Latest / Best of LinkedIn: Digital Transformation & Tech / Best of LinkedIn: Green ICT & Sustainable AI CW 30/ 31

Transcript

Best of LinkedIn: Green ICT & Sustainable AI CW 30/ 31 · Best of LinkedIn: Digital Transformation & Tech

• 0:00
  Speaker 0: This episode is provided by Thomas Alguyer and Frennus, based on the most relevant LinkedIn posts about green ICT in sustainable AI from CW-Thirtyandthirtyone.
• 0:09
  Speaker 0: Frenness supports ICT enterprises in the form of delivering precise ICT market and pricing intelligence that analyst subscriptions an existing databases cannot provide.
• 0:18
  Speaker 0: you can find more info in description.
• 0:20
  Speaker 1: Welcome to The Deep Dive everyone!
• 0:22
  Speaker 1: We are exploring top Green ICT and Sustainable AI trends surfacing across Linkedin right now.
• 0:28
  Speaker 1: And if you've been tracking digital transformation lately, the narrative has completely shifted.
• 0:33
  Speaker 1: Yeah
• 0:33
  Speaker 0: totally shifted.
• 0:34
  Speaker 0: we're looking at a really hard pivot from these ambitious sustainability goals to just cold hard arithmetic.
• 0:41
  Speaker 0: So our core mission for this deep dive is tracking the physical footprint of AI,
• 0:45
  Speaker 1: right?
• 0:46
  Speaker 1: Because it's no longer Just a theoretical debate about carbon offsets on some corporate slide deck.
• 0:51
  Speaker 1: ai Is hitting very real Very unyielding physical limits mostly local power grids and honestly community tolerance.
• 0:58
  Speaker 0: Yeah that local pushback as massive.
• 1:00
  Speaker 0: I saw a stat from bull steer on This but i actually had To read twice to believe
• 1:04
  Speaker 1: oh beth The block projects.
• 1:05
  Speaker 0: exactly in the first quarter of this year, local community pushback blocked or delayed at least seventy five U.S.
• 1:12
  Speaker 0: data center projects.
• 1:14
  Speaker 0: we're talking about infrastructure worth over one hundred and thirty billion dollars.
• 1:19
  Speaker 0: That matches the total projected value for all of twenty-twenty five just halted in a single quarter.
• 1:24
  Speaker 0: I mean, is The Grid really at its breaking point or Is this Just A Localized PR Issue?
• 1:28
  Speaker 1: No it's a genuine Breaking Point and the primary friction is actually shifting away from just sheer electricity demand.
• 1:34
  Speaker 1: It's becoming an acute public health And air quality issue.
• 1:37
  Speaker 1: now
• 1:37
  Speaker 0: Wait Really Air Quality?
• 1:39
  Speaker 1: Michael Cork pointed out a staggering development.
• 1:42
  Speaker 1: There are at least eighty-two gas burning power plants currently clanned or underdevelopment specifically to power data centers.
• 1:49
  Speaker 1: When the grid can't handle a load, they're literally building dedicated fossil fuel plants just to keep the servers
• 1:54
  Speaker 1: running.".
• 1:55
  Speaker 0: That is... I mean.. The regulatory speed happening behind scenes there must be wild!
• 1:58
  Speaker 1: It Is Michael noted that in Texas permits for these gas plants to serve data centers Are being fast tracked and an average of eighteen days
• 2:07
  Speaker 0: Eighteen days.
• 2:08
  Speaker 0: Yeah,
• 2:08
  Speaker 1: compared to a traditional power plant air permit which usually takes about two hundred and eighty-five days They're essentially bypassing standard environmental scrutiny because enterprise demand for compute is just that aggressive
• 2:20
  Speaker 0: Which directly drives the rise of what's known as behind-the-meter fossil fuel generation right?
• 2:25
  Speaker 0: Because they can't pull the goats from the main grid without browning out neighborhoods They just build fossil fuel plants on site
• 2:32
  Speaker 1: exactly right behind the utility meter to guarantee uptime.
• 2:36
  Speaker 1: Boris Gamazichukov highlighted estimates projecting that US behind the meter capacity could exceed forty gigawatts by twenty-twenty eight.
• 2:43
  Speaker 1: That's
• 2:43
  Speaker 0: insane!
• 2:44
  Speaker 1: It has become such a systemic issue, that trackers are now actively scoring hyperscalars specifically on their fossil fuel use.
• 2:51
  Speaker 0: Okay wait let me stop you there.
• 2:53
  Speaker 0: We have the tech industry aggressively branding itself as a clean digital future of work, but they are building forty gigawatts behind-the-meter gas plants just to power AI models.
• 3:02
  Speaker 1: Pretty much!
• 3:03
  Speaker 0: That is like buying cutting edge electric vehicles and powering it with diesel generator straps to roof.
• 3:08
  Speaker 0: It completely undermines the entire premise of sustainable technology.
• 3:12
