Interview: Zoho CEO Sridhar Vembu bats for technology sovereignty, focuses on creating domain-specific LLMs

If you look at that landscape, you have one extreme of everybody working in a centralised office, then you had everybody working from home, those two extremes we saw. But actually, we have to find a happy medium. That is where Zoho Corp itself is heading as a company. We are launching hub and spoke offices.

This hub and spoke model is centred around rural areas where there's one dominant hub office, and smaller satellite offices around, this model is very viable in that people work not necessarily from home, because that long term has issues -- of acculturation, bringing new people on board, all that. The small offices work, we have 20-30 people, connected to a larger office, maybe 500 or 1,000 people. That model is very viable. This model really needs the tools to go with.

How do we collaborate? This is where Cliq comes in. In fact, we tailored it for this type of problem. I think this is increasingly relevant. Bengaluru - take the example. Last week we had what three-hour traffic jam. That's why there's Cliq rooms - it's very easy collaboration. You see your colleagues on a large TV and your mobile phone itself acts as your microphone and speaker. So this is about as easy as it gets.

What's your broader vision for Zoho? It has evolved in so many ways from a workspace suite to this offering, you're also talking about the hardware side, LLMs and servers. You've also experimented with Zoho Meet and a messaging app Arattai. When one steps back and looks at everything that you're doing, what's the common factor?

It comes from the sheer complexity and the breadth and depth of technologies we depend on to work and to live today. We depend on a large number of tools to get our work done in the office, and even at home.

These technologies are also merging like WhatsApp is used for both business and personal use. In the same way, you have smart TVs, and they will integrate with our work. We are integrating with work, right? That's what necessitates all these technologies. What we want to be is a broad -spectrum technology provider that brings state-of-the-art technology, but at some very affordable and accessible price points to the global audience like the Middle East, Latin America, and Africa, we have a very large, growing customer base now.

Our growth strategy is about bringing all this technology together. This is a lot of heavy capability R&D we have to build, and a lot of nifty system integration. Cliq Rooms is an example of the integration, the Smart TV, phone, and the cloud, are all that nicely integrated, you don't realise they are integrated so well.

Like the famous Steve Jobs phrase, it just works. That's what we're aiming for.

Looking at the global situation, last week you raised a cautionary alarm on the demand for SaaS. Again, is that a North America problem? We are seeing pockets of concern emerge everywhere. How are you insulating yourself? Does it help to also have a growing India market?

Our strongest growth has been in India for quite some time. India's growth is continuing, we are still seeing tremendous growth while global growth has slowed down, particularly in the developed world -- US, EU, etc.
What is happening in China is really a bust at this point. We can see a real estate bust, infrastructure companies getting in trouble, and all that in China.

The economic news coming out of China is very grim and that is driving a lot of the global slowdown as well, because commodity, all of that, everybody is dependent on China for the demand, for the infrastructure, all that.

The war just adds a new dimension, an unpredictable dimension, we just cannot know -- oil prices for example. We are still bullish on India but the key here is we have to build the domestic capabilities. You mentioned 'insulate', I will use the word 'resilient'. Indian economy has to become more resilient and the only way we become resilient is to become much more self-reliant, the whole atmanirbhar movement is critical to technologies -- we have to acquire capabilities here. That is the key and we are contributing our part there.

Do you think India missed the whole AI wave? Because if you look at what's happening in the US, hundreds of millions of dollars being ploughed into AI startups, if you look at UAE strategy, China's strategy they've been buying up chips. Where are we here?

I wouldn't say India missed the bus so much so -- you can say that about Germany, Japan or South Korea. They would all say 'maybe we missed the bus, too'. I don't see it that way. I see it as we have to catch up on these capabilities, and we will. Part of it is seen in sometimes you get overhyped. Already you see the AI tools, the usage has come down from the peak. I think it's true for all of us. It's true. We started using it heavily. And then we know our usage intensity has dropped.

It's moved to the application layer because Google, Meta, etc are integrating it into all their products.

Definitely, that will continue and we are doing that integration. I'm saying that hype will wear off a little bit, durable uses will be found. We still have important technology challenges -- first is the hallucination problem to solve. Corporate and privacy protection, both are really linked because you cannot just -- in the name of AI -- violate randomly everybody's copyright and all of that. That's also important. For example, when it emits code, was it really just simply regurgitating some code it already memorised? And companies that use that code, will we be held liable for violating somebody's IP?

These are fundamental questions. Just because there's a magical new tool, these questions don't automatically get resolved. This is where a lot of work has to happen. I think there is time to catch up on all of this.

Have enterprises banned you from using ChatGPT?

Definitely, people want the ability to turn it off. We give them the option to turn it off. We don't just say we use it default. You turn it off. That's what we're doing.

Zoho was building its own large language model. Can you give us an update on that?

Right now, we are focused on very domain-specific models. The broad LLMs that are out there like ChatGPT, or Google Bard, are now rumoured to be in the trillion parameter range, not just hundreds of billions, but towards a trillion. At that point, they're memorising everything that humans have ever created. But when you come to specific domains, whether it's in content, legal, or medical, smaller models can do the same job. Domain-specific models are what we are focused on right now because that's pragmatic. They require smaller models and therefore smaller GPU count, etc. Often, the scaling is not linear. For some of these, there's an n squared scaling, where you throw a large number of parameters, it's not just simply linear scaling at that point.

So, you have to have a GPU strategy. The pragmatic strategy given the reality that GPUs are in short supply, is to focus on these domain-specific models. That's what we are focused on. We are also working with AMD, they are coming up with new GPUs.

Are you planning to acquire compute?

Definitely, as we have to make sure the models work on AMD. That's something that requires software work. AMD also has to come out with the chips they've announced. They'll soon be in the market. Those are all things in the future but we are working on this actively.

We recently spoke to Arvind Krishna of IBM, Jensen Huang of Nvidia. They stressed this point on AI sovereignty for India. What are your views on this?

I'd say broader technology sovereignty for India. We are a large nation. Twenty-three million kids born every year, that's the number I focus on. We have to think about that, that means we are twice China plus America plus Japan. That's the size of India's talent pool, that's how we should think. That means we should aim to do all of the technologies, because we are the people.

We have to create the jobs for our young people. Working on this technology is a very good job to have. That's on the supply side, that we have the talent pool to spare to this. On the other side, it also builds resilience, that we don't have to be threatened by somebody. Our history tells us that we should not assume that the global backdrop will be always favourable to us, because we have a 1000 year history of this. Which is why we have to have technology sovereignty.

That should be our strategy -- not just AI, but in drones, in medical equipment, advanced biotech, nanotech, etc. We have to have a technology sovereignty strategy.

Can give us an outlook for the SaaS space? Is it going to be yet another year of not-so-great growth?

There's a short term called the funding winter. Long term, consolidation.
From a typical business to enterprise point of view, they don't want to have too many vendors. They don't have to manage 600 SaaS vendors. It's just hard, and then integrate all this. They want the tools to work together. Ideally, they want fewer vendors in the pipeline. That is what is driving it. There is the demand side, too. How many customers, how many vendors would an enterprise want to have? That means inevitably we are going to have consolidation, and that consolidation is a long term backdrop.

The short term funding aside, the consolidation is coming. That's what I predicted. We see smaller acquisitions happening, but we still have a lot of consolidation ahead.

Are you looking at acquisitions?

We may do it, we have not grown through acquisition before. That's because the way that we see backdrop is valuations are too bloated. People are throwing money at it. When sanity returns, we might look at acquisition