> eBook

Enterprise
adoption trends
Examining AI adoption
and its impact on the
data center
Is your data
centre ready for
the growth of AI?

Power.
Cooling.
Capacity.
Growth in AI use impacts them all.

Schneider Electric’s sustainable AI workload solutions

can help you prepare your data centre infrastructure
for upcoming technology innovations - whether you are
retrofitting an existing facility or planning a new data
centre build.

se.com/datacentre
Contents

4 Introduction

5 Data center trends as the digital

world accelerates automation
5
8 AI: Friend or foe for the data center
industry?

10 Setting the table for the new

energy landscape

14 Overcoming the complexities of

HPC deployments

15 Powering industry 4.0 with the

intelligent Edge

18 Industry trailblazers

20 Evolving powertrain design to 24

better support AI workloads

21 What’s next for data center

cooling in the era of generative
AI?

24 Three ways for colocation data

centers to cut carbon emissions in 15
the AI world

27 Developing AI infrastructure:
Design considerations for critical
workloads

28 Under the surface: Defending your

data center from lurking threats

32 Q&A with Schneider Electric

21

3 | DCD eBook • datacenterdynamics.com

Introduction

There is no question figuratively? As the backbone

artificial intelligence (AI) is of this new digitized existence
taking the world by storm. in which we find ourselves, it’s
From manufacturing and no longer ‘business as usual’
entertainment, to healthcare for the facilities responsible for
and education, our lives effectively running the world.
are becoming increasingly
automated. But as technology Whether it’s data center power,
continues to proliferate, how cooling, management or design,
is this impacting the complex in this eBook we examine the
infrastructure underpinning it industry trends we’re seeing as
all? a result of the AI boom, taking a
deeper dive into the challenges
With data centers under more and opportunities it presents,
pressure than ever before, how and what sort of changes we can
do they go about handling expect increased AI adoption to
the heat, both literally and impress upon the sector.

4 | DCD eBook • datacenterdynamics.com

Accelerating
automation
Data center trends as the
digital world accelerates
automation
Steven Carlini,
Schneider Electric

N
ever has there been The difference between AI previously unthinkable pace. But
so much speculation and the last great technology for all the hype, the reality is that
for the advancement wave – the Internet – is for AI to become more powerful,
and permeation of that the Internet was about corresponding computational
AI into everything we do – communicating. It connected power and network speed and
work, learn, play, travel and people – to information and capacity must be deployed.
even into the decarbonization to technology that leveraged
of our industries. As AI moves content. The AI wave is more However, I am seeing many
from science fiction to reality, about machines taking action – AI pundits claiming that the
a significant transformation creating, predicting, automating data center architecture needed
of systems and processes in and optimizing. is widespread and available.
all industries will occur in this Sorry to disappoint you but
latest, and potentially great, We are embarking on a step that simply is not the case.
technology wave. change in what’s possible at a For the step change to occur,

5 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

shortage of utility power is only

“Governments around the world need to figure out a expected during peak use times.
better way to bring more distributed renewable power
Going forward, I see data
onto the grid.”
center operators collaborating
Steven Carlini more with their local utilities.
Schneider Electric Large data centers have
multiple utility feeds and
long backup power capacity.
By cooperating with electric
substantial, dedicated data Germany are also requiring utilities, data center operators
center capacity will be needed a certain amount of heat can offload a significant
with much more at the Edge – that is generated be reused portion of demand during
closer to the users and data. constructively. high-demand electric heat or
cooling times.
Many trends, both macro and More collaboration with
micro, are affecting the data local utilities More distributed
center industry and the ability to renewable power
support AI as the world evolves. One of the top concerns of
Let’s take a look at some of them. adding data center capacity Governments around the world
is the perceived impending need to figure out a better
Sustainable operations shortage of utility power. With way to bring more distributed
data center capacity seemingly renewable power onto the grid.
Sustainable operations are growing at a much faster rate A main barrier is that most
a core requirement for new than utility power is added, countries use processes and
capacity. To secure permitting many renewable projects have applications designed for one
for data centers, many countries been delayed due to supply or two grid connections per
or local jurisdictions require chain, legal, cost and capacity year and now the requests are
that data center designs be as of installation equipment issues. for hundreds or thousands
efficient as possible, use water However, adding data center of distributed, renewable
conservatively and operate capacity does not have to be a connections resulting in long
off renewable utility sources. one-to-one relationship with connection queues (many
Restrictive countries like adding utility power. A potential years.)

6 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

Once we have this renewable can be stored by data center Edge computing will see
power on the grid, most operators and not curtailed. a boost from the need for
data center operators will This renewable energy storage AI models close to the user
have to adjust to the reality can be used vs carbon- and data called Edge AI.
that renewable power has emitting generators. We’re currently experiencing
intermittent capacity related an increase in workflow
to when the wind is blowing Proliferation of Edge automation, where AI models
or the sun is shining. The computing and liquid at the Edge will work to
utilities will try to normalize cooling make existing systems and
this concern with grid storage processes more efficient for
batteries. Finally, we have the proliferation many industries – transport,
of Edge computing and liquid manufacturing and medical for
However, critical facilities cooling. I believe we’re now in example.
like data centers will need a time where real catalysts exist
electricity service levels that for both. For liquid cooling, Data centers are
are much more stable. To there will be servers with up the enablers for AI’s
satisfy this requirement, I to 16 GPUs, 2 CPUs and many promise
predict data center operators DPUs for AI applications. Server
will start adding multiple hours manufacturers are installing The promise of AI is well
of on-site energy storage. This input and output piping for documented and data center
onsite storage will benefit not liquid because these servers capacity is indeed the enabler
only in terms of resilience, but simply generate too much heat of that promise. Will data
also at times when there is a to be efficiently rejected by fans centers rise to meet the AI
surplus renewable supply that and air. challenge? I believe they will.

