Bahrain Telegraph - Sarvam AI Faces Scrutiny Over Hosting Chinese Models on Token Factory Platform

NYSE - LSE
AZN -1.41% 176.28 $
CMSC -0.03% 19.085 $
RIO 0.67% 76.41 $
CMSD -0.11% 18.759 $
BCC -1.05% 65.06 $
GSK -1.1% 48.05 $
BCE -6.29% 20.745 $
RYCEF 1.98% 17.7 $
RBGPF -0.46% 75.3 $
RELX -0.48% 38.83 $
VOD -0.88% 14.72 $
JRI -0.03% 10.847 $
BP 0.86% 43.62 $
BTI -2.06% 57.32 $
NGG -0.87% 81.5 $
Sarvam AI Faces Scrutiny Over Hosting Chinese Models on Token Factory Platform
Sarvam AI Faces Scrutiny Over Hosting Chinese Models on Token Factory Platform

Sarvam AI Faces Scrutiny Over Hosting Chinese Models on Token Factory Platform

Bengaluru-based artificial intelligence startup Sarvam has come under fire following the launch of Token Factory, a commercial inference platform that provides access to global AI models. The controversy centers on the inclusion of six Chinese-developed models alongside Google’s Gemma and Sarvam’s own offerings, prompting critics to question whether government-subsidized computing resources are being used to serve foreign technology.

Text size:

Bengaluru-based artificial intelligence startup Sarvam has come under intense scrutiny following the launch of Token Factory, a commercial inference platform designed to allow developers and businesses to access global AI models hosted on local computing infrastructure. The backlash focuses on the company’s decision to include six Chinese-developed AI models in its catalogue, raising questions about the alignment of its commercial activities with India’s broader goals for sovereign artificial intelligence.

Launched on November 18, Token Factory is a commercial inference stack that enables startups, enterprises, and developers to run various AI models on Sarvam’s computing infrastructure located within India. The platform aims to reduce the significant costs and time associated with building AI products from scratch by providing access to graphic processing units (GPUs), compute infrastructure, and coding expertise through a single application programming interface (API).

The model catalogue available on Token Factory includes DeepSeek V4.1 Flash 1017, Kimi K3.1 Pro, Sarvam Prism, Qwen 3.8 27B, GLM 5.3, GLM 5.3 Flash, Clef, EmbeddingGemma 2, Gemma 4 31B, DeepSeek V4 Pro, Sarvam 105B, and Sarvam 105B Conversations. The platform also features Google’s Gemma and Sarvam’s own foundational model, Sarvam 105B.

Critics have questioned whether the company is utilizing government-subsidized GPUs to serve foreign-developed AI models, particularly given Sarvam’s recent high-profile engagement with the Indian military. The controversy emerged just days after the company signed a deal with the Headquarters Integrated Defence Staff (HQ IDS) of the Ministry of Defence to develop and deploy AI applications for the Indian Army, the Indian Navy, and the Indian Air Force.

Sanchit Monga, CEO of the US-based AI inference company RunAnywhere, publicly criticized Sarvam’s strategy on social media platform X. He highlighted that Sarvam was selected by the government in April 2025 to build India’s own foundational model and received a GPU subsidy worth Rs 98.68 crore for 4,096 H100 units. Additionally, the company raised $234 million at a valuation of $1.5 billion in June.

Monga argued that with Token Factory, Sarvam has opened its platform to everyone, hosting eight models: one proprietary, six Chinese open-weight models from developers including DeepSeek, Kimi, and GLM, and one from Google. "So India’s sovereign AI lab is mostly serving Chinese models from Indian GPUs. That’s a hosting business, and plenty of companies already run one," Monga wrote. He expressed disappointment that the company, after receiving such significant capital and subsidized compute, was not acting as a lab that ships frontier research out of India.

Other critics raised concerns about infrastructure centralization. Srikanth Lakshmanan, founder of CashlessConsumer, warned that Sarvam’s role as an inference provider risks creating a de facto national inference corporation. He argued that if the Government of India intends to boost the AI economy through GPU subsidies, it should decentralize infrastructure to districts to ensure equitable and distributable growth.

In response to the social media backlash, Sarvam AI clarified that Token Factory runs entirely on privately procured GPUs and does not rely on government subsidies for its inference operations. Pratyush Kumar, co-founder of Sarvam AI, explained that while the company received subsidised access—a 40% discount—to GPUs in India through the IndiaAI Mission to train models, this arrangement was strictly for model development.

"The outcome was the creation of open weight models, which are published on HuggingFace and AIKosh, with permissible licences. Our access to these GPUs ended in March 2026," Kumar told The Hindu businessline. He emphasized that Token Factory is intended to serve as a domestic alternative to foreign inference platforms, offering rupee pricing and data sovereignty. "Inference runs on Sarvam-operated GPUs in India. Requests are not forwarded to the original model developer’s overseas API," the company stated in its FAQ.

Industry observers have pushed back against the criticism, characterizing Token Factory as a standard commercial move. Nikhil Pahwa, founder and editor-in-chief of MediaNama, stated he saw no issue with Sarvam launching an inference platform or hosting Chinese models. He noted that Sarvam needs to build monetization and compete in the open market, targeting coders with its primary product-market fit.

Pahwa pointed out that the pricing for DeepSeek V4.1 Flash on Token Factory is similar to market pricing on other platforms like OpenCode and Command Code, suggesting there is little reason for users to switch solely based on price. Regarding the subsidy debate, Pahwa argued that Sarvam’s obligations are strictly defined by its government agreements rather than public sentiment.

"You have to check what the Indian government got contractually for giving Sarvam discounted GPUs. They’re a business, and they have to work in shareholder interest. Nationalism is the pitch that got them the GPUs, but they’re not expected to deliver anything beyond their contract," Pahwa said. He added that launching an inference stack does not mean the company is abandoning foundational model research.

Hemant Mohapatra, a partner at Lightspeed India, which is an investor in Sarvam, noted that global AI companies, including those developing Chinese models, are generating significant revenue by running their models on GPU infrastructure hosted in India. Amid a global GPU shortage, Mohapatra suggested that Token Factory represents a timely attempt by Sarvam to capitalize on this opportunity.

"Just because a company launched product-x, doesn’t mean that’s where the launches are going to stop?" Mohapatra asked, implying that the inference platform is just one part of Sarvam’s broader business strategy.

N.al-Kaabi--BT