How does enterprise AI adoption in the APJ region differ from trends in North America and Europe?

I look after APJ, which spans three geographies: Japan, APAC, and India. Most hyperscalers or big cloud players are either from the US or China. Some smaller regional players may be big in their countries, but not globally. We know Gen AI is useful, and at scale, with lots of data and infrastructure, it can do magical things. But how can we use it more effectively for society?

AI started with large language models (LLMs). But different languages lack the dataset and the level of maturity as in English LLMs. The big trend is devising a local language model catering to non-English speakers. This requires non-English data. The workloads in APJ are trying to create mechanisms to gather this data. There is a lot of context and cultural elements in the data, which will be used for training and learning to develop a more focused, sovereign AI.

The workload trend is trying to create a subset catered to specific countries or cultures. We are in the transition stage, looking beyond global scale or English to more specific domains.

With the maturity of language models, the language barrier suddenly breaks down. Both parties can speak their native language, and communication improves as a result of this technology. This dialogue creates new opportunities, especially for enterprises, as language is no longer a barrier. One can expand and provide their services or benefits at a wider scale, and AI is all about scale, as with HPC. The more scale and data, the better the results.

Within the APJ region, which countries are leading enterprise adoption of AI?

There are different levels of maturity in each country. In certain countries, semiconductor manufacturing is advanced, such as Taiwan and Korea, Japan in the automotive industry, or Singapore in finance. The differences in enterprise maturity translate into the speed of AI adoption across industries. The more mature the base industries, the more they look into advanced technologies to propel them to the next stage. The maturity of certain industries across countries in the region is balanced and diverse. For instance, in Korea, both because of advanced industries and language challenges, the government is pushing for sovereign AI, while large enterprises are pushing to adopt AI to streamline operations, improve quality, and better understand their real strengths. AI can look at data from a perspective humans can’t, providing the necessary insight to move to the next stage of evolution.

Where do Indian enterprises currently stand on AI adoption, given the mixed views on whether India is leading adoption or still has significant ground to cover?

India is relevant as it is trying to provide the large-scale infrastructure that AI requires. Many new data centres are power-hungry and need water for liquid cooling, and India is among the top countries providing this. They are starting investments, not just in the public sector but also in concert with the private sector, to ensure the necessary infrastructure to propel India forward.

In terms of actual consumers, the overall challenge, not just in India but across the APJ region, is the lack of local-language or locally specific data. We are still in the data accumulation stage, and need to figure out how to break the data silos and ensure data availability for AI training.

Having advanced data centres to support advanced AI infrastructure is a challenge for most countries. For instance, Japan has legacy data centres with not enough power or that don’t support liquid cooling. Suddenly, this new requirement is a power-hungry infrastructure that requires liquid cooling. You first have to deal with this legacy. In India, the number of legacy data centres is smaller, so it’s easier to leapfrog to advanced data centres and move ahead of other countries in terms of infrastructure readiness. That’s a prerequisite.

The biggest challenge, especially for developed countries like Japan, is coping with legacy infrastructure. For other countries, it’s more about how they can leapfrog into this new age because they can start from zero. This is the key challenge and an opportunity for India.

When do you see enterprises moving beyond AI POCs and committing to production-scale deployments?

The results of a small POC may be inaccurate, misleading, or hallucinatory. This is the worst way to start the AI journey because for it to work accurately, there needs to be some scale and access to data. A POC without this is doomed to fail.

Enterprises are starting to realise they can try out the technology in the cloud using open-domain data, then bring it into the enterprise on-prem using their own confidential data. But there’s always a risk of data reuse for AI when you use it in the cloud. You start on the cloud and see potential in the POC, but how do you bring it home? That goes back to the data challenge. You need to break the data silos within the company and ensure that the on-prem AI platform has access to company data. Then, POCs can be done with meaningful scale and data. You see success, value, and it’s easier to progress into production. Most POCs don’t go into production because they didn’t have enough access to their own data, train enough, or understand the technology enough to benefit from it. Premature POCs need to move toward more structured, mature POCs that have real access, not just sample or synthetic data, but real data. That’s the key challenge. Once customers realise that, the transition to production becomes smooth.