7 | DCD eBook • datacenterdynamics.com

AI: Friend or
foe for the
data center
industry?
The European data center
industry finds itself at an
intriguing juncture Thierry Chamayou
Schneider Electric

A
s of September 2024, about the implications of key driver. It is dramatically
our industry will have AI on our industry, in terms changing the entire IT
to comply with strict of how can we continue to landscape.”
and specific new EU ensure power availability for
energy-efficiency regulations, increasingly energy-hungry The speed in proliferation of
while simultaneously having data centers, while at the same AI adoption is quite staggering
to find new ways to cope time keeping the industry too. To give you an idea on
with an increase in energy on track with new energy- the scale of growth in AI, the
demands being driven by the AI efficiency targets. market is set to reach $407
revolution which is now in full billion by 2027 (approx. €382
swing. The game-changer that is billion), with expected annual
AI growth of 37.3 percent between
Kickstart: The networking now and 2030.
event that examines the A quick read of our recent White
digital landscape Paper 110 and you’ll see just Because data centers serve
how profoundly AI is disrupting as the critical infrastructure
It’s this conundrum that I’ve the way data centers have been supporting the AI ecosystem,
been reflecting on, even long traditionally designed, built, AI is completely reliant on
before I attended Kickstart and managed. As Andreas data centers. But the flip side
Europe 2024 in Amsterdam this Herden, senior vice-president to this is, data centers are now
February. At this year’s event, sales Germany & Continental relying on AI and data analytics
it was inevitable that AI would Europe for Green Mountain to help power themselves in a
be the hot topic dominating put it so succinctly at Kickstart, more effective way. And this
discussions, and I was intrigued “We have so many requests brings us neatly to the topic of
to hear from my fellow peers for AI deployments. AI is the sustainability.

8 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

The road to net-zero

“Data centers are already leveraging AI-powered
As I’ve touched on, new energy- automation, data analytics and machine learning, to
efficiency regulations such as make gains in energy efficiency and decarbonization.”
the European Commission’s
Energy Efficiency Directive Thierry Chamayou
Schneider Electric
(EED) are about to come into
force, which will require
data center companies to
annually report on their energy explained to us in Amsterdam, surge in AI, “You want to build
performance. “We have a commitment as efficiently as possible, you
to 24/7 carbon free energy, want to use storage to enable
And while the EED’s stated and the industry is already more renewables on the grid,
goal of reducing energy acknowledging the progress and you want utilities driving
consumption in Europe by we’re making in response to our more capacity for renewables.”
11.7 percent by 2030 sounds climate-neutral commitments.”
ambitious, it’s encouraging All of this industry feedback There’s no doubt then, AI is
to know that our industry is tells me that climate neutrality here for the long haul, and it’s
working hand-in-hand with for our industry can be forcing our industry to find
the EU to make this target achieved, but only through new ways of driving data center
attainable. ongoing dialogue and in energy efficiency. On a personal
close partnership with our level, I’m confident we’ll succeed
As Lex Coors, president of customers, partners, right up in this, because traditionally
the European Data Centre to government level. In fact, our industry has been strong at
Association revealed, “In the industry partnerships are more working together to come up
last few years, the EED has had important than ever for our net- with innovative solutions that
an open consultation with our zero ambitions, now that AI is so mitigate new challenges.
industry, working with us as obviously here to stay. What’s more, I’m proud, not only
trusted advisors.” to work for a company that’s
Accelerating new best taking the lead in sustainable
Leveraging the power of practices best practices, but also to
AI represent an industry that’s
I believe AI can act as a positive working together to provide
So, given the acceleration in AI accelerator for our data center sustainability blueprints that
and its insatiable demand for industry, propelling us faster to can help other energy intensive
data center computing power, where we need to get. As my industries to decarbonize.
how is it going to be possible to Schneider Electric colleague
achieve a net-zero data center Jim Simonelli, CTO for data Learn more about Kickstart
industry? Well, data centers are centers put it, to cope with the Europe here.
already leveraging AI-powered
automation, data analytics and
machine learning, to make
gains in energy efficiency and
decarbonization. AI-based
cooling and e-waste recycling
are two more examples that
highlight how our industry is
tapping into the power of AI to
drive sustainability.

Alongside AI, renewable

energy is another central theme
that was under the microscope
at Kickstart 2024. As Martjin
van Wijngaarden, global
energy transaction manager
for Iron Mountain Data Centers

9 | DCD eBook • datacenterdynamics.com

Setting the
table for the
new energy
landscape
As high-performance
computing and the
proliferation of AI transforms
data centers, preparing for
the new energy landscape
involves strategic planning
and innovative solutions, Farah Johnson-May

from modular architectures DCD

to advanced cooling

H
igh-performance HPC deployments and provide Evaluating readiness for
computing (HPC) insights on how to overcome high-density computing
is advancing them.
multiple sectors, but As we set the table for the
integrating and operating these Vance Peterson, global future of data centers, it’s
technologies in data centers solutions architect at Schneider crucial to prepare every
presents significant challenges, Electric, brings over 20 years element carefully to support
requiring strategic and of experience in the mission- high-density computing
innovative solutions. This raises critical industry. At Schneider, workloads. This involves a
the question: What are the real- he collaborates with industry comprehensive assessment
world challenges in managing leaders to help customers of existing infrastructure,
advanced compute systems? address their most complex covering everything from
challenges, particularly as the utility service connections
June 2024 saw DCD team up data center sector is driven to the detailed components
with Schneider Electric, a global by digitization and the need within the data center, such
leader in energy management to enhance sustainability as electrical, mechanical, UPS,
and automation solutions, to while addressing capacity cooling, and power distribution
explore the complexities of constraints. systems.

10 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

requirements related to AI
and HPC implementations

5) As the final step before

broad adoption, pilot a small
project: Start with a small-
scale project to test and
validate the use case, the use
of AI and HPC in real-world
scenarios, and use that pilot
to validate the economics or
return on investment against
projections.

