Latest / Best of LinkedIn: Digital Transformation & Tech / Best of LinkedIn: Sustainability & Green ICT CW 10/ 11

Transcript

Best of LinkedIn: Sustainability & Green ICT CW 10/ 11 · Best of LinkedIn: Digital Transformation & Tech

• 0:00
  Speaker 0: This episode is provided by Thomas Allgaier and Frennus based on the most relevant LinkedIn posts about sustainability in green ICT, in CW-XI.
• 0:09
  Speaker 0: Frenness supports ICT enterprises with market and competitive intelligence decoding green software developments benchmarking emerging standards tracking regulatory shifts and analyzing competitor strategies.
• 0:23
  Speaker 1: All right and welcome to the deep dive.
• 0:25
  Speaker 1: We are so glad you could join us today.
• 0:26
  Speaker 0: Yeah, thanks for tuning in.
• 0:27
  Speaker 0: So our mission today is to make sense of some massive And I mean massive operational shifts happening right now In sustainability and green ict
• 0:36
  Speaker 1: Right?
• 0:37
  Speaker 1: And we're basing this entirely on the trends in The raw conversations we've seen from professionals On linkedin across calendar weeks ten and eleven
• 0:44
  Speaker 0: Exactly!
• 0:45
  Speaker 0: You know what's really obvious right off the bat From looking at these sources Is it sustainability?
• 0:49
  Speaker 0: It completely moved past that phase Of just being a corporate PR ambition.
• 0:53
  Speaker 0: Oh,
• 0:53
  Speaker 1: absolutely It's not a marketing exercise anymore.
• 0:56
  Speaker 1: it is A hard very real operational engineering problem.
• 1:00
  Speaker 1: Yeah
• 1:00
  Speaker 0: and to really get why I want you to imagine something picture stepping onto the site of this newly built multi billion dollar AI Super computer right?
• 1:12
  Speaker 0: Okay i'm picturing it.
• 1:13
  Speaker 0: its packed with You know those advanced silicon in human history.
• 1:17
  Speaker 0: But then you walk around the back of the building And you don't see a modern power grid connection.
• 1:22
  Speaker 0: Instead, you see this massive fleet of diesel generators just strapped to flatbed trucks burning fossil fuels.
• 1:29
  Speaker 0: Just keep the lights on inside which
• 1:31
  Speaker 1: sounds completely dystopian right?
• 1:33
  Speaker 1: It does.
• 1:34
  Speaker 0: but that is literally what's happening in the data center industry Right now.
• 1:37
  Speaker 0: it's like buying a sprawling twenty bedroom mansion and then realizing The town you built it doesn't have the plumbing or power grid.
• 1:47
  Speaker 1: That's a great analogy because before we can even talk about the solutions, you really need to understand that massive bottleneck.
• 1:54
  Speaker 1: The digital infrastructure is just hitting a physical
• 1:56
  Speaker 1: wall.".
• 1:57
  Speaker 0: Right and Scander Garum shared this brilliant breakdown of his exact issue on LinkedIn.
• 2:02
  Speaker 0: he pointed out this wild mathematical mismatch.
• 2:05
  Speaker 0: He said it takes roughly one year to physically build a data center.
• 2:08
  Speaker 1: okay
• 2:09
  Speaker 0: but it can take up to seven years, actually connected to the grid.
• 2:13
  Speaker 1: Seven years?
• 2:14
  Speaker 1: I mean that is a defining crisis of this sector right now because upgrading high voltage lines getting permits building substations Incredibly
• 2:22
  Speaker 0: slow.
• 2:22
  Speaker 1: Right It's low physical work but AI demand is exponential.
• 2:27
  Speaker 1: Developers cannot just sit on billion dollar real estate investment for seven years waiting for local utility company catch-up
• 2:33
  Speaker 0: Which why like Garum noted they're literally trucking in gas turbines today Yeah which obviously shatters any green commitments those companies made.
• 2:42
  Speaker 0: And the crazy thing is this capacity crisis, it's only getting worse.
• 2:47
  Speaker 0: Boris Kamazachikov brought up this report from the EPRI The Top Electricity Think Tank in the US.
• 2:52
  Speaker 0: Oh right!
