Small Models, Own GPU Infra: Sridhar Vembu Reveals Zoho's AI Strategy

Sridhar Vembu, founder of Zoho, revealed on December 16 that the company is developing smaller artificial intelligence (AI) models. These models, based on 7 billion to 20 billion parameters, aim to address specific domain issues for their customers.

Vembu stated at the CNBC-TV18 and Moneycontrol Global AI Conclave that they have discovered smaller models to be more effective for domain-specific issues, which is why they are currently not pursuing the 500 billion parameter models. Additionally, they are interested in acquiring their own GPU infrastructure, as it proves to be more cost-effective in the long run.

"There are two forms of capital. Brain capital is the bigger one. If that is solved, financial capital will come. It's not that $50 million or $100 million can't be found in Bengaluru. Now, AMD is also coming with AI chips. We are working with them. The supply chain issues of chips will be resolved," Money Control quoted him as saying.

"That 80 percent could be automated. That is the low-hanging fruit," he was quoted as saying.

Recently, the renowned software-as-a-service (SaaS) provider, Zoho, introduced a collection of 13 AI-powered extensions and integrations for its applications, all driven by ChatGPT. Zoho is advancing its AI functionalities by incorporating third-party intelligence into Zia, its proprietary secure cloud-based AI engine.

In June 2023, Zoho revealed its plans to develop its own extensive language model (LLM), akin to the GPT model of OpenAI and Google's PaLM 2. This project is under the supervision of Vembu and is in progress by Zoho's Research and Development team based in India.