Setting Up Data Centers In India: Assessing The Systemic Tailwinds & Possible Beneficiaries

Summary

In a discussion on the future of India's data center industry, Sunil Gupta, co-founder and CEO of Yota Infrastructure, and Harid Kapadia, an analyst at Ilara, explore the impact of AI and government initiatives on data center growth. Gupta highlights the current 1200-megawatt capacity, expecting it to double by 2027 and reach 3000 megawatts by 2030, with AI potentially driving even greater expansion. They also delve into operational capacities, market segments, and the technological evolution of data centers, including the shift towards more efficient cooling systems for handling the heat generated by powerful AI chips.

Takeaways

• 📈 The Indian data center industry has seen significant growth, with a current size of about 1200 megawatts and expected to reach 2000-3000 megawatts by 2027 and 2030, respectively.
• 🌐 The rise of hyperscalers in India, along with the growth of e-commerce, mobile penetration, and cloud adoption, has been a major driver of this expansion.
• 🔌 Yota Infrastructure, co-founded by Sunil Gupta, currently has an operational capacity of over 100 megawatts, with a campus approach that allows for scalability up to 700 megawatts in the future.
• 💡 Both colocation and cloud segments are growing in India, with the collocation market seeing significant expansion due to large capacity builds and leasing space to hyperscale cloud operators or large enterprises.
• 💻 The traditional workloads served by CPU-based servers are evolving with the advent of AI and generative AI, which require training machines with vast amounts of data to make decisions and creations.
• 🚀 The potential of AI to drive data center capacity is immense, with some projections suggesting a doubling or tripling of the current capacity worldwide due to AI alone.
• 🛠️ Data centers of the future will likely be very dense, requiring high power per rack and advanced cooling technologies, moving from air-based to water-based or immersion cooling methods.
• 💡 The shift towards AI and GPU-based processing will necessitate a change in data center design, with a focus on handling the increased power and cooling requirements of these advanced chips.
• 💰 The cost to set up a data center is estimated to be around 60 to 70 crores per megawatt, with potential returns of 12-18% and a payback period that can range from 2.5 to 5 years.
• 🛑 The discussion highlights the dynamic changes in the data center industry, with a potential revolution in how data is stored and processed due to technological advancements.
• 🔍 The interview also touches on the opportunity for companies in the capital goods sector, such as those involved in electrical systems, transformers, and substations, to benefit from the data center boom.

Q & A

What is the current size of India's data center industry?

As of the time of the transcript, the Indian data center industry is approximately 1200 megawatts.

What is the expected growth of data center capacity in India by 2027?

The data center capacity in India is expected to double, reaching about 2,000 megawatts by 2027.

What is the projected data center capacity in India by 2030?

By 2030, the data center capacity in India is expected to grow to around 3,000 megawatts.

What is the potential impact of artificial intelligence on India's data center capacity?

The push from artificial intelligence, particularly generative AI, could lead to a significant increase beyond the 3,000 megawatts projection, possibly even reaching 17,000 megawatts by 2030.

What is Yota Infrastructure's operational capacity in terms of data center size?

Yota Infrastructure has an operational capacity of more than 100 megawatts spread across campuses in Delhi and Mumbai.

What is the business model of Yota Infrastructure?

Yota Infrastructure operates on a campus approach, offering wholesale and retail colocation services, as well as building its own sovereign clouds and providing managed services, including GPU as a service.

What is the cost to set up a data center in India?

The cost to set up a data center in India is estimated to be around 60 to 70 crores per megawatt.

What is the expected payback period for a data center investment?

The payback period can vary from as early as 2.5 to 3 years if an anchor customer is secured before the data center goes live, to 4 or 5 years for more speculative investments.

What are the potential returns on investment for a data center?

The returns on investment for a data center are typically around $80 to $100 per kilowatt per month, which can provide a reasonable return of around 12 to 18%.

How will the advent of AI and generative AI affect traditional data center workloads?

AI and generative AI will lead to a significant increase in data center capacity needs, as these technologies require training with large amounts of data and more powerful processing capabilities.

What changes are expected in data center design to accommodate AI and generative AI technologies?