“For example, direct-to-chip

liquid cooling or immersion
cooling can address increasing
densities associated with
accelerated compute. This
Just as a well-prepared host organization is from a systems may present some skill gaps in
ensures every dish is perfectly and technology standpoint. an organization, and as such,
paired, Schneider provides tools In other words, take a good operators and maintenance staff
and expertise to guide these hard look at the organization’s will need to scale up to support
evaluations: technology infrastructure – its the needs of HPC.”
data capabilities and its skill sets.”
“Schneider can offer support Accountability: Many
and guidance from our agnostic Peterson goes on to describe hands make light work
trade-off tools that are available the foundational steps that an
on the public website. We enterprise might want to take to Schneider Electric actively
can help clients in a number prepare for high performance shares its expertise and fosters
of ways: conducting detailed compute: partnerships to support the
power audits, PUE effectiveness design of HPC and AI-ready
assessments, load and capacity 1) Define business objectives: operations. As Peterson notes,
analysis, thermal imaging, and Clearly outline the goals and the complexity of HPC presents
computational fluid dynamics expected outcomes from the challenges that the industry is
simulations to understand their adoption of HPC and AI collectively working to address:
current capabilities, and also
forecast within a digital twin 2) Assess data readiness: “For instance, Schneider
environment, what their reality Ensure data quality and has partnered with Nvidia to
might look like the day after availability, plus security combine our expertise in energy
tomorrow.” in place to support the management and digitization.
applications Together, we are developing
Foundational steps for optimized data center reference
adopting HPC 3) Invest in skill designs that will pave the way for
development: Identify gaps high-performance computing
Preparing for HPC and AI in training investments, and and accelerated advancements
adoption involves several hire talent specifically with in Edge artificial intelligence and
foundational steps to ensure expertise in AI and HPC digital twin technologies.”
that organizations are not technologies
only equipped with the This collaboration is not just
right technology but also 4) Conduct infrastructure about meeting current needs,
have the necessary skills planning: Evaluate and but also about setting new
and infrastructure. Peterson plan for the necessary industry standards:
explains: computational infrastructure
and resources to support “These reference designs will
“Clients really need to workloads, and address any help redefine the benchmarks
consider where exactly their regulatory or compliance for AI and operations within

11 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

modular, scalable architectures

“Schneider has partnered with Nvidia to combine our is crucial for navigating the
expertise in energy management and digitization. increasing densification of
Together, we are developing optimized data center data center environments.
infrastructure that will pave the way for high-performance This approach benefits both
purpose-built facilities and
computing and accelerated advancements in Edge
retrofitted infrastructures,
artificial intelligence and digital twin technologies.”
ensuring readiness for future
Vance Peterson demands. According to
Schneider Electric Peterson, several best
practices are critical in this
context:
the data center, and it marks with the tools and the resources
a significant milestone in our they need to integrate new and - Capacity planning:
industry’s evolution, with AI evolving AI solutions into their Thoroughly assess current
applications gaining traction infrastructure, allowing them and future power and cooling
across industries.” to enhance deployment and requirements to determine the
efficiency and ensuring reliable necessary capacity upgrades
Peterson emphasizes lifecycle operations.”
that the goal of this - Advanced cooling solutions:
partnership: The proof is in the Integrate liquid cooling
pudding technologies to efficiently
“Through this collaboration, dissipate the heat generated
we really want to provide data As we prepare for the new by high-density computing
center owners and operators energy landscape, adopting elements

12 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

- Power system upgrades: Install The ability to elevate accurately predict how best to
additional power feeds, busways, infrastructure management not only modify, but optimize
or upgraded PDUs to ensure from a previously reactive to their architectures going
reliable power delivery at higher proactive approach, means forward.
voltages and power densities that the value of real-time
architectural insights cannot be “The evolution and
- Energy efficiency measures: understated. widespread adoption of HPC
Utilize variable speed drives, will likely make advanced
hot/cold aisle containment, and As the old adage goes, you monitoring and management a
airflow optimization to enhance can’t manage what you don’t must-have rather than a nice-
overall efficiency and reduce monitor, and the unique – to-have. Electrification and
cooling loads and not to mention dense – decarbonization have increased
nature of HPC environments electrical demand, and this,
- Modular architecture: Deploy makes the implementation along with the increased
scalable, modular components of advanced monitoring demand driven by digitization,
that can be easily expanded an essential tool in the is really setting the table, if
to meet increasing power and effective maintenance and you will, for the new energy
cooling demands management of this complex landscape.”
infrastructure.
- Advanced monitoring and To find out more, check
management: Implement From power consumption out the full DCD>Broadcast
systems to continuously track to environmental constraints, over on the next page, or visit
power usage, temperatures, and by leveraging digital twin ‘Transitioning to AI ready
humidity, enabling proactive and other advanced metering data centers’ on the Schneider
maintenance and optimization. technologies, operators can website.

13 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

Overcoming the
complexities of
HPC deployments
with Vance Peterson,
Schneider Electric

CLICK
TO
WATCH

14 | DCD eBook • datacenterdynamics.com

Powering
Industry
4.0 with the
intelligent Edge
Why companies are starting to
realize the value of the Edge
Shahid Ahmed
NTT Data

T
he advent of AI-based of AI applications in the Amidst this challenge lies a
solutions is fueling a enterprise. Low latency between solution, the intelligent Edge or
data-driven landscape applications at the Edge and Edge AI, forming a critical pillar
where the demand for the cloud results in slow of Industry 4.0 that is poised
faster processing has reached computations – this is a critical to revolutionize industrial
a tipping point, placing a challenge when use cases need automation and enhance real-
significant strain on traditional immediate action, e.g. safety use time, data-driven insights.
IT infrastructure. case inside an airport hangar.
Enterprises also face increased Edge computing and Edge
These devices – deployed costs from backhauling large AI
everywhere from factories amounts of data both from
to hospitals to smart cities – an infrastructure and cost- Edge computing, with its
capture real-time data at an per-gigabyte of storag. This localized processing power
unprecedented scale, while is especially true for the and storage, offers a practical
powering AI-based applications manufacturing and industrial approach to overcoming the
including intelligent sectors. limitations of cloud architecture.
machine vision, digital twins,
autonomous vehicles, and more.
And it’s not slowing down any
time soon. IDC estimates there
“With 70 percent of businesses actively adopting Edge
will be approximately 41.6 billion technologies, enterprises are recognizing the potential
IoT devices, generating up to to drive Industry 4.0 transformation while enhancing
79.4 zettabytes of data, by 2025. resiliency in their operations.”

As data creation continues

Shahid Ahmed
NTT Data
to surge, corporate networks
need help managing all kinds

15 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

It brings processing power and Businesses see the promise connectivity is a challenge for
storage closer to data sources of Edge and leverage IoT most in the manufacturing
on local devices, enabling and private 5G technologies vertical.
enhanced security. to drive efficiencies across
various sectors through Driving efficiencies
Edge AI – or smaller and automation. With 70 percent with private 5G and IoT
more lightweight, bespoke of businesses actively technologies
AI models deployed on Edge adopting Edge technologies,
devices – capitalizes on Edge enterprises are recognizing the The advantages of intelligent
computing capabilities by potential to drive Industry 4.0 Edge are powered by fast,
enabling these new specialized transformation while enhancing reliable, and secure connectivity
devices to process vast resiliency in their operations. solutions such as private 5G
amounts of data from multiple networks. These networks
devices and endpoints. This Organizations are already provide a robust and secure
allows them to all speak seeking out these Edge infrastructure for connecting
the same language which solutions, including leveraging IoT devices, ensuring seamless
enables ‘Actionable AI’ in the modular data centers on-site communication and data
given environment and thus to bring data management, transfer between Edge devices
faster, more autonomous processing, and collection to and central systems. These
decision-making. challenging and remote sites technologies support the
where reliable, high bandwidth much-anticipated revolution in
For instance, intelligent
machine vision systems in
manufacturing can analyze “Edge computing, with its localized processing power
and react to visual data on-site, and storage, offers a practical approach to overcoming
reducing the lag that would the limitations of cloud architecture.”
occur if data had to be sent
to and processed in the cloud Shahid Ahmed
from multiple devices across NTT Data
multiple systems.

