00:00:00
[Music]
00:00:00
in just a short space of time AI has
00:00:03
rapidly transformed from a niche area of
00:00:05
computer science into a ubiquitous Force
00:00:08
reshaping Modern Life AI regulation
00:00:11
generative AI I just wait for the AI to
00:00:13
do it but to fuel AI it needs to process
00:00:16
data lots of data and that needs scores
00:00:20
of massive data centers across the world
00:00:23
construction projects are underway but
00:00:25
building data centers is only one part
00:00:29
where do we get the power to run them
00:00:31
this is the central question facing the
00:00:33
future of AI because these AI operations
00:00:36
need power and a lot of
00:00:39
it the main bottleneck for AI was
00:00:42
initially computing power were there
00:00:44
enough chips to process the data but now
00:00:47
energy is being seen as the critical
00:00:49
Factor how do we power the data
00:00:52
[Music]
00:00:53
sensus I think we would probably build
00:00:56
out bigger clusters than we currently
00:01:00
can if we could get the energy to do it
00:01:03
the term data center is becoming ever
00:01:05
more vague it largely describes a
00:01:08
building or section of a building that
00:01:10
is used to store Digital Data along with
00:01:12
the compute power needed to manage and
00:01:14
distribute this information this is
00:01:16
normally accessed remotely via some sort
00:01:18
of network the most obvious example of
00:01:21
this being the internet having large
00:01:23
racks of computers working 24 hours a
00:01:25
day to manage this data drives the large
00:01:28
power drawer power consumption increases
00:01:31
exponentially with AI data centers
00:01:33
firstly it requires a huge amount of
00:01:35
information to be stored fed through and
00:01:37
created when training the machine
00:01:39
learning
00:01:40
algorithms secondly the end AI tools
00:01:43
require massive computational resources
00:01:46
including large amounts of gpus to
00:01:48
process and analyze large data sets when
00:01:51
generating the end result large language
00:01:53
models like chat GPT and image
00:01:56
generation models like Dary and mid
00:01:58
Journey cannot be processed on devices
00:02:00
like laptops and phones they require
00:02:03
communication with cloud data centers
00:02:05
for every
00:02:06
transaction James Walker is the chief
00:02:09
executive officer and director of Nano
00:02:11
nuclear energy Inc a company looking to
00:02:14
provide Cutting Edge portable micro
00:02:16
reactors and other micro reactor
00:02:18
Technologies we've spoken to a number of
00:02:21
um people that are trying to build data
00:02:23
centers at the moment and some of the
00:02:25
projections they require will go up to
00:02:28
sort of 2 gaw and those are just the
00:02:30
conversations we're having I'm sure
00:02:31
there are other conversations that are
00:02:33
being had at the moment where they are
00:02:35
asking for a lot more than that
00:02:37
President Trump and his team Envision a
00:02:38
futur in AI development where the US
00:02:41
leads the world powering everything from
00:02:43
autonomous vehicles to military
00:02:46
AI but let's talk numbers a typical AI
00:02:50
data center can consume as much
00:02:52
electricity as a small City and when you
00:02:54
multiply that by hundreds or even
00:02:57
thousands of centers the demand starts
00:02:59
to look well pretty intense founder of
00:03:02
dig Economist Alex deis tells us how
00:03:05
these Digital Trends have accelerated
00:03:07
energy consumption during the previous
00:03:10
decade they were responsible for at
00:03:12
least 1% of global electricity
00:03:14
consumption and that percentag is like
00:03:16
increasing fast over the past few years
00:03:19
due to Trends such as digital currency
00:03:21
mining uh and also recently artificial
00:03:24
intelligence which is also extremely
00:03:26
energy hungry as well so you know in the
00:03:29
coming years it's already been estimated
00:03:31
that data centers are going to be
00:03:32
responsible for you know 3 to 4% of
00:03:35
global electricity consumption uh so
00:03:38
it's quite significant to put that into
00:03:41
perspective estimates suggest that data
00:03:44
centers are currently consuming more
00:03:45
than the total energy consumption of
00:03:48
France we spoke to technology analyst
00:03:51
Jack gold who tells us why this is such
00:03:53
a massive problem Elon Musk built the
00:03:56
new xai plant in Nashville I believe it
00:03:59
was we looked at the power required to
00:04:02
to run that thing and it's the
00:04:04