• 2:52
  Speaker 0: The Forecaster Report Yeah they just revised their twenty-thirty U.S.
• 2:56
  Speaker 0: data center electricity forecast and They didn't just bump it up a little...they raised by sixty percent.
• 3:01
  Speaker 0: Wait,
• 3:02
  Speaker 1: sixty percent in a single forecast cycle?
• 3:04
  Speaker 1: Yeah.
• 3:05
  Speaker 1: I mean that's almost unheard of.
• 3:06
  Speaker 1: and utility planning.
• 3:07
  Speaker 1: Totally!
• 3:07
  Speaker 1: And that jump...I mean it is almost entirely driven by AI because AI workloads don't act like normal web servers.
• 3:13
  Speaker 1: you know they do not idle.
• 3:15
  Speaker 1: They draw this continuous maximum base load power for training runs that go on for months.
• 3:22
  Speaker 1: EPRIs saying data centers could use up to seventeen percent total U.S electricity by twenty thirty.
• 3:28
  Speaker 1: Seventeen
• 3:28
  Speaker 0: percent just for data centers.
• 3:30
  Speaker 0: It's
• 3:30
  Speaker 1: staggering, but you know while power gets all the headlines there's another constraint that's hitting a breaking point just as fast...
• 3:36
  Speaker 0: Water!
• 3:37
  Speaker 1: Exactly Robert Stevens and Thylak Sambandham both did deep dives on this.
• 3:42
  Speaker 1: Stephens pointed out that A large data center can use five million gallons of water per day.
• 3:48
  Speaker 1: Wow
• 3:48
  Speaker 0: five million gallons.
• 3:50
  Speaker 0: That's what is that comparable to?
• 3:51
  Speaker 0: It's
• 3:51
  Speaker 1: roughly the daily water use of a town of fifty thousand people, that
• 3:54
  Speaker 0: is insane.
• 3:55
  Speaker 1: and The kicker is how it's used.
• 3:57
  Speaker 1: we think of water being pumped directly into server racks To cool them.
• 4:01
  Speaker 1: but up to eighty three percent Of the data centers Water footprint Is actually indirect
• 4:06
  Speaker 0: means meaning
• 4:07
  Speaker 1: its tied to the electricity generation miles away.
• 4:10
  Speaker 1: You know power plants boiling water to spin turbines or cooling their condensers.
• 4:15
  Speaker 1: Oh
• 4:15
  Speaker 0: right so every megawatt the server uses is literally evaporating water off-site.
• 4:20
  Speaker 1: Precisely!
• 4:21
  Speaker 1: Sambendham noted that hyperscale water consumption just in the US, is projected to hit thirty three billion gallons a year by twenty twenty eight.
• 4:29
  Speaker 1: Wow We're already seeing these facilities clash with local communities and places where water is scarce.
• 4:35
  Speaker 0: Yeah but I do want throw an account of perspective here because Gabriel Collins weighed in on this exact debate.
• 4:41
  Speaker 0: In the sources,
• 4:42
  Speaker 1: okay?
• 4:42
  Speaker 1: What did he say?
• 4:43
  Speaker 0: well He argued that while the water numbers are huge Water isn't necessarily The ultimate deal breaker.
• 4:49
  Speaker 0: He looked at the global water use of the massive hyperscalars Google Microsoft meta right and his takeaway was That their collective global water Use is about equal to just one major pipeline delivering water to San Antonio for a year.
• 5:02
  Speaker 1: Okay So it's a lot but on a municipal scale It's a solvable problem
• 5:05
  Speaker 0: exactly.
• 5:06
  Speaker 0: Collins makes the point that you can fix water issue locally.
• 5:10
  Speaker 0: You know, you could use reclaimed wastewater or closed loop cooling...or even air-cooling.
• 5:15
  Speaker 0: But you can't easily fix a broken, overloaded power grid.
• 5:19
  Speaker 1: That's a fair point!
• 5:20
  Speaker 0: For him electricity is still the much bigger structurally harder problem to solve...
• 5:25
  Speaker 1: Which naturally forces industry get creative right?
• 5:28
  Speaker 1: Because if you cant wait seven years for that grade and your hitting limits on resources what do u
• 5:32
  Speaker 1: do?!