16 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

industrial automation and data-

driven decision-making. “While motivations differ, real-time data access,
automation/AI, operational efficiency, and IoT device
Combining these management have been identified as the top drivers
technologies offers several behind Edge investment.”
key advantages for today’s
increasingly distributed Shahid Ahmed
enterprises: NTT Data

- Enhanced security:
Centralized data storage identified as the top drivers data demands of the latest AI-
presents a single point of failure, behind Edge investment. powered innovations isn’t a one-
making it a prime target for person job. What’s clear is that AI
cyberattacks. Edge computing The compatibility challenge is driving the demand for Edge
allows for local data processing, technologies.
minimizing the attack surface Despite the benefits, the sheer
and reducing the potential for variety of devices, protocols, To meet this demand,
large-scale data breaches. and platforms in the IoT organizations will need to
landscape creates a significant collaborate internally between
- Actionable AI: Combining compatibility hurdle. Businesses IT and business teams, and
Edge computing with Edge AI are hesitant to invest in Edge externally with managed service
allows for faster data processing solutions that might not providers (MSPs) who can
and Actionable AI, improving integrate seamlessly with their help navigate legacy systems
performance in critical current infrastructure or future and protocols. Leveraging the
applications, where identifying deployments. knowledge of MSPs will be
and addressing equipment integral to finding the most
issues early can prevent costly However, to truly realize the efficient and effective ways for
downtime. promise of Edge computing an enterprise to deploy and
and Edge AI, enterprises need leverage Edge computing.
- Increased resiliency: Edge to modernize their networks.
computing systems often Reliable connections and low- By embracing the intelligent
operate with some degree of latency environments are Edge, businesses can unlock
autonomy, making them less essential for data movement a myriad of benefits from
susceptible to network outages and real-time data processing operational efficiency to real-
that can cripple centralized required by AI-based solutions. time Actionable AI – the
systems. perfect foundation for agile and
What’s clear is that the adaptable operations.
- Improved safety: Safety is promise of Edge will drive
critical to real-time applications companies to improve their As more enterprises look
at the Edge. Leveraging network infrastructure so that to adopt the latest Edge
these technologies unlocks they can take advantage of technologies, this foundation
new solutions to identify local, real-time compute and will be critical to ensuring
unauthorized users in unsafe solutions. In fact, almost 40 seamless data processing,
environments. percent of enterprises planning scalability, and the ability to
Edge deployments acknowledge adapt to evolving business goals,
Early adopters in sectors like a need to upgrade their network but this demands orchestration
manufacturing, transportation to support the expected spike across IT and business
and logistics, and healthcare, in connected devices and the functions.
are already reaping the rewards. demands of new applications.
They’re focusing on applications Keep in mind that going in
that improve employee safety, Building a successful, alone on the journey can prevent
customer experience, and actionable AI journey enterprises from realizing the
operational efficiency. While full potential of Edge. Embrace
motivations differ, real-time Successful Edge deployments collaboration, harness the power
data access, automation/AI, drive businesses to treat it of the intelligent Edge, and
operational efficiency, and IoT as an integral part of their watch Industry 4.0 aspirations
device management have been business strategy. Meeting the transform into reality.

17 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

> Industry trailblazers

Enterprise adoption trends

in the wake of AI

Hyperscalers account for 41% of

worldwide data center capacity

Hyperscalers now account for 41 percent of

worldwide data center capacity, according
to data from Synergy Research Group
Comparatively, non-hyperscaler colocation
capacity accounts for another 22 percent,
while on-premise data centers make up the
remaining 37 percent.
Just six years ago, 60 percent of data center
capacity was on-premise. Synergy predicts
that this change will only accelerate, with
hyperscalers expected to account for more than
60 percent of capacity by 2029 and on-premise
dropping to just 20 percent.

Schneider Electric collaborates with Nvidia on

AI data center design

Schneider Electric will leverage its expertise

in data center infrastructure and Nvidia’s Hyperscale data
advanced AI technologies to introduce the center capacity by
first publicly available AI data center reference Country/Region
designs. These designs are set to redefine the
benchmarks for AI deployment and operation
within data center ecosystems, marking
a significant milestone in the industry’s
evolution.
JLL: Edge data center market to surpass $300bn in
The rise of AI has spurred notable
transformations and complexities in data The global Edge data center market will hit a valuatio
center design and operation, with data center consulting firm JLL. JLL notes that a surge in data-de
operators working to swiftly construct and will drive the market growth, which can be supported
operate energy-stable facilities that are both The report cites the demand for technology that r
energy-efficient and scalable. for the projected growth in value of the market.

18 | DCD eBook • datacenterdynamics.com

 >> DCD eBook | Enterprise adoption trends

NTT Data and Schneider Electric join AWS, Digital Realty, Google,
forces to drive AI innovation at the Meta, Microsoft & Schneider
Edge Electric call for supplier
action
NTT and Schneider Electric have unveiled a
first-of-a-kind co-innovation that empowers The governing body of the
enterprises to harness the power of Edge iMasons Climate Accord, a
computing. The strategic partnership program of Infrastructure
introduces a unique solution that seamlessly Masons, is calling on all
integrates Edge, private 5G, IoT, and modular suppliers serving data centers
data centers, providing unparalleled to support greater transparency
connectivity and supports the computational in Scope 3 emissions as part of
demands of generative AI applications broader efforts to reduce the
deployed at the Edge. industry’s carbon footprint.
Consisting of AWS, Digital
Realty, Google, Meta, Microsoft,
and Schneider Electric, the
governing body released an
open letter that explains the
importance of widespread
adoption of environmental
product declarations (EPDs),
which are standardized, third-
party-verified documents
reporting the embodied
emissions of a product.

