India's sovereign AI lab Sarvam hosts mostly Chinese models on subsidized H100s
ayushthakur0 · x · 2026-11-17
Sarvam AI was picked by the Indian government in April 2025 to build India's own foundation model, receiving a ₹98.68 crore GPU subsidy for 4,096 H100s and raising $234M at a $1.5B valuation in June. Its Token Factory inference platform is now open to all with 8 models: only 1 is Sarvam's own, 6 are Chinese open-weight models (DeepSeek, Kimi, GLM) plus Google's Gemma. @sanchitmonga22 argues this makes Sarvam a hosting business rather than the frontier lab India hoped for; @ayushthakur0 counters that deep-tech progress is non-linear, and that running inference at scale — beyond just deploying vLLM/SGLang — is itself hard.
More from Companies & People
- NYU professor calls Jensen Huang's return to Nvidia a 'fantasy' — GaryMarcus · 2026-11-17
- SpaceX uniquely owns the ride to orbit, the global network, and the AI computers at once — ccerrato147 · 2026-11-17
- SpaceX catches a returning spaceship, projects $33–37B revenue in Q1 2027 vs $5.6B in Q3 2025 — ccerrato147 · 2026-11-17
- Scoble: NVIDIA's parking lot is the world's best recruiting ground, echoing Gates' Borland limo ploy — Scobleizer · 2026-11-17
- WSJ goes inside Nvidia's culture: Huang's harsh employee eviscerations and rising 'royal family' — kevinnbass · 2026-11-17
- Perplexity CEO: AI for automating scientific research is the best compute investment — dr_alphalyrae · 2026-11-17