equivalent of about 55 or 60,000 homes
00:04:07
and where you going to get that from
00:04:09
most uh electrical generation at least
00:04:12
in in the US is already uh on the edge
00:04:17
that that is to say that you can't
00:04:19
really increase it by 10 or 15 or 20% it
00:04:22
the plants won't handle it and we
00:04:24
already have brown outs right in the
00:04:25
summer when it gets real hot and
00:04:27
everyone's got their air conditioning on
00:04:28
the energy grid is an interconnected
00:04:30
distribution network of power lines and
00:04:32
substations delivering current from the
00:04:34
source of generation to your house while
00:04:37
managing the voltage and amplitude to
00:04:39
power home electronics at a consistent
00:04:41
and accepted level modern power
00:04:43
generation is made up of multiple
00:04:45
different methods and Technologies
00:04:47
primarily including coal gas nuclear
00:04:50
solar and wind each of these offer
00:04:53
benefits and downsides often being
00:04:55
implemented for different use cases and
00:04:58
locations for example
00:05:00
solar works better in Sunny climates and
00:05:02
wind turbines are often placed offshore
00:05:05
where there is more uninterrupted wind
00:05:08
whether it's a few hundred megawatts or
00:05:10
multiple gws these generation systems
00:05:13
are only designed to be able to provide
00:05:15
a certain amount of power to Any Given
00:05:17
area when a higher amount of energy is
00:05:20
being consumed than expected it creates
00:05:22
a strain on the available resources and
00:05:25
can limit Supply to end users building
00:05:27
new plants can be slow costly and
00:05:31
complicated so the question arises does
00:05:34
the United States or any developed
00:05:36
country have the energy infrastructure
00:05:39
to support massive expansion it's true
00:05:42
some parts of the US are already
00:05:44
experiencing power shortages because of
00:05:46
data census States like California and
00:05:48
Arizona which have long struggled with
00:05:50
electricity demand are seeing increased
00:05:53
strain on their grids due to the rapid
00:05:54
expansion of data censors and while AI
00:05:58
is pushing us to The Cutting Edge ofch
00:05:59
technology it's also pushing our
00:06:02
existing energy grids to their limits
00:06:04
whenever a big energy consumer shows up
00:06:06
on your power grid whether it's Crypt
00:06:08
Miner or artificial intelligence based
00:06:11
data center they start consuming a whole
00:06:14
lot of power and the supply typically
00:06:16
doesn't just go up like it takes a long
00:06:19
time to increase your available power
00:06:22
supply the only thing that might happen
00:06:24
is that there are some redundant energy
00:06:28
sources that you can react activate some
00:06:30
old um coal based power plants that you
00:06:33
can maybe trigger to to provide the
00:06:36
power that you need but one of the
00:06:38
issues they're having is that they just
00:06:40
can't extend uh or increase their power
00:06:42
supply anymore they're already maxed out
00:06:44
so in that case the only consequence is
00:06:47
like okay they're just going to be
00:06:48
paying more for uh Power because this
00:06:52
are simple laws of supply and demand
00:06:54
Supply is the same demand goes up so
00:06:56
prices go up as AI accelerates it's Al
00:06:59
only make things more challenging some
00:07:02
US states are actively pushing to become
00:07:04
the next tech hubs for these AI data
00:07:06
centers because they have Surplus power
00:07:09
States like Wyoming Texas and North
00:07:11
Dakota have vast SES of untapped energy
00:07:14
potential Wyoming has been making big
00:07:17
moves with wind energy while Texas is
00:07:19
already home to some of the world's
00:07:21
largest data
00:07:22
centers these states have a unique
00:07:25
opportunity to attract datadriven
00:07:27
Industries but it's not just about
00:07:29
having extra power they'll need the
00:07:31
infrastructure to support it too of
00:07:34
course building new data centers isn't
00:07:36
all smooth saling the big cloud
00:07:39
providers haven't been able to build
00:07:40
data centers in areas where they would
00:07:42
like to because there isn't enough power
00:07:43
to power them nuclear is one way to do
00:07:48
that it is also risky in the sense that
00:07:53
it takes a long time for regulations to
00:07:56
be put in place to allow you to build a
00:07:58
new plant it can take a decade the
00:08:01
really big plants that we built in the
00:08:02
past take a long time to construct even
00:08:05
if once you get the regulation nuclear