Data center companies

confirmed for UK
International Investment
Summit

Data center investors and

operators are among the first
attendees confirmed for a
UK government investment
summit as new chancellor
Rachel Reeves looks to attract
more digital infrastructure
providers to Britain. Reeves
flew out to New York to meet
with companies including data
center provider CyrusOne and
investment fund Blackstone.
The new government is
keen to see more data centers
value by 2026 built to help grow the UK’s
digital economy. Government
on of $317 billion by 2026, according to a report from real estate estimates suggest the UK data
ependent technologies, including AI and the Internet of Things (IoT), center sector employs 17,000
d by Edge data centers. people and generates £4.6 billion
requires low latency, high computing, and generative AI as reasons ($5.86 billion) in revenue each
year.

19 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

Evolving
powertrain design
to better support
AI workloads
with Stuart Sheehan,
Schneider Electric

CLICK
TO
WATCH

20 | DCD eBook • datacenterdynamics.com

What’s next
for data center
cooling in the era
of generative AI?
The expectation from some
is that soon traditional
air-cooled data center
technologies may become
obsolete, especially as AI
adoption and supercomputing Paul Finch
Kao Data

advancements gathers pace

T
he era of generative AI will require a plethora of more requests for capacity related
is well and truly upon energy-efficient data center to artificial intelligence (AI)
us. According to JLL, designs. requirements, with most of
it is among the top these expected to come from
three technologies expected to From an industry service providers and AI start-
have the biggest impact on real perspective, the numbers are ups, as opposed to members
estate, having reached record indeed staggering. Analysts at of the hyperscale and cloud
investments of up to $4bn in AI- TD Cowen have stated the AI communities.
powered property technologies wave has led to approximately
(PropTech) in 2022. 2.1GW of US data center Now, as AI descends into all
leases, while CBRE’s European aspects of technology products,
JLL’s 2024 report also found Real Estate Market Outlook services, and solutions, many
that AI energy demands – 2024 found that data center are asking if the data center
ranging from 300-500MW+ – providers will see an uptick in industry is truly ready to

21 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

accommodate its requirements.

The answer, for many operators, “The expectation from some is that soon, traditional
is no. air-cooled data center technologies may become
obsolete, especially as AI adoption and supercomputing
Cooling the AI workloads advancements gather pace.”
of the future
Paul Finch
Today Nvidia, the world’s Kao Data
leading authority on high-
performance computing (HPC)
and AI, is estimated to be more than 40 watts per square The future of data center
responsible for more than 95 centimeter will face significant cooling
percent of machine learning challenges in the future.
workloads and remains the The discussion around cooling
dominant manufacturer of This is no wonder, with methodologies remains
GPU-accelerated technologies. the latest Nvidia SuperPOD one of the most divisive
packing up to eight H100 GPUs conversations within the
Last year, the company per system, and all connected industry. In one camp are
shared news it had won a $5 by Nvidia NVLink. Each DGX those who advocate for air-
million grant to rearchitect H100 is expected to provide cooled systems and recognize
the data center landscape up to 32 petaflops of AI the benefits of free air cooling
and build an advanced liquid performance, around six times over that of a liquid-cooled
cooling system to address more than its predecessor the approach – one which often
many of the challenges that DGX A100, which was already requires a major investment
legacy data centers, including placing limits on traditional in capital expenditure, a
on-premises, enterprise, and data center capabilities. retrofit of a legacy data center
older colocation facilities, are architecture.
facing. To add further context from a
design and energy standpoint, In the other are the owners
Funded by the US an Nvidia SuperPOD can and operators already
Department of Energy, the include up to 32 DGX H100 undertaking proof of concept
COOLERCHIPS program has systems with the associated (POC) projects and deploying
been positioned as one of the InfiniBand connectivity hybrid environments – those
most ambitious projects the infrastructure, drawing up to who are developing high-
industry has ever seen, and at a 40.8kW of power per rack. By performance infrastructure
time when processor heat and today’s standards, that might be systems precision-engineered
power capabilities are soaring considered an obscene amount to accommodate compute-
as Moore’s Law and data center of processing power and AI intensive applications on an
designs reach their physical capability, but rack and power industrial scale.
limits. densities are only expected to
increase. The benefits of liquid
The expectation from cooling
some is that soon, traditional Interestingly, Nvidia’s new
air-cooled data center Blackwell GPUs are set to With rack densities now
technologies may become enable businesses to build and expected to surpass 100kW, it’s
obsolete, especially as AI run real-time Generative AI clear that liquid cooling will
adoption and supercomputing applications and large language become increasingly popular.
advancements gather models at up to 25× less cost For those embracing the
pace, and that Nvidia’s and energy consumption than technology, the benefits are
cooling system could cost its predecessor, paving a new significant. Many of today’s
approximately five percent less path for data centers engineered liquid cooling solutions
and run up to 20 percent more for AI. The question remains, leverage the higher thermal
efficiently than air-cooled how will data centers need to transfer properties of water
approaches. It also expects evolve to accommodate the and other fluids to cool high-
that cooling technologies cooling requirements of AI, and density racks more efficiently
may begin to reach their which organizations will be the and effectively than legacy
limitations, as heat loads of winners in the race? measures.

22 | DCD eBook • datacenterdynamics.com

 Approaches such as this are therefore able to support greater toward liquid cooling as the
also reinforced by studies from CPU and GPU densities while only option for future GPU-
organizations such as Iceotope offering significant heat re-use powered compute.
and Meta, which confirmed capabilities.
the practicality, efficiency,
and effectiveness of precision Organizations within the
liquid cooling in meeting sector have also predicted
the cooling requirements of that liquid cooling can be up
hyperscalers, where liquid to 3,000 times more effective
cooling has gained something than using air, all of which
of a bias among members of points towards liquid cooling
the community. having the potential to
become the preferred cooling
With DTC liquid cooling, architecture of the future and
between 70-75 percent of the something that will be vital to
heat generated by the rack meet data center sustainability
equipment is removed via requirements.
water with the remaining 25-
30 percent removed via air. The future for generative AI
Because direct-to-chip cooling is both exciting and unknown,
is more effective from a heat but if Moore’s Law is reaching
transfer perspective, it is its physical limits, all roads lead

