NEW DELHI: Google Cloud was a distant third to AWS and Microsoft Azure when Thomas Kurian made the surprise move from Oracle, where he was president, to lead Google’s cloud business in 2018.

Google couldn’t have asked for better.

The Bengaluru boy revitalised the business.

Revenue surged.

In the June quarter, Google Cloud posted a second consecutive profitable quarter, with an operating profit of $395 million and a revenue of over $8 billion.

But now, with Microsoft taking the lead in generative AI, Kurian and his boss, Sundar Pichai, have a new challenge.

Kurian, who is currently on his first visit to India as Google Cloud CEO, spoke about the transformative power of genAI, and Google Cloud’s offerings in the space.

ExcerptsDo you see genAI as a defining moment in technology?There's no question.

It's a defining moment.

The capabilities of these models and what they're able to do, has astonished users in many countries.

It's really a very transformative experience of what a model can do.

We are being bold and responsible, meaning we're making substantially forward-looking product investments.

But there's no question to us that this is a seminal moment in computing.Google and Microsoft are locked in a battle for AI supremacy.

Both of you are trying to put generative AI to work across a variety of areas.

In this, what advantage does Google Cloud bring to the company?As part of Google Cloud, there are three primary things people want from generative AI – infrastructure for people who want to build their own models for genAI, a platform called Vertex for people who want to develop applications with genAI, and third, a set of capability for people who want to use genAI, we call that Duet.

At the infrastructure layer, our differentiation is largely around scale, performance, and efficiency of the infrastructure.

We offer, for example, two types of accelerators, one set from Nvidia the other from Google TPUs (tensor processing units).

Our performance and efficiency are twice as much (as competitors), which essentially means you get double the value for the same amount of spend.

The proof of it is, 50% of every AI-funded startup is a customer of Google Cloud and 70% of every AI unicorn is a customer of our cloud.

To develop applications, Vertex offers a full suite of services, and you also get a choice of over models, you have Google's models, open-source models, models of partners.

That's way more than anyone else.India has a thriving developer base.

With genAI, there’s a rise of AI pair programming (systems generating code).

There are worries this could lead to big job losses.What we have seen is that the AI pair programmers, what we call Duet, are being used to do things that software engineers don't like to do.

As an example, software engineers don't write to like documentation, the AI model can generate documentation.

Software engineers don't like to write unit tests.

Unit tests are tests that they must write, to make sure that their software is quality controlled.

The AI model can generate those tests.

Our AI models are being used for code review.

For instance, I'm a software engineer, I've written a piece of code, before I'm allowed to submit it, I need to have it reviewed by an expert.

That’s a bottleneck.

I may have to wait in line to get my manager or somebody else to review it.

Now the model can do it.

Lastly, Duet is also being used for migrating an old piece of code.

It may have been written in Cobol and I may not have anybody who understands that anymore, and I want to migrate it to a modern language like Java.

Now, we can automatically convert it.

And so, none of these necessarily mean that we are replacing programmers with the model, we are helping programmers go faster with the model.With genAI, is there a FOMO among enterprise customers? Are they rushing in even though revenue models are still just emerging.We are seeing a lot of interest from customers across a whole range of industries.

If you look at banking, people are using it for a variety of reasons.

And think of it as helping banks have a digital research assistant that's doing their research.

We also have them using it for customer service – a virtual agent assisting the human being in answering questions.

It improves speed of customer service; it improves cost efficiency and customer service.

The work we've done with Apollo Hospital, for example, we're helping doctors and nurses find the next best action, which is, I have treated this person, what is the next best thing that this person should do after I prescribe the medication.There are lots of concerns around AI.

What guardrails are you putting in place?We've had a deep commitment to responsible AI for a long time.

I'll give you three examples of things that are very concrete.

The first one, when we introduced our platform Vertex, along with models from us and from third parties, we said your data as a company is your data solely.

You can keep your data, your intellectual property fully private to you.

That's different from what other companies do.

Second, we provide developers with a service that allows them to compare an answer that they got from a model with two additional things.

One is, where was that answer derived from? So that's called citation.

Can I compare this answer with other sources of data to make sure it's factually valid? That's a service called Grounding that we've introduced that allows you to get higher quality answers from models.

In Vertex, we have a capability to allow you to have 18 different kinds of controls that are built in.

For example, you can protect yourself from abusive speech, you can guarantee, for example, the model does not generate abusive language, you can guarantee the tone of the model used in conversation.

And we've put all those controls into the platform so that customers have full control of the models.How will genAI pricing models evolve over time for different markets? Google has unveiled AI tools for enterprise customers at $30 a month in the US.Pricing is relatively simple – you pay for what you use.

So, when you ask the model a question, for every request, you pass in a set of tokens to ask it a certain question.

And then it generates a set of tokens on the output, basically pricing based on the request response.

That is a single price that we announced last week for the US market.

We are looking globally and what their local pricing should be.

We have different pricing for small and medium businesses versus large enterprises.Google’s new large learning model – Palm 2 AI model – has wider data sets, with more capabilities than its predecessor…We’re updating our Palm model once every four months.

So, we are at the second generation called Palm 2, we announced a variety of different things with it.

If you look at our text model, we announced both quality improvements, but also an increase in what we call the token size that the model can take.

This means how many characters can you feed a model.

So, we've increased it by four times from the previous generation.

That’s important when an AI model is summarising a book.

For the image model, we announced Imagine, the new image model will do this substantial improvement in the quality of the images.

We also added a feature called watermark that allows you to tag the image with an indelible setup, which means you can tell whether an image was generated by the model or whether it was actually a photograph.

That's important for areas like copyright.

We also introduced enhancements to our coding model, and answers to our speech model, which is speech to text and text to speech.