00:08:07
seems like a good option for power
00:08:09
generation as it doesn't create
00:08:10
greenhouse gases however these plants
00:08:13
are complicated from both a legal and
00:08:16
construction standpoint as well as
00:08:18
highly expensive and timec consuming to
00:08:20
build not to mention aspects like
00:08:22
nuclear waste management and the expense
00:08:24
of eventual
00:08:26
decommissioning for example the new
00:08:28
hinley points SE site in the UK is
00:08:31
currently predicted to cost £46 billion
00:08:34
or $58 billion and not to be completed
00:08:37
until
00:08:39
2031 that's a 6-year delay on completion
00:08:42
and a staggering 28 billion over the
00:08:45
original £18 billion
00:08:48
budget renewable energy sources like
00:08:50
wind and solar are growing but they
00:08:53
can't provide the consistent energy data
00:08:55
centers demand Renewables like solar and
00:08:58
wind are heavily dependent dependent on
00:08:59
environmental conditions no wind means
00:09:02
the turbines can't spin too many clouds
00:09:05
and the solar panels become
00:09:07
inefficient this creates an
00:09:08
unpredictable and inconsistent method of
00:09:11
power generation energy storage
00:09:13
facilities are being widely developed to
00:09:15
account for this
00:09:16
inconsistency including lithium ion
00:09:19
battery plants and renewable more
00:09:21
eco-friendly Alternatives like gravity
00:09:23
batteries and hydro storage
00:09:26
facilities however these are systems
00:09:28
that need to rely on a considerable
00:09:30
amount of redundancy both in storage and
00:09:33
for extra generation in optimal
00:09:35
conditions which takes up more space and
00:09:37
resources to
00:09:39
achieve the problem with them is that
00:09:41
they are too intermittent the data
00:09:43
centers in AI need very little downtime
00:09:45
on an annual basis and you can obviously
00:09:48
supplement that by having a a very large
00:09:50
footprint dedicated to say solo with big
00:09:52
storage facilities but then you're
00:09:54
getting into quite a lot of expense and
00:09:57
you're still not drisking it and you
00:09:58
still need to have a lot of thought back
00:09:59
systems for when power is intermittent
00:10:02
and it is not the right season and
00:10:03
conditions aren't optimal and your
00:10:05
storage facility is being run down on
00:10:08
and and then you run the risk of the
00:10:09
entire operation shutting down and
00:10:11
needing to fall back on something like
00:10:13
diesel that's why Tech giants like
00:10:15
Google Amazon and Microsoft are backing
00:10:17
startups looking to provide the next
00:10:19
generation of nuclear power plants there
00:10:22
are a number of startup companies are
00:10:23
trying to build smaller nuke plants and
00:10:25
they think they can build it very
00:10:26
quickly the question becomes regulations
00:10:29
will the government allow it uh and and
00:10:31
that's to be seen now with the new Trump
00:10:33
Administration coming on board it's much
00:10:36
more likely they're going to be
00:10:38
favorably inclined to allow that kind of
00:10:40
technology in my opinion it's actually a
00:10:42
good idea smrs or small modular reactors
00:10:46
seem to be the magic answer to a lot of
00:10:49
these power
00:10:50
concerns companies like last energy and
00:10:52
Rolls-Royce are looking to provide a
00:10:54
bolt-on solution where data centers are
00:10:57
able to have their own personal new
00:10:59
nuclear power plant to directly generate
00:11:01
electricity for their needs the modular
00:11:04
aspect allows them to scale up as needed
00:11:08
there's also talk about implementing
00:11:09
these into areas like container ships to
00:11:12
address shipping emissions numbers are
00:11:14
currently being thrown around like the
00:11:16
ability for a ship to run for 30 years
00:11:18
without refueling they've essentially
00:11:20
done an assessment of all the available
00:11:23
Power Systems that are available to them
00:11:24
and and decided that it's really only
00:11:26
feasible that that amount of power can
00:11:28
be generated by nucle
00:11:29
it has additional benefits too like high
00:11:31
base load power you can put it wherever
00:11:33
you want but their energy projections
00:11:34
are huge like I think over the the next
00:11:37
20 years they're going to require
00:11:40
somewhere in the region about 30 to 40%
00:11:42
of um increase in power that the current
00:11:45
the United States is currently utilizing
00:11:47
although in development these smrs still
00:11:50
have a long way to go predominantly in
