AI computing requires 10x more energy, but we source over 50% from green alternatives: Anil Pawar, Yotta Data Services

In an exclusive interview with Express Computer, Anil Pawar, Chief AI Officer, Yotta Data Services, discusses the company’s recent partnership with the Telangana Government to build India’s largest AI supercomputer. Pawar highlights the immense potential of AI in transforming the data centre landscape and emphasises Yotta’s role in advancing AI-driven infrastructure in India. He elaborates on the company’s commitment to green energy, the challenges of integrating AI into data centres, and their efforts in reskilling employees. Pawar also touches on the critical role of sustainable cooling solutions and the growing energy demands of AI, offering insights into Yotta’s plans to address these challenges with innovative solutions.

Yotta recently launched India’s largest AI supercomputer in collaboration with the Telangana Government. Can you elaborate on this initiative and how it positions Yotta at the forefront of AI infrastructure in India?

The Telangana Government has done a fantastic job with this initiative, especially considering the immense demand and the technological transformation India is undergoing. The AI race is arguably the biggest since the nuclear race, and the country that leads this technological race will dominate in terms of innovation and progress. Both the Central and state governments are focusing heavily on this, and Telangana has taken a significant first step.

The agreement we have with the Telangana Government involves building a data centre that will eventually host 25,000 of the latest GPUs. To begin with, we’re aiming to have this data centre up and running in two years, starting with 4,000 GPUs. This infrastructure will serve a wide range of users—startups, enterprises, and various government departments within Telangana. I believe this will be a game-changer for the Telangana Government.

As Yotta, we’ve already executed large-scale projects like this, so we’re excited to contribute to the growth of the state and the country as a whole. We’re very bullish about the Telangana Government’s initiative and feel fortunate to be selected as a partner in such an important project.

How do you see AI transforming the data centre landscape, and what unique role does Yotta play in advancing AI-driven data centre technologies in India?

AI is transforming the data centre landscape by touching every business and industry, something we haven’t seen with previous technologies, which were usually specific to certain verticals. At Yotta, we’re leveraging AI to optimise our data centre operations in several ways. For instance, in physical security, we use AI-enabled cameras to detect suspicious activities, enhancing safety measures. But the biggest game-changer has been AI-assisted energy management. Energy accounts for about 30% of our costs, both in capex and opex. By using AI to manage energy consumption—like adjusting GPU states based on workload—we’re achieving significant savings.

Additionally, AI has improved our operational efficiency. We’ve automated monitoring, eliminating the need for manual oversight and allowing AI to notify the team only when necessary. While we’re just scratching the surface, the gains we’ve seen so far have been remarkable. I believe AI-assisted data centres will be essential for everyone moving forward.

What are some of the most significant challenges in integrating AI into data centres, and how has Yotta addressed these hurdles to ensure optimal AI performance and scalability?

We’re developing several AI products, and we’ve observed two significant challenges when integrating AI into data centres. First, there’s the organisational challenge. Many organisations are not prepared to adapt to new technologies, and this requires a major mindset shift. Everything is becoming automated, proactive, and self-service, especially with AI. This shift can be difficult for people within the organisation as it requires a new way of thinking.

Second, there’s the challenge of skills and competencies. Since AI is a new technology, many organisations struggle with a lack of the necessary expertise. However, we’ve been addressing these challenges by transforming Yotta into an AI-native organisation—what we call “AI Yotta.” We are embedding AI across all functions, from HR to administration, ensuring that AI is a core part of our operations. While these challenges are common during technological transformations, overcoming them is essential for success.

I believe you mentioned that change management is one of the major hurdles in incorporating AI. As the head of AI at Yotta, do you have any training sessions or programmes planned to reskill or upskill your employees in this particular area?

We’ve mandated an internal AI certification for employees, which is tied to their performance appraisal. This incentivises everyone to complete it, as there are rewards for those who achieve it. We’ve also arranged expert training sessions, collaborating with industry leaders like Nvidia to help reskill and upskill our employees. As the AI business unit, we’re heavily investing in training across the organisation.