• 5:32
  Speaker 1: You
• 5:32
  Speaker 0: re-engineer everything Exactly.
• 5:34
  Speaker 1: we're seeing this massive wave of physical fixes.
• 5:37
  Speaker 1: its pretty incredible
• 5:38
  Speaker 0: And it goes down into actual buildings.
• 5:40
  Speaker 0: Coral Raib highlighted this structural shift that I thought was super interesting, like we usually picture data centers as these ugly concrete and steel bunkers.
• 5:49
  Speaker 1: Yeah, brutalist architecture.
• 5:50
  Speaker 0: Right but Ray pointed out there's a huge move toward using cross-laminated timber or CLT for the buildings instead.
• 5:58
  Speaker 1: Oh engineered wood instead of concrete?
• 6:00
  Speaker 0: Exactly because concrete production is terrible for carbon emissions.
• 6:04
  Speaker 0: It's not just the fossil fuels burning to heat the kilns it's The actual chemical process that releases tons of CO too.
• 6:11
  Speaker 1: right by using clt Which is basically these layers of timber glued together so its as strong steel you avoid all those emissions, plus the wood itself acts as a carbon sink.
• 6:24
  Speaker 1: And Raib said using prefab tumor parts makes construction way faster too.
• 6:28
  Speaker 0: That's so smart!
• 6:29
  Speaker 0: So the building itself goes from being a carbon emitter to a carbon sink, and then inside the building they're rethinking power architecture because if you want to use solar or wind on site You hit the intermittency wall.
• 6:40
  Speaker 0: The sun goes down, the winds stops.
• 6:42
  Speaker 1: And standard lithium ion batteries only give what?
• 6:44
  Speaker 1: Four hours of backup
• 6:46
  Speaker 0: Exactly...so What happens If it is cloudy for three days?
• 6:49
  Speaker 1: You're back to the diesel trucks.
• 6:50
  Speaker 0: Right, but Aaron Patterson shared this major update about a Google and Excel energy data center in Minnesota.
• 6:57
  Speaker 0: They are totally skipping lithium ion and putting in a three hundred megawatt iron-air battery.
• 7:01
  Speaker 0: Okay
• 7:02
  Speaker 1: Iron air that's the one that uses like reversible rusting.
• 7:04
  Speaker 1: Yeah
• 7:04
  Speaker 0: The chemistry is wild.
• 7:05
  Speaker 0: it basically rusts an unrest to store and release power.
• 7:09
  Speaker 0: But the key metric Is how long it lasts?
• 7:12
  Speaker 0: This system can soar power for a hundred hours.
• 7:15
  Speaker 1: Yes It completely changes the math.
• 7:17
  Speaker 1: You could run a massive AI facility, two-thirty seven on clean energy with zero fossil fuel backup.
• 7:24
  Speaker 0: That is massive.
• 7:25
  Speaker 0: so it basically turns unreliable solar into solid baseline power
• 7:30
  Speaker 1: Exactly and its not just The Power Generation being rethought.
• 7:33
  Speaker 1: Shane Snyder talked about networking inside of the facility.
• 7:37
  Speaker 1: People don't realize how much power takes to move data back & forth between GPUs during an AI training run.
• 7:44
  Speaker 1: It's
• 7:44
  Speaker 0: a huge energy drain,
• 7:45
  Speaker 1: right?
• 7:46
  Speaker 1: And Snire pointed out that Microsoft is shifting away from traditional lasers to micro IDs for their networking cables.
• 7:53
  Speaker 1: it's A great mental model.
• 7:54
  Speaker 1: Traditional lasers are like pushing water through high pressure fire hose.
• 7:57
  Speaker 1: takes a ton of energy and gets really hot
• 7:59
  Speaker 0: makes sense
• 8:00
  Speaker 1: But micro-LEDs use thousands of parallel channels, so it's more like letting that same amount of water flow gently down a wide river.
• 8:07
  Speaker 0: Oh!
• 8:07
  Speaker 0: So its like wide and slow instead narrowing fast?
• 8:09
  Speaker 1: Exactly you get the same massive data speed but with way less resistance.
• 8:13
  Speaker 1: Snyder said this shift could cut networking power by fifty percent.
• 8:17
  Speaker 0: That is wild.