Data centers will need to become denser, with higher power per rack and different cooling technologies, possibly including water-based cooling or immersion cooling for chips, to handle the increased heat generated by AI and generative AI workloads.

Outlines

00:00

📈 India's Data Center Growth and Future Projections

The first paragraph discusses the rapid growth of India's data center industry, which has expanded from 200-300 megawatts to approximately 1200 megawatts due to factors like the entry of hyperscalers, the rise of e-commerce, mobile penetration, and cloud adoption. Sunil Gupta, co-founder and CEO of Yota Infrastructure, provides insights into the current size and future expansion plans of the data center business in India. He mentions that the capacity is expected to double to 2000 megawatts by 2027 and reach 3000 megawatts by 2030, with the potential for further growth driven by artificial intelligence and generative AI technologies. Yota Infrastructure's operational capacity is highlighted, with over 100 megawatts spread across campuses in Delhi and Mumbai, ready to scale up to meet future demands.

05:01

💡 Evolution of Data Centers and Technological Advancements

The second paragraph delves into the operational capacity of data centers, focusing on the colocation and cloud segments that are experiencing significant growth. The discussion covers the business models of data center operators, including wholesale and retail colocation, as well as managed services and Sovereign Cloud offerings. The conversation also touches on the technological evolution of data centers, with a shift towards denser facilities requiring higher power per rack and advanced cooling solutions. The potential impact of AI on data center capacity is explored, suggesting that the worldwide capacity for traditional loads could at least double due to AI, leading to a significant increase in demand for data center resources.

10:02

🛠️ Technological Shifts in Data Center Infrastructure

The third paragraph addresses the technological shifts in data center infrastructure, particularly the transition from air-based cooling to more efficient water-based cooling methods. The discussion highlights the challenges of retrofitting existing data centers to accommodate powerful AI chips and the need for new 'AI factories' designed specifically for these advanced technologies. The potential of immersing chips in liquid coolants to extract heat more effectively is mentioned, indicating a significant evolution in data center design and cooling technologies. The conversation also briefly touches on the emerging market for coolants and the potential for this technology to become more established in the industry.

Keywords

💡Data Center

A data center is a facility that houses a large number of servers, storage systems, and network devices. In the context of the video, data centers are critical infrastructure for hosting, managing, and distributing data, especially as India's digital economy grows. The script discusses the expansion and technological advancements in data centers due to the rise of AI and cloud services.

💡Megawatt

A megawatt is a unit of power equal to one million watts. It is used in the script to measure the capacity and size of data centers. The script mentions that India's data center capacity is expected to grow significantly, from the current size to thousands of megawatts by 2030.

💡Hyperscalers

Hyperscalers refer to companies that operate at a massive scale in the cloud computing industry, typically offering services like Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). The script highlights the impact of hyperscalers on the growth of India's data center industry.

💡E-commerce

E-commerce refers to the buying and selling of goods or services using the internet, and the transfer of money and data to execute these transactions. The script mentions the growth of e-commerce as one of the tailwinds driving the demand for data center capacity in India.

💡Cloud Adoption

Cloud adoption refers to the process of businesses and individuals using cloud computing services to store and manage their data and applications. The script discusses how cloud adoption has become a significant factor in the expansion of data centers in India.

💡Co-location

Co-location in the context of data centers means renting space, power, and bandwidth to customers who want to house their servers in a data center. The script mentions co-location as a growing segment in India's data center market, where companies build large capacities and lease out space to hyperscale cloud operators or large enterprises.

💡Artificial Intelligence (AI)

Artificial Intelligence refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. The script discusses the impact of AI, particularly generative AI, on the demand for data center capacity and the need for more advanced data center designs to handle AI workloads.

💡Enterprise Workloads

Enterprise workloads refer to the applications and processes that are critical to a business's operations. The script discusses how the traditional enterprise workloads are being supplemented by AI, leading to an increased demand for data center capacity.

💡GPUs

GPUs, or Graphics Processing Units, are specialized electronic circuits designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer for output to a display. In the script, GPUs are highlighted as key components in AI and generative AI applications, requiring significant power and cooling solutions in data centers.