On a broader scale, I believe India as a whole needs to focus on upskilling, especially as AI begins to disrupt traditional business models. Companies like Wipro and TCS are already investing heavily in upskilling to remain competitive. If IT businesses don’t become AI-native, they will face significant challenges. With India’s young population, we have a unique advantage, but we need to build strong platforms to train and prepare our workforce for the AI-driven future.

Comparing our data centres in India to those in cooler regions like Europe and the Nordics, we face higher cooling costs due to the climate. Considering the increasing power demands of AI data centres, what sustainable cooling methods can we implement?

When we started at Yotta, transitioning from a standard data centre to an AI data centre seemed like a normal upgrade. However, we quickly faced significant challenges. AI computing is about 10 times more energy-intensive, which also means it requires far more cooling. You’re right in pointing out that countries in Europe and the Nordics benefit from cooler climates, reducing their cooling needs, whereas in India, with its hot and humid environment, cooling demands are much higher.

To address this, we’re leveraging India’s strength in green energy, particularly solar and wind. At Yotta, more than 50% of our energy is already sourced from green alternatives, and we’re continuously working to increase that to 70%, with the ultimate goal of becoming carbon-free in the coming years. While the climate poses challenges, we’re fortunate to have access to abundant renewable energy, and we’re focused on improving the technology to harness it efficiently.

What type of cooling method do you use?

Initially, we used standard cooling methods for earlier CPU generations in the cloud. However, when we built our AI data centre and added GPUs, which consume significantly more energy, we needed a different cooling approach. Currently, we use air cooling, where the heat is extracted from the rear of the rack using a rear door heat exchanger (RDHX). This method minimises energy consumption for cooling. As we progress, we’re transitioning to liquid cooling, especially with the next generation of GPUs, which will require around 150 kilowatts per rack. So, right now, it’s a combination of air and liquid cooling, but we’ll eventually shift to full liquid cooling.

With increasing energy demands due to automation, AI, and GenAI, and the rise of AI data centres, do you think nuclear energy or small modular reactors are the most viable options to sustainably power this growing computing load?

Absolutely. I believe nuclear energy, particularly in the form of smaller modular reactors, is a viable option. As GPU consumption grows, so does the energy demand. While we’re moving towards green energy, it’s not yet possible to meet 100% of our needs through traditional renewable sources. That’s why nuclear energy, especially in a modular form, is gaining traction. The energy demand is increasing exponentially, and nuclear energy seems to be the most practical solution to address this growing gap.

However, setting up nuclear reactors for data centres will require new infrastructure, policies, and a specialised skill set. There are potential side effects, and we need to address all of these factors carefully. At the same time, companies like Nvidia, AMD, and Intel are working on reducing the energy consumption per GPU, which is a critical metric. So, addressing the energy challenge requires efforts from both the technology and data centre sides, focusing on consumption, cooling, and cost management.

Looking ahead, what AI innovations or upcoming projects at Yotta are you most excited about, and how do you envisage them shaping the future of AI in both India and globally?

At Yotta, we envisage building a comprehensive AI ecosystem in India, beyond just offering GPUs. Our goal is to provide end-to-end AI solutions for startups, universities, and researchers, with the aspiration that the next big AI model, like ChatGPT or OpenAI, emerges from India. We believe we can achieve this by meeting the country’s massive compute requirements at affordable prices. To support this vision, we’re innovating products like our AI Lab—a cloud-based solution that universities and companies can access quickly and affordably. This helps overcome the high costs of GPU infrastructure, making it easier for India to create its own AI models and technologies.

With roughly 20% of the world’s data generated in India, but only about 3% to 4% data centre capacity, how do you plan to capitalise on this deficit?

Yes, you’re absolutely right. India leads in global data consumption, but we significantly lack data centre capacity—currently at just 3%. This presents a huge opportunity, and that’s why we’re seeing a surge in companies setting up data centres in India. To support the country’s unprecedented digital transformation, which no other nation is matching, we need AI-specific data centres.

We’re actively expanding to bridge this gap. We began with a 50-megawatt data centre in Mumbai, which can scale up to 250 megawatts. Then we launched a data centre in Delhi and recently opened another in Gift City. We’re also building the Telangana Government Data Centre. This expansion is critical to India’s growth, and we’re focusing on building green data centres.