23 | DCD eBook • datacenterdynamics.com

Three ways
for colocation
data centers
to cut carbon
emissions in
the AI world
Exploring what sustainable
decision-making for data
centers means with a focus
on the transformative power
of carbon accounting and Carsten Baumann

circularity Schneider Electric

A
here is no question today without addressing Aligned’s chief innovation and
that wholesale data sustainability initiatives. Cloud technology officer, to discuss
centers, like hyperscale and colocation providers are ways data center providers can
ones, are facing rising taking note of technological reduce carbon emissions while
capacity demand while trying advancements that enable them meeting growing capacity
to operate more sustainably. to measure and curtail carbon demand from wholesalers and
This balancing act has become emissions, water consumption, enterprise tenants.
even more challenging since AI and other environmental waste.
has taken center stage across Indeed, a holistic sustainability In our Innovation Talk,
all industries, requiring greater strategy is now inextricable from “Empowering wholesale
and greater computing power. the business strategy at large. colocation providers through
In fact, data centers output four carbon accounting and
percent of the world’s carbon One innovator forging circularity,” we explore what
emissions. the path ahead for the sustainable decision-making
colocation provider market for data centers means,
We all know that data is Aligned Data Centers. I focusing on the transformative
centers, considered “energy had the opportunity to meet power of carbon accounting
hogs” by some, cannot operate with Phill Lawson-Shanks, and circularity.

24 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

A pioneer in sustainable data heat, steam, or cooling), the real this challenge by using unique
center design and operations, challenge lies with accounting identifiers to tag and account
Aligned Data Centers, which for Scope 3 emissions. for every piece of equipment in
caters to wholesale and the data center throughout the
enterprise clients that need That’s because these are facility’s life cycle.
sustainable data center capacity, the embodied emissions in
delivers leading PUE, WUE, and all building materials, such as This embodied carbon
CUE metrics across six million concrete and steel, and the tracking system gives clearer
sqft and 1.3GW of power. manufactured equipment inside visibility of embodied carbon
a data center. across the data center – from
Here are three key takeaways steel trusses to construct the
from our conversation: Most cloud and colocation facility to electrical and cooling
providers struggle with a infrastructure sources to
1. Start with carbon accounting lack of reliable supplier data, equipment materials. Selecting
quantitative tools to measure and sourcing the right materials
We will say this time and time Scope 3 emissions, and a clear and equipment is imperative
again at Schneider: You can’t accounting and reporting from the beginning.
measure what you can’t see. methodology. As Phill points
Accurate carbon accounting out about the building and Additionally, scrutinizing
gives you the transparency you construction phase: other key embodied carbon
need for reporting throughout contributors to data center
the entire data center lifecycle. “The hardest thing to do operations is an essential step
This information is vital for is look at discrete items and in reporting and ultimately
enabling colocation providers to determine their embodied mitigating Scope 3 emissions.
communicate carbon emissions carbon footprint.”
across the supply chain. Tenants This “Recommended
who are working toward Accordingly, as operators inventory for data center
sustainability goals can leverage move toward renewable Scope 3 GHG emissions
this information and map it energy sources and efficient reporting,” is helpful for carbon
against their own sustainability technologies for operating the accounting and reporting.
KPIs. It also allows them to data centers, Scope 3 emissions
select one data center provider continue to be the largest source Aligned Data Centers takes its
over another. of data center GHGs. carbon reduction commitments
a step further by actively
While operators can more At Schneider, “We believe investigating non-concrete
readily identify and measure understanding and reporting and alternative materials in
Scope 1 emissions (those from on Scope 3 emissions is the its buildings and innovating
direct operations) and Scope next frontier for data center cooling solutions that cut down
2 emissions (those associated industry key performance GHG emissions even more
with the purchase of electricity, indicators (KPIs).” Aligned solves during operation.

Schneider Electric

25 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

components are placed in

“Working in partnership is integral to empowering the middle of the racks. That
cloud and colocation providers’ stakeholders to solve means that with 1MW of
the elimination of data center carbon emissions closer power, a traditional approach
to cooling would need to
to zero, today.”
spread the load over 260 racks.
Carsten Baumann
Schneider Electric Today, by using liquid
cooling technology, Aligned
is able to increase the power
density to 20 racks for the
2. Focus on opportunities for accelerating the transition same amount of power. It uses
circularity to a more sustainable future. onsite generation of renewable
The benefits of circularity energy to further reduce GHG
There is a tremendous include improved operational emissions.
opportunity for data center efficiency, cost savings, and
operators to embrace a circular enhanced reputation. Working in partnership is
economy. In addition to integral to empowering cloud
reusing key pieces of electrical 3. Build for the future and colocation providers’
infrastructure and other through partnership stakeholders to solve the
components, this effort could elimination of data center
include expanding take-back, It is fair to assume that if carbon emissions closer to
recycle, and refurbish programs you build a data center zero, today. At Schneider, for
designed to reduce the cloud infrastructure today, the instance, we are proving that
and colocation provider’s building will be there for repeatable architectures for
e-waste. Initiatives to extend 20+ years. So you need to sustainable data centers are
the life of equipment by using make sure what you are faster to deploy.
predictive analytics for better, building today is future-
more accurate maintenance proofed and has sustainability Our driver is to accelerate
also advance a circular outcomes in mind – not just the path to decarbonization
approach. as an afterthought but as key by making sure sustainable
deliverables. data centers built today can
The potential of circularity meet the same performance
principles is promising Aligned Data Centers is characteristics they will
in achieving resource driven by its mission to build need as capacity demands
efficiency, waste reduction, and operate sustainable data skyrocket.
and a more sustainable data centers now and well into the
center ecosystem. As Phill future, placing high priority on This is especially important
explains in the on-demand innovation for carbon-friendly as the collective mindset of
webinar, Aligned keeps track design and operations. As Phill sustainability takes firm hold
of the lifespan and usage of emphasizes: worldwide and as regulatory
equipment and products in its requirements come into play.
facilities. He notes: “It used to just be that just
one person was overseeing Together with cloud and
“We might not be able to sustainability, and they didn’t colocation providers such as
reuse the materials in our own interact with our team very Aligned Data Centers, we can
work, but they could be used in much. Now they are in the design, deploy, and operate
another industry.” room for the decision-making what Phill calls an “ecologically
and have veto rights over invisible data center.” For
For example, lithium-ion contracts.” the sake of our planet,
batteries might not be good which is facing catastrophic
for an EV again, but they could One of the key areas that climate change. It is a shared
be good for another use. He Aligned is innovating in is responsibility we are proud to
sees this circularity effort cooling, which is especially take on with Aligned and the
as the responsibility of both important as chips are getting data center sector overall.
Aligned and the equipment hotter and creating more
vendors, including Schneider, heat. Traditionally, the hottest It’s time to act now.