💡Liquid Cooling

Liquid cooling is a method of cooling that uses a liquid with a high specific heat capacity to absorb heat from a system and then dissipate it. The script discusses the potential shift from air-based cooling to liquid cooling in data centers to handle the heat generated by powerful AI chips more efficiently.

💡AI Factories

AI Factories is a term used in the script to describe the future data centers that are specifically designed to handle the computational demands of AI and generative AI. These facilities will be different from traditional data centers, with advanced cooling technologies and higher power densities to accommodate the needs of AI chips.

Highlights

The rise of AI and the data revolution, along with government support, could drive a significant opportunity in India's domestic Data Center business.

Sunil Gupta, co-founder, MD, and CEO of Yota Infrastructure, and Harid Kapadia, an analyst from ilara, discuss the future of data centers in India.

India's data center capacity is projected to hit 17,000 megawatts by FY2030, according to Jeff's belief.

The Indian data center industry has grown from 200-300 megawatts to approximately 1200 megawatts, with expectations to double by 2027.

The growth in data center size is attributed to the entry of hyperscalers, e-commerce growth, mobile and fiber penetration, and cloud adoption.

Yota Infrastructure has an operational capacity of over 100 megawatts across campuses in Delhi and Mumbai.

Data center industry in India is divided into colocation and cloud services, with the colocation market experiencing significant growth.

Yota Infrastructure offers a hybrid model including wholesale and retail colocation, as well as sovereign cloud and managed services.

The cost to set up a data center is estimated to be around 60 to 70 crores per megawatt.

Payback periods for data center investments can range from 2.5 to 5 years, depending on customer acquisition strategies.

Returns on investment in data centers are expected to be around $80 to $100 per kilowatt per month.

Genset manufacturers, transformers, and capital goods companies are poised to benefit from the data center boom.

Technological advancements like Nvidia's next chip could revolutionize data centers by replacing traditional processors with more powerful and cost-efficient alternatives.

The data center capacity worldwide is expected to at least double due to the adoption of AI and generative AI.

Future data centers will need to be denser and equipped with more advanced cooling technologies to handle the power requirements of AI chips.

Current data centers are exploring retrofitting to accommodate AI chips, but the data centers of the future will be purpose-built to handle these chips more efficiently.

Liquid cooling is an emerging technology that could become the standard for cooling AI chips due to its efficiency in heat extraction.