• 8:18
  Speaker 0: And speaking about heat...that brings us to what I think is most elegant physical fix we found in the sources
• 8:25
  Speaker 1: The waste heat.
• 8:26
  Speaker 0: Yes because no matter how efficient your servers or cables are, all that power eventually turns into heat.
• 8:33
  Speaker 0: And normally we spend billions of dollars in all the water just dumping the heat into air.
• 8:38
  Speaker 1: Just literally throwing it away?
• 8:40
  Speaker 0: Exactly!
• 8:41
  Speaker 0: But Samson Tam and Martin Rankis both brought up this huge shift-treating waste heat as a utility... not a liability.
• 8:49
  Speaker 0: Right, Rinkus made such a good point.
• 8:51
  Speaker 0: he said you can't separate power generation and cooling anymore.
• 8:54
  Speaker 0: if your expelling half of energy is heat You're basically running thermal power plant.
• 8:58
  Speaker 0: that just happens to be doing math.
• 9:00
  Speaker 1: Yeah That's brilliant way to phrase it And Samson Tam talked about how these facilities are now capturing the heat with heat pumps and sending into local district heating system.
• 9:09
  Speaker 0: So instead being drain on town.
• 9:12
  Speaker 0: The data center actually provides heating for neighborhood.
• 9:15
  Speaker 0: It completely changes how community views facility
• 9:18
  Speaker 1: It really does, but you know as cool is all these hardware upgrades.
• 9:21
  Speaker 1: are the timber The batteries.
• 9:23
  Speaker 1: The heating.
• 9:24
  Speaker 1: it's really just putting a better engine in a car if the software running on those servers Is fundamentally bloated?
• 9:30
  Speaker 1: You're still driving with the parking brake on
• 9:32
  Speaker 0: exactly.
• 9:33
  Speaker 0: Which is the perfect transition into the whole green software and Green Ops theme?
• 9:37
  Speaker 0: right because it feels A little silly to build a billion dollar green data center If the AI models running inside it or just horribly inefficient.
• 9:47
  Speaker 1: Right, we have to look at the code itself.
• 9:49
  Speaker 1: Yeah
• 9:50
  Speaker 0: and The gains they're seeing here are immediate.
• 9:52
  Speaker 0: I was blown away by this research.
• 9:54
  Speaker 0: Sanjay Potter shared.
• 9:55
  Speaker 0: We always think we need new microchips But he showed that just optimizing the design of an AI prompt can drop a model's energy use And co-two emissions By thirty two to forty eight percent.
• 10:06
  Speaker 0: wait
• 10:06
  Speaker 1: almost fifty percent Just from changing the prompt?
• 10:09
  Speaker 0: Just by typing the prompt.
• 10:10
  Speaker 0: better no hardware changes.
• 10:11
  Speaker 0: That's
• 10:12
  Speaker 1: incredible
• 10:13
  Speaker 0: And the logic makes sense.
• 10:14
  Speaker 0: If you give an AI and open-ended prompt without a strict limit, The GPU has to keep all these neural weights loaded burning max power while it tries to figure out when to stop generating text.
• 10:25
  Speaker 0: But if you just give it rigid constraints You cap the token generation...the hardware spins down way faster.
• 10:31
  Speaker 1: That directly ties into what Gary Graves from Earth AI was saying in his post.
• 10:37
  Speaker 1: He is super critical of this brute force mindset and I right now
• 10:41
  Speaker 0: the whole just buy more GPUs thing
• 10:43
  Speaker 1: exactly.
• 10:44
  Speaker 1: Graves pointed out that a single modern AI chip draws as much power As A window AC unit running non-stop
• 10:51
  Speaker 0: well,
• 10:52
  Speaker 1: And he said That Just Scaling Up That Brute Force Is Basically An Engineering Failure.
• 10:56
  Speaker 1: His Exact Quote Was More Compute Is not The Same Thing as Better Design?
• 11:01
  Speaker 0: He's Totally Right.
• 11:02
  Speaker 0: it has to be About Lightweight Smarter architecture.
• 11:05
  Speaker 0: And there was this mathematical breakthrough shared by Jason Michael Knet that really blew my mind, he talked about this framework called Bloom Theory.
• 11:12
  Speaker 1: Oh yeah!