26 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

Developing AI
infrastructure: Design
considerations for
critical workloads
with Rob Bunger,
Schneider Electric

CLICK
TO
WATCH

27 | DCD eBook • datacenterdynamics.com

 Under the
surface:
Defending
your data
center from
lurking threats

When threats are headed for

your IT infrastructure, stay
calm and manage on with
DCIM and lessons from Jaws Kevin Brown
Schneider Electric

28 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

I
recently revisited the Unlike in Jaws, there is more They are grappling with
thriller, Jaws. The movie than one shark lurking under cyber threats and potential
was released 49 years ago the surface. attacks while trying to make
on June 20, 1975, and the sure firmware is up-to-date and
book 50 years ago. Like many For CIOs, multiple threats that they are following their
others I love this story – the are lurking company’s security policies.
water looks calm, but danger If that wasn’t enough, we
lurks under the surface as a We know CIOs need to argue, they will be increasingly
great white shark makes local improve how they manage questioned on the energy
waters its home, terrorizing a their sprawling, hybrid IT consumption of their IT
beach community. Characters infrastructure and understand infrastructure and may, if global
with different personalities the number of assets they have trends continue, need to report
approach the threat differently, on the network. They need their sustainability metrics.
leading to chaos. to maintain resiliency and
avoid outages that may disrupt Many CIOs are turning
I believe this is a great business continuity, resulting to DCIM (Data Center
analogy to how organizations in lost revenue and potential Infrastructure Management)
may have different reputation damage. software for answers. I believe
‘personalities’ in responding
to the ongoing threats to their
IT infrastructure. These threats “When it comes to the threats facing your IT
are like the shark in Jaws – you infrastructure, are you like Hooper, Quint, or Brody?”
know they are out there, you
generally can’t anticipate them, Kevin Brown
and when they attack, they Schneider Electric
may kill you.

29 | DCD eBook • datacenterdynamics.com

Jaws – the personalities of its determined to defeat the shark, a shark cage. In the book, unlike
central characters and the threat which swims around, probing the movie, Hooper dies after
of the cinematic shark that for vulnerabilities to exploit. going in the cage.
made everyone scared to go in Each character is essential to the
the water – offers a few lessons outcome of the story, with their Quint is a modern-day
about managing a fearsome own unique approach to finding Captain Ahab. He is obsessed
threat. I’m not sure spoiler alerts and killing the shark. with sharks and, in another
are needed for a film that is famous scene, he details
nearly 50 years old but consider Oceanographer Hooper is his experience on the USS
yourself warned. fascinated by the shark and is an Indianapolis when it sank in
intellectual. More comfortable July 1945. Quint is ‘old school’.
Probing for IT on yachts than fishing vessels, He rejects Hooper’s technology,
infrastructure Hooper tends to depend on relies on an old, trusty boat
vulnerabilities to exploit technology to address the threat. with barrels, and is hell bent
He has the latest electronic on revenge of all sharks. He
Jaws has three central equipment and, in a famous respects the shark but believes
characters, and they are all scene, he sinks into the water in in, and only in, his tried-and-

30 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

identify whether devices are in

“DCIM is a tool that can help CIOs face the sharks compliance with policies, we
lurking out there as they constantly move, feed, and are working on AI algorithms
look for their next victims.” to identify anomalies, and we
have added new sustainability
Kevin Brown capabilities to help with
Schneider Electric reporting metrics. We are
leaning into new technologies
and identifying new threats to
true approach. Quint gets eaten believing that technology alone help our customers.
by the shark in both the movie will defeat the shark because it
and the book. isn’t enough. At the opposite Like Quint, we are building
end of the spectrum is Quint, the basics into our DCIM tools.
Police Chief Brody is arguably who knows so much and has For example, we make it easy
the most traditional character. loads of experience but learns to deploy firmware updates, are
He sees the chaos the shark is his old school ways aren’t improving our usability and
wreaking on the community sufficient for the threat. Then user interfaces, and are striving
and works to restore authority there is pragmatic Brody, who to simplify our portfolio and our
and calm from the chaos. He evolves as the film progresses. business. The EcoStruxure IT
is concerned the community Learning from Hooper and team is focused on getting the
won’t survive without the Quint, Brody perseveres, basics right and continually
summer season tourism. He is overcomes his fears, and lives improving because having
pragmatic and working hard, to tell the story. a solid boat with a bunch of
trying to find ways to solve the barrels is part of the overall
problem. To create a well-rounded approach.
approach, I believe we need
Drawback on Brody – he fears to pull from the best traits of Like Brody, we recognize
the unknown – he doesn’t know Hooper, Quint, and Brody: we need to be pragmatic.
how to swim and tends to be embrace new technology like We understand the world is
slow to move. Yet he eventually Hooper; respect the tried and evolving, which is why we
faces these fears and brings true that has worked in the past have invested in obtaining IEC
his pragmatism to the other like Quint; and be pragmatic, 62443 certification for our
characters, telling Quint, in hardworking, and diligent like Network Management Card.
perhaps the most famous line Brody. I would argue that most We are working on ISO27001
from the movie, “you’re gonna IT organizations swing between certification for our IT Expert
need a bigger boat”. Brody these three personalities, cloud products and FIPS 140-
learns from Quint and Hooper driving a multitude of debates 3 certification for on-premises
and becomes the hero who kills on the best approach, like the Data Center Expert, and we
the shark and survives to fight chaos the shark generates in are implementing the features
another day. the beach community. needed to deploy at scale for the
new challenges of the hybrid IT
Is your organization like How DCIM can help environment.
Hooper, Quint, or Brody? CIOs with their IT
infrastructure The combination of all three
So, what can we learn from ‘personalities’ assures us that
the approaches of the main DCIM is a tool that can help we are helping customers meet
characters – what do their CIOs face the sharks lurking out their IT infrastructure challenges
personalities mean for the there as they constantly move, and deal with unknown threats,
threat and what is a best feed, and look for their next so they aren’t afraid to go in
practice? When it comes to victims. the water. Don’t let the summer
the threats facing your IT pass by without learning more
infrastructure, are you like Like Hooper, our team at about our EcoStruxure IT
Hooper, Quint, or Brody? Schneider Electric is using offer and how it can help your
the latest technology in our organization or, if you need a
Hooper is forward-thinking EcoStruxure IT DCIM portfolio “Brody” for help in navigating
by looking to technology for to address threats. We have the waters, contact one of our
answers, but he is arrogant in built in new tools to help EcoStruxure IT experts.