Transcripts

okay the rise of AI the data Revolution

the government push could all drive a

decadal opportunity in India's domestic

Data Center business to talk more about

it we have Sunil Gupta co-founder MD CEO

at Yota infrastructure and we also have

the analyst of ilara with us harid

kapadia thank you gentlemen both of you

for joining in let me take off from the

last point that my colleague vakshi left

it at Jeff believes the data center

capacity in India will hit 17,000

megawatt by fi20 three uh so the first

question is um you know if you could

just tell us Mr Gupta what is the

current size right now and how much is

already in the work 177,000 may be the

opportunity but right now the you know

the plans are in place to hit what by

you know which

year so yeah good morning everybody so

Indian data center industry which was

just about 200 to 300 megawatt till

about 201314 got a big Philip when

hyperscalers came into India and

suddenly all the other Tailwinds like

grow of e-commerce growth of mobile

penetration uh fiber penetration

everything started growing Big Time

Cloud adoption became a very very big

thing at that time from that time to now

the current Indian data center size has

already become about 1200 megawatt uh it

is expected to become double based on

announced capacities and the capacities

which are uh under construction it is

expected to become about 2,000 mega

award by 2027 and based on the

projections uh by 2030

this capacity is expected to become

about 3,000 megawatt now there is a

Delta between this 3,000 megawatt and

17,000 and that Delta is coming because

of the artificial intelligence

generative artificial intelligence push

which is coming on uh which is coming on

the top of this 3,000 megawatt which was

supposed to be based on Enterprise

workloads and cloud

services uh right Sunil hi morning har

morning uh what is what is the sort of

operational capacity at your company at

toota so we have operational capacity

which is about more than 100 megawatt

right now spread across our you know

campuses in Delhi and Mumbai and uh the

way we did it is we we sort of

anticipated this uh growth which is

going to come up in the next couple of

years for hypers or otherwise and we

took a campus approach so that the same

campus which today possibly have 50

megawatt can be scaled to let's say 700

megawatt in future you know just by

keeping on adding up more data center in

the same campus

So when you say uh so you know in terms

of tency of these data centers what is

is this uh when you say hyper

hyperscalers you mean for cloud

providers right uh yes or or is it more

collocation where I mean a company

builds a Data Center and then kind of

sort of you know rents out the space uh

it's it's so in India which segment will

will will grow and how do you see that

going yeah so essentially both the

segments are growing uh the way we

divide as the industry uh the market is

uh one is co location and second is the

cloud and manage services on the top of

that the industry which is growing very

very big in the last couple of years is

collocation Market where uh you are

building up large capacities you know

and leads out the space with Power and

Cooling and somebody else like a

hyperscale cloud operator or large

Enterprises or let's say gcc's they will

come and they will put their hardware

and software and the essentially the it

in your data center essentially data

center operator acting as Co location

however some of the Legacy players who

have been operating for last 20 years or

so as well as my own company Yota we

adopted a model where uh besides giving

wholesale and Retail Co location we also

build our own Sovereign clouds we are

delivering a whole lot of managed

services and recent for which is known

by Yota is on gpus to give uh GPU on a

as a service model now that is something

which is coming on the top of Co

location so in the same building if you

want to adopt that hybrid model you can

offer wholesale cocation to hyperscale

Cloud operators you can give retail

cocation to Enterprise customers and you

can also have uh a capacity Reserve

which can be used to deliver your

Sovereign Cloud uh uh AIG gpus and other

manage services okay all right hi Sunil

good morning and thanks a lot for

joining in give us uh uh you know a

couple of reference points then the cost

I believe to set up a data center is

around 60 to 70 cres per megawatt would

that be uh a decent number to work with

yes uh in case we take l Center at a

collocation level that is the right

number right okay all right so what is

then the payback period you know what

kind of returns do you expect and you

know by when will it break even if

you're going to be spending 60 to 70

cres or per

megawatt so essentially it's so much

depends on the F Factor also so many

times if you're lucky you are able to

get an anchor customer even before your

data center goes live so your paybacks

become as as early as two and a half to

three years also but in case you are

making a speculative bit and you want to

have uh you know Enterprise customers

will be coming in only after the

building has gone live and they'll be

giving you retail business and not

really a very wholesale business the

payback period can extend to four or

five years uh uh on this investment

around 60 to 70 CR per megawatt of it

load the returns typically are around I

I'm talking in dollar terms around uh

$80 to uh $100 per kilowatt per month so

if we just do the mass this is something

which uh can give you a reasonable

return of around 12 to 18%

C uh hasid com in on the discussion

right vaki told us that there are many

ways to play this entire boom there are

Genet manufacturer Transformers capital

goods which companies according to you

are best place to take

advantage morning uh so on the capital

goods side which is largely related to

do with electrical systems we have seens

we have ABB uh so every product has a

different company to play with so if I

want to play Genet probably com in is

the best play on the Genet side followed

by Kosar engines and there are some

unlisted players as well on the switch

gear side or substation side we have ABB

seens Schneider but part of the

Schneider is also unlisted the listed

entity is largely into medium voltage

switch gears uh that's how and Hitachi

energy on in terms of substation related