• 11:13
  Speaker 1: This is probably the most fascinating technical piece we read.
• 11:16
  Speaker 0: Totally So.
• 11:16
  Speaker 0: for decades AI has relied on a thing like Markov Property.
• 11:20
  Speaker 0: Right Without getting too technical The Markov property basically forces the AI to squish its entire history into single current state so computer doesn't crash as data gets longer It
• 11:30
  Speaker 1: compresses everything.
• 11:32
  Speaker 1: Yeah
• 11:32
  Speaker 0: But Net who?
• 11:34
  Speaker 0: By the way, as a physics background working on The Higgs Boson he realized that compressing all of history is actually super inefficient.
• 11:42
  Speaker 1: Because Bloom Theory gets around the Markov property completely
• 11:45
  Speaker 0: right?
• 11:45
  Speaker 0: Exactly!
• 11:46
  Speaker 0: It uses this dynamically branching graph thing.
• 11:48
  Speaker 0: and what it means for hardware.
• 11:50
  Speaker 0: you get linear growth in your compute needs not exponential
• 11:55
  Speaker 1: Which is
• 11:55
  Speaker 0: huge!
• 11:56
  Speaker 0: It really is.
• 11:57
  Speaker 0: Nets benchmarks showed model used to take over three minutes.
• 12:01
  Speaker 0: train on standard architecture took point seven seconds using gloom theory.
• 12:05
  Speaker 1: From minutes down to less than a second, the energy savings there are just astronomical.
• 12:11
  Speaker 1: but and this is The Big Hurdle how do you actually get developers to care about this?
• 12:15
  Speaker 0: That's the million dollar question because devs or judged on speed in accuracy nobody Is checking the carbon footprint of their code right.
• 12:24
  Speaker 1: And Andrea Hendricks nailed This.
• 12:26
  Speaker 1: she bluntly said if You don't measure Energy Nothing Changes.
• 12:30
  Speaker 1: Companies are great at tracking cloud costs because it's right there on the dashboard.
• 12:35
  Speaker 1: But until raw energy use is sitting next to dollar amount as a key metric, sustainability will just stay in the reporting department and won't reach actual engineers if
• 12:45
  Speaker 0: you don't measure or manage it
• 12:46
  Speaker 1: Exactly!
• 12:47
  Speaker 1: And that what the Green Ops movement is trying fix.
• 12:50
  Speaker 1: Kevin Leslie highlighted that these certifications, like the Greenox Academy are standardizing this right now.
• 12:56
  Speaker 1: They want to give developers real-time data on the carbon intensity of their code while they're writing
• 13:02
  Speaker 0: it... So they see the impact immediately?
• 13:04
  Speaker 0: Yeah
• 13:05
  Speaker 1: forcing efficiency right at the source.
• 13:07
  Speaker 0: Okay so we've covered the hardware We've covered software.
• 13:10
  Speaker 0: Let's zoom out into a big picture here Because speed in this shift is wild.
• 13:18
  Speaker 0: Why is everything suddenly accelerating right now?
• 13:20
  Speaker 1: Because the incentives have completely flipped.
• 13:23
  Speaker 0: Yeah, Nikol Shlapsman really broke this down.
• 13:25
  Speaker 0: he pointed out that Europe Is grabbing over half of the global climate tech funding Right Now and it's not because they're just feeling super environmentally friendly.
• 13:34
  Speaker 0: He said Climate Tech isn't even about The climate anymore.
• 13:38
  Speaker 0: It's About Who Gets Paid when the whole world is forced to upgrade its industrial base.
• 13:43
  Speaker 0: It's
• 13:44
  Speaker 1: pure economic strategy!
• 13:45
  Speaker 0: Exactly, governments are funding the infrastructure that companies will literally have to buy over the next ten years.
• 13:51
  Speaker 0: And
• 13:51
  Speaker 1: you see it in the policies too.
• 13:53
  Speaker 1: Lucia Lopez-Narayega did a great deep dive into the EU Industrial Accelerator
• 13:58
  Speaker 0: Act.
• 13:58
  Speaker 0: Oh yeah
• 13:58
  Speaker 1: For a long time EU policy was all about punishments carbon taxes strict rules Basically the stick, but this act is a massive shift to the carrot.