31 | DCD eBook • datacenterdynamics.com

Q&A: Enabling
AI adoption with
Wendy Torrell and
Steven Carlini |
Schneider Electric
Senior research analyst Wendy
Torrell and vice president of
innovation and data center
Steven Carlini answer our
questions on all things AI

How is AI currently
“Traditional data center power, cooling, and racks in the infiltrating people’s
IT space aren’t sufficient for AI training clusters with everyday lives? (i.e,
GPU-based servers that have extreme power capacity healthcare, education,
and rack density demands. With a single AI server manufacturing,
consuming 5.7kW in 1U, it’s feasible to reach densities entertainment etc.)
over 100kW.”
AI, and specifically generative
Steven Carlini AI, is reshaping our daily
Schneider Electric lives. Three segments we see
prioritizing AI opportunities

32 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

are healthcare, banking/

finance, and government/
public sector. In healthcare,
generative AI can be used to
analyze medical data, generate
drug candidates, and create
personalized treatment plans.

Banks can leverage it for

crafting tailored financial
products, improving customer
service through chatbots,
and detecting fraudulent
activities. And government
operations can streamline
processes, enhance decision- Steven Carlini
making, and improve citizen
services, through applications
such as crisis management,
environmental protection, and either multiple standard 60/63A foundational models, we will
infrastructure optimization. Its rack PDUs or one larger custom see a proliferation of IT stacks
potential applications are vast rack PDU becomes necessary. deployed on-premises.
and impactful. We also will see a shift from
air cooling to liquid cooling as Inference workloads will
Why are traditional data the densities of these clusters generally be deployed at lower
center power, cooling and exceed the capabilities of air power consumption and lower
racks no longer sufficient cooling, roughly at 40kW/ rack density than training
for high-density AI rack depending on the server’s workloads, and in many cases,
workloads, what are its airflow requirement. Densities at the Edge, near the points
limitations? are also driving a need for racks of use. These workloads, to a
with greater weight-bearing large degree, can be supported
Traditional data center power, capacity, and more space for by the same power and
cooling, and racks in the IT the cables, piping, etc. within cooling solutions that have
space aren’t sufficient for AI the racks. been deployed for traditional
training clusters with GPU- workloads.
based servers that have Before the power of AI
extreme power capacity and can truly be harnessed, With so many data centers
rack density demands. With foundational models need out there already, what
a single AI server consuming to be trained. What are kind of design changes
5.7kW in 1U, it’s feasible to the key data center design could a pre-existing
reach densities over 100kW. changes or trends we facility implement if it
The use of 120/208 V can expect to see going wanted to add AI clusters
distribution (in NAM) is no forward as a result of to its arsenal?
longer sufficient, and instead, this?
240/415 V distribution is a If there is available power
minimum requirement to limit Training foundational AI capacity, heat rejection capacity,
the number of branch circuits models demands significant and floor space, deploying an
within these high-density computational resources AI training cluster in an existing
racks. And even at that higher which require extreme data center facility is feasible.
voltage, it is still a challenge to power capacities and rack We developed a reference
provide sufficient capacity with densities. Massive build outs design, in collaboration with
a single standard off-the-shelf are leading to insufficient Nvidia, that shows three retrofit
rack PDU. utility power capacity and scenarios of adding roughly
constrained supply chains. 1MW AI cluster to an existing
Once densities exceed But as more businesses deploy facility, that have trade-offs in
34.6kW/rack in NEMA countries inference applications based space usage, efficiency, and
and 43.5kW in IEC countries, on fine-tuned or augmented cost.

33 | DCD eBook • datacenterdynamics.com

>> DCD eBook | Enterprise adoption trends

• Retrofit 1 demonstrates
Wendy Torrell
using air-cooled servers and a
density of 40kW per IT rack.

• Retrofit 2 uses liquid-

cooled servers configured at
73kW/rack and liquid-to-air
CDUs, which allows the heat to
be removed with their existing
(air-cooled) cooling system.

• Retrofit 3 uses liquid-

cooled servers configured at
73kW/rack and liquid-to-liquid
CDUs that removes the heat
by tapping into pipes of an
existing chilled water-cooling
system.
Why is liquid cooling a As well as the tech, in the
We also published a white key component in the wake of AI, are we having
paper, The AI Disruption: adoption of AI? to ‘redesign’ or restructure
Challenges and Guidance the workforce?
for Data Center Design, AI servers will increasingly
which discusses the physical be designed with direct-to- As AI rapidly evolves, there is
infrastructure design changes chip liquid cooling. Factors a growing need for specialized
that are necessary with high- such as server density talent in AI development, data
power, high-density AI training (power per u-height), power science, and machine learning.
clusters. consumption, and specific Businesses are investing
use case will determine the heavily in upskilling and
Is the adoption of AI optimal cooling method. reskilling existing employees to
enabled infrastructure For less demanding AI adapt to AI-driven roles. New
easier for new data center applications, air cooling might positions like AI ethics officers
deployments as they can still suffice. and prompt engineers are
‘get it right from the start’, emerging. However, the talent
or are new builds faced However, as AI models gap remains significant. To stay
with a different set of continue to grow in competitive, organizations must
challenges? complexity and power foster a culture of continuous
consumption, liquid cooling learning, embracing AI as a tool
The availability of utility power is expected to become to enhance human capabilities
must be a top consideration increasingly prevalent for AI rather than replace them.
and may drive the decision on servers.
whether it is feasible to build
a new data center. A new data Direct-to-chip liquid
center design can be cost- and cooling is currently the
energy-optimized to support most prevalent type of liquid
the high density, high-capacity cooling used in AI servers.
AI workloads. For instance, This method involves
as shown in our new build IT directly cooling the hottest
room reference design, placing components of the server,
the CDUs on a separate, high- such as the CPU and GPU,
temperature chilled water loop with a liquid coolant. Air
increases free cooling hours. cooling is still necessary for
High temperature chillers the remaining components
make it possible to operate within the server, and other
this chiller-based cooling loop air-cooled equipment in the
at temperatures that provide cluster like the networking
unmatched cooling efficiency. gear.

34 | DCD eBook • datacenterdynamics.com

Adding an edge
to education with
Schneider Electric and RMD

Watch the case study

se.com/datacentre

35 | DCD eBook • datacenterdynamics.com