting Kimi, GLM, DeepSeek open models — itsOmSarraf_ · 2026-11-16
- Sarvam launches Token Factory, hosting open-weight models like DeepSeek in India — itsOmSarraf_ · 2026-11-16
- India's Sarvam AI launches Token Factory, hosting open-weight models locally — itsOmSarraf_ · 2026-11-16
- Sarvam AI launches Token Factory, hosting open-weight models in India — itsOmSarraf_ · 2026-11-17
- Sarvam launches Token Factory: India-hosted open-weight models via unified API — itsOmSarraf_ · 2026-11-17
- Sarvam AI launches Token Factory serving stack with Kimi and GLM on B200s — itsOmSarraf_ · 2026-11-17
- Sarvam AI launches Token Factory, hosting top open-weight models in India — itsOmSarraf_ · 2026-11-17
- Sarvam launches Token Factory: India-hosted open-weight model inference with 99.95% SLA — itsOmSarraf_ · 2026-11-17
- Voice Agent Built on Sarvam-Hosted Gemma 31B for Automated Receptionist Calls — itsOmSarraf_ · 2026-11-17
Episode 2 · India's sovereign AI lab Sarvam accused of mostly hosting Chinese open models (2026-11-17, 2 posts)
Sarvam, picked by India to build sovereign foundation models with major GPU subsidies and a $1.5B valuation, is facing criticism that it mainly hosts Chinese open-source models.
- India's sovereign AI lab Sarvam hosts mostly Chinese models on subsidized H100s — ayushthakur0 · 2026-11-17
- India's sovereign AI lab Sarvam slammed for mostly hosting Chinese models — itsOmSarraf_ · 2026-11-17