Google on Thursday said that it will be taking India’s digital public infrastructure (DPI), which includes Aadhaar, Digiyatra, among others to different countries.
The company has launched ‘DPI in box’ model in collaboration with the Nandan Nilekani-backed EkStep Foundation and digital governance firm Protean Technologies.
It will be a plug and play kind of model where countries can take the DPI solutions by India and Google will come on board as an enabler to facilitate that.
“We are not going to build products in DPI, our partners may do so. This is a model of partnership. This is about saying, what can we do to accelerate India’s leadership in this space,” said Bikram Bedi, vice president and country managing director of Google Cloud India.
Notably, India’s DPI is hosted on Google Cloud and analysts said through expansion to other countries, the company could look at higher revenues.
According to Bedi, certain countries will have different requirements with regard to DPI. Some would want the solution on the public cloud and some would want it on the sovereign cloud. “So, how you design for different requirements would be really about building the right product. Apart from that, the partners will also work on a go-to-market strategy to take DPI globally,” Bedi said.
Lately, the government has been in touch with some countries to export DPI stack.
At the Google for India event on Thursday, the company also announced that it will be introducing Gemini 1.5 Flash in India in the coming months with a capability to enable Indian organisations to store data in India and run complex AI models locally. The AI model Gemini 1.5 Flash will introduce cost-efficient compute costs with right performance for AI startups and companies in the country, Bedi said in a media roundtable on the sidelines of Google for India event.
Comments from Bedi assume significance as the current cost for training and running machine learning models, performing inferences, among other tasks is very high. Higher costs make it difficult for startups and early stage companies in AI to have a right relation between costs and performance.
Notably, data localisation is a key regulatory requirement for some industries such as finance.
“This is based on our customers’ ask because there are certain workloads where customers can use models globally, but there are certain workloads where they would like the model to be in India,” Bedi said.
“Price performance starts to become a very important factor, when you (AI companies) start to go into production (of AI models in real-world applications) and you’re starting to scale. And hence, Google’s ability comes to provide the right forms of price,” Bedi said.
Google showcased Gemini 1.5 Flash a few months back at the Google I/O 2024 event. The company said this is its fastest and versatile multimodal AI model. Using APIs, startups and companies can integrate Gemini models and integrate advanced AI capabilities for their applications.
For startups and AI companies, models like Gemini could help reduce the computational costs by optimising performance for common tasks, without requiring expensive infrastructure like large GPU clusters.
“It is a question of price performance because you may not need a 780 billion parameter model for something which could be solved with a 10 billion or an 8 billion parameter model,” Bedi said.
According to Bedi, Google’s AI stack includes hardware – GPUs and TPUs (tensor processing units), software layer with Vertex AI studio and its first party applications such as Google Workspace, translation, etc, and that puts the company in a unique position.