00:11:52
regulation after all safety is a major
00:11:55
concern with anything nuclear for the
00:11:58
time being we're seeing in companies
00:11:59
like Microsoft striking Direct Energy
00:12:01
deals with the likes of the 3M Island
00:12:04
nuclear facility this will provide them
00:12:06
power when it reopens properly in 2028
00:12:10
after suffering the worst nuclear
00:12:11
incident in US history in
00:12:14
1979 in the meantime there's the
00:12:16
question of how to balance energy usage
00:12:19
for consumers versus AI operations
00:12:22
initially the new data centers
00:12:24
envisioned by Stargate will need power
00:12:27
from somewhere they will come up with
00:12:28
interest
00:12:29
Solutions I think they're probably going
00:12:31
to use gas problem with that obviously
00:12:33
is that that's not feasible in all
00:12:35
locations to run gas to remote locations
00:12:38
um to do data centers and that kind of
00:12:40
thing it's would be the same problem if
00:12:42
you were to try to install transmission
00:12:44
lines from grids to to for your data
00:12:46
centers and a lot of the grids cannot
00:12:48
afford to lose that much power to a Data
00:12:50
Center and I think that happened to
00:12:52
Amazon very recently where what they
00:12:54
wanted to buy Off the Grid got shut down
00:12:56
essentially and the utility provider was
00:12:58
not allowed to provide them that power
00:12:59
because it was it was too significant a
00:13:01
portion of of what the utility company
00:13:04
was able to provide to the the local
00:13:06
population will people have enough power
00:13:09
in their homes or will we have to choose
00:13:12
between keeping the lights on and
00:13:13
powering the AI of the future but it's
00:13:16
not just power that data centers consume
00:13:18
at an alarming rate all the power
00:13:20
consumption creates Heat and the GPU
00:13:23
chips processing AI data needs to be
00:13:25
called that means fresh water and a lot
00:13:28
of of it an average data center is
00:13:31
estimated to use about 300,000 gallons
00:13:34
or 1.2 million L of water every day just
00:13:38
for cooling that's roughly the
00:13:40
equivalent to the water usage of 100,000
00:13:43
homes one Innovative idea is to build
00:13:46
underwater data centers and use the
00:13:48
oceans to call the AI processing chips
00:13:51
naturally Microsoft has already
00:13:53
experimented with an underwater data pod
00:13:56
in the waters of Scotland the container
00:13:58
behind me it fits on a it fits on a
00:14:00
trailer it fits on a cargo ship and it
00:14:02
allows us to actually build up this data
00:14:05
center to any size that we want one of
00:14:07
the upsides of the Stargate initiative
00:14:09
touted by President Trump is hundreds of
00:14:12
thousands of jobs that may be the case
00:14:14
in construction but data centers need a
00:14:17
few people to operate them once built
00:14:19
they are not a Powerhouse for employment
00:14:21
as Sweden discovered they gave power
00:14:23
subsidies to data centers for like 6
00:14:26
years 7even years since 2017 and they
00:14:29
top last year um then they realized like
00:14:32
okay we're not getting a lot of jobs
00:14:33
from that uh we would prefer to put this
00:14:36
power in steel manufacturing which is
00:14:38
more labor intensive there's no easy
00:14:40
answer here but one thing is certain
00:14:43
ai's power demand will only keep growing
00:14:46
can the US and the rest of the world
00:14:48
handle it it depends on what time
00:14:50
Horizon you're talking about I mean in
00:14:52
in one year things could be dramatically
00:14:55
worse than today in five years you would
00:14:58
expect that things have kind of you know
00:15:01
calmed down and in 10 years or that's so
00:15:04
far away
00:15:06
like the world could be looking
00:15:08
completely different by then indeed in
00:15:11
the short term president Trump's
00:15:13
Stargate initiative will be a massive
00:15:15
challenge for power generation in the US
00:15:18
even with some states having free energy
00:15:19
Supply capacity the overall strain on
00:15:22
the grid could push it to Breaking Point
00:15:25
with grid upgrades on the table the
00:15:27
future of AI powered America will depend
00:15:29
on how we Balance power technology and
00:15:33
the
00:15:35
environment remember to hit the
00:15:36
Subscribe button and ring that Bell to
00:15:38
stay updated with our latest content and
00:15:41
while you're here why not check out
00:15:43
another one of our exciting videos
00:15:45
thanks for watching and see you in the
00:15:47
next one