work so these are the companies that one

could benefit on Transformers

specifically we have volte Transformers

on the distribution side

uh followed by Crompton ABB and seens so

these are largely mnc's who dominate the

space on the uh Genet on the uh data

center

site uh Sunil you know just one uh sort

of question this so for example I was I

was listening to this interview uh by

the Nvidia CEO Jensen hang right uh and

he was talking about how their next chip

all right is uh is is is going to be is

going to essentially replace data

centers I mean you know all the

processors in a data center because it's

going to be so powerful and when we talk

a chip we usually imagine like a small

little thing is not I mean it's a big

thing it costs uh I think he said about

$200 million but he said it is going to

be so cost efficient that uh you know

it's going to over over his lifetime you

know just the co cost of copper wires in

the data center will be will be more

than uh will be more than this I mean

you know the what the chip will do for

you over over the life of it operation

technologically is it is it are we is it

all

settled or will we see a revolution or

is it only Evolution just wanted to get

a quick sense I think the market is

changing dynamically right now we can

clearly see that the traditional

workload which were there in our

professional and uh I would say personal

life till date the erps or the

e-commerce applications websites email

servers for majority of the use cases

which we have in our lives till date uh

are getting served by CPU based servers

uh before the Advent of AI and more

specifically generative AI now in case

of generative AI you are actually

training the machines with a huge amount

of data so that the machine based on

some algorithm uh you know becomes

artificially intelligent and start

making decisions for you start doing

some Creations for you my Sunil sorry

since we have I'm just my question was

will the will what data centers right

now look at typically will they change

dramatically so a lot of what is riding

as as

opportunity will it all be there or

because the technology itself will

change in terms of how you store data

and you don't need these huge things so

I'm yeah go in terms of the capacity it

is being said by various people that

whatever capacity which worldwide has

been put in by for data centers for

traditional loads will become at least

double which means AI will present as

much more capacity opportunity for data

centers while in India we are talking

about from 3,000 megawatt to I just now

heard from your program that somebody

has also talked about 177,000 but nobody

is able to actually assess that once

this huge number of gpus and a

widespread adoption of AI starts

happening our life what type of growth

data centers we will see but it is very

sure that whatever capacity of data

centers have been built in India or

worldwide you will see at least double

or triple of that capacity coming in

purely because of artificial

intelligence number one number two the

type of data centers we are making today

which are essentially on a per unit of a

rack cabinet you are giving let say 6 to

8 kilow of power you will end up giving

50 Kow 100 Kow or more power per rack

essentially the data center will become

very very dense the technology the

engineering which we are using to give

power to the equipment and also and more

importantly to cool this equipment will

change rically uh just to give you idea

we are using some sort of cooling called

air uh air based cooling which will

basically change to a water-based

cooling where you're taking water right

next to the chip maybe putting water

right on the chip or immersing the chip

into the water itself or some sort of a

liquid show that the heat can be

extracted as early as possible so the

data centers while as in today uh

industry worldwide is trying to retrofit

the current data centers to handle this

chips but the data centers of tomorrow

which Jensen is calling as AI factories

will be very very different which will

be purpose made for the purpose of

handling these chips where the uh the

technology especially the cooling

technologies will be very very different

from what we are using

today U got that and and uh we're

talking about coolants right right is

that is that is that we just we had the

management of Gulf oil lubricants in the

morning and they were telling us that

right now it's very small I mean it's a

new technology it's emerging but uh but

maybe it'll but you're saying it's an

established technology and it's much

more efficient as compared to air

conditioning right that's uh no so so so

for as I said that majority of the data

center the widespread most commercially

used technology still uh air cool air

you know air based cooling there are new

technologies which are emerging where

you are still taking water right close

to the chip possibly putting from a

cooling plate and then doing but you are

not actually immersing the chip into the

water or some of liquid itself the new

technology which you talking about is

where you are dipping the chip into a

liquid itself now that technology is

evolving uh people are using proofs of

concept they doing some poets they are

putting their CPU servers in that but uh

but Nvidia is yet to yet to certify

putting in their gpus into that liquid

so uh the G the chip which is going to

use the maximum amount of power and

which require most amount of cooling is

yet to be certified for those liquids

but I I what I understand from my

discussions with with various players

including Nvidia that uh some sort of uh

you know certifications and and and that

you know the testing work is on uh my

personal feeling is that that is the

technology which will take over because

uh uh you know uh to to cool that amount

of heat generated by gpus the best way

is to actually dip the chips into that

liquid itself so that water can the heat

can be extracted right then and there

all right uh you know a lot more

questions but no time but so we what

we'll do is we'll make this part one of

our discussion on data centers and we'll

have both of you back uh soon again uh

to talk about this evolving new market

not new but uh evolving Market uh suil

and harid great speaking with both of

you here thank you very much for joining

us here on CNBC TV8