• 14:09
  Speaker 1: They're actively guaranteeing markets for low carbon products.
• 14:12
  Speaker 1: Their linking decarbonization straight-to industrial dominance
• 14:15
  Speaker 0: because they know if they don't own The green supply chain there gonna be completely dependent on other countries For future tech.
• 14:21
  Speaker 1: precisely it's a global arms race.
• 14:24
  Speaker 1: Tim Schumacher pointed out that China's fifteenth five year plan looks at the climate transition purely as A matter of energy sovereignty.
• 14:31
  Speaker 0: energy sovereignty Right.
• 14:33
  Speaker 1: Their bet is that whoever controls the clean power and batteries to store it, will control next era of manufacturing and AI.
• 14:41
  Speaker 0: So its an arms race for power.
• 14:44
  Speaker 0: But if this becomes a mandatory market thing How do we stop companies from just greenwashing?
• 14:49
  Speaker 0: You know If tech giant says their new AI data center is one hundred percent renewable how actually?
• 14:55
  Speaker 0: no they aren't lying.
• 14:57
  Speaker 1: While the accounting rules are changing, Asim Hussein is pushing hard for something called hourly carbon accounting to replace the old system of annual averages.
• 15:06
  Speaker 0: Wait so with annual averages could a company just buy a tonne cheap solar credits in summer run their servers on diesel all winter and still legally say they're hundred percent green?
• 15:18
  Speaker 1: Yes that's exactly the loophole!
• 15:20
  Speaker 1: The annual average totally hides reality.
• 15:23
  Speaker 1: But hourly carbon accounting forces transparency, it matches your power use with the power generation on the grid hour by hour.
• 15:30
  Speaker 1: Oh wow!
• 15:30
  Speaker 1: It proves whether your data center was actually using clean energy at.
• 15:34
  Speaker 1: you know two points AOM On a Tuesday when the wind wasn't blowing
• 15:37
  Speaker 0: Which has got to be terrifying for a brand.
• 15:38
  Speaker 0: if they get caught faking it
• 15:40
  Speaker 1: It's becoming commercially dangerous to fake it.
• 15:42
  Speaker 1: Will Nordberg shared data showing that sustainability isn't just nice to have anymore It's actually a leading indicator of commercial performance.
• 15:50
  Speaker 1: Really?
• 15:51
  Speaker 1: Yeah, A bad sustainability perception can predict the drop in market share twelve to twenty-four months in advance.
• 15:56
  Speaker 1: Verifiable Sustainability is the baseline expectation.
• 15:59
  Speaker 1: now If you cant prove it You lose revenue Period.
• 16:03
  Speaker 0: Okay so if we synthesize all this All these insights from past couple weeks What does main takeaway for listener right now?
• 16:11
  Speaker 1: I think core take away that green IT no longer a compliance checklist.
• 16:15
  Speaker 1: It is the ultimate test of your operational competence.
• 16:18
  Speaker 1: Yeah, whether you're dealing with a seven-year wait for grid power or rethinking your AI math to save compute time Or getting ready for strict hourly carbon audits Efficiency is survival.
• 16:28
  Speaker 1: now.
• 16:29
  Speaker 1: if your infrastructure's wasteful You won't scale and you don't compete.
• 16:32
  Speaker 1: The era of cheap infinite compute Is over
• 16:35
  Speaker 0: it really?
• 16:36
  Speaker 0: Which leaves me With a final thought for you to mull over as we wrap up today.
• 16:40
  Speaker 0: If energy use is rapidly becoming the ultimate constraint and hourly tracking exposes every single wasted megawatt.
• 16:47
  Speaker 0: Will we soon see a day where enterprise software is priced not by the features it offers, but purely by the carbon footprint that leaves behind?
• 16:55
  Speaker 1: Ooh!
• 16:55
  Speaker 1: That's great question to leave off on... If
• 16:57
  Speaker 0: you enjoyed this episode new episodes drop every two weeks.
• 17:01
  Speaker 0: Also check out our other editions of Cloud, Digital Products & Services Artificial Intelligence and ICT in Tech Insights HealthTech DefenseTech.
• 17:09
  Speaker 1: Thanks so much for joining us to explore these sources.
• 17:11
  Speaker 1: Stay curious and remember to subscribe,