  Speaker 1: I know...it's that exact structural contradiction we're seeing.
• 3:14
  Speaker 1: Yeah, it's driving this macro trend that Oliver Cronk and Chris Adams call the recarbonization of the Internet.
• 3:22
  Speaker 0: Recarbonisation right?
• 3:23
  Speaker 1: The sheer physics of the AI compute build out is actively reversing the clean energy progress we've made.
• 3:30
  Speaker 0: I guess the natural counter argument from tech leaders.
• 3:33
  Speaker 0: as always you know That AI will eventually optimize its own energy use.
• 3:37
  Speaker 0: But then we run into the Jevons paradox.
• 3:39
  Speaker 0: Oh,
• 3:40
  Speaker 1: Arwell Owen's point.
• 3:41
  Speaker 1: Ray
• 3:41
  Speaker 0: Arwell brought this up pointing out that as AI tech becomes more efficient it actually drives over all usage and consumption.
• 3:48
  Speaker 0: The efficiency makes it cheaper which perversely increases a total energy burned across the network
• 3:53
  Speaker 1: Exactly.
• 3:54
  Speaker 1: if models get twice as efficient We just run four times many queries.
• 3:58
  Speaker 1: You can't rely on chip level efficiency to save you when demand curve is exponential.
• 4:03
  Speaker 0: So if these hyperscalers can't just throw up massive gas plants without facing billions in pushback, they have to figure out how to squeeze more compute into less power hungry footprints.
• 4:13
  Speaker 0: And the biggest drain on that power is keeping servers from melting down.
• 4:26
  Speaker 1: as the grid power issue.
• 4:27
  Speaker 0: Yeah, Aparna K shared a brutal reality check on this...
• 4:36
  Speaker 1: which is terrifying when you map that globally!
• 4:39
  Speaker 0: She actually pointed to the Platumata dispute in Kerala where local communities successfully challenged global beverage companies over groundwater depletion.
• 4:48
  Speaker 0: She framed it as direct preview of tensions.
• 4:51
  Speaker 0: we're going see about AI infrastructure and drinking supplies…
• 4:55
  Speaker 1: seeing those plans collide with reality.
• 4:58
  Speaker 1: Xiaolei Ren highlighted a planned open AI data center in Australia, it originally aimed to use recycled water for cooling...to spare the drinking supply.
• 5:07
  Speaker 0: Makes sense on paper!
• 5:08
  Speaker 0: Right
• 5:09
  Speaker 1: but pipeline constraints reportedly forced them to pivot to Waterless Cooling
• 5:13
  Speaker 0: Which sounds like great environmental win until you understand the thermodynamics.
• 5:17
  Speaker 0: Waterless cooling requires significantly higher electricity demand especially during heat waves right?
• 5:22
  Speaker 1: Yes Without evaporative water cooling, you are relying heavily on mechanical refrigeration cycles.
• 5:28
  Speaker 1: Basically massive air conditioners.
• 5:31
  Speaker 1: so you trade a water crisis for localized power crises
• 5:35
  Speaker 0: just shifting the burden right back onto the grid when it's already under maximum thermal strain which explains why The industry is finally looking to change the cooling architecture entirely with liquid cooling.
• 5:46
  Speaker 1: Allison Freeman and Peresh Patel both noted that moving from traditional air cooling to liquid systems like the IR- seven thousand rack drops.
• 5:53
  Speaker 1: The power usage effectiveness or PUE, from around one point four down between one point one at one point two.
• 6:00
  Speaker 0: And for anyone managing data center white space those physics are game changing.
• 6:04
  Speaker 0: Air is a terrible conductor of heat.
• 6:06
  Speaker 0: We spend massive amounts of electricity just spinning giant fans.
• 6:09
  Speaker 1: Exactly, liquid is exponentially denser and transfers heat directly away from the chip.
• 6:14
  Speaker 1: You cut out the fans And AC units
• 6:16
  Speaker 0: which drops the PUE.
• 6:17
  Speaker 0: The density implications are massive.
• 6:19
  Speaker 0: Conventional air-cooled racks usually max between five to twenty kilowatts But with liquid cooling you can push rag densities upto fifty or even one hundred plus kilowatt
• 6:28
  Speaker 1: per rack Drastically more compute in exact same physical footprint.
• 6:33
  Speaker 1: But some are taking that constraint to an absolute extreme.
• 6:36
  Speaker 1: Mohammed El-Saheb detailed China's world first underwater AI data center off the coast of Shanghai.
• 6:42
  Speaker 0: Wait, under water?
• 6:43
  Speaker 1: Yeah they literally dropped a hundred ninety two racks ten meters below the surface powered entirely by an offshore wind farm uses zero freshwater because of passive seawater cooling and operates at a PUE below one point zero five.
• 6:57
  Speaker 0: Okay wait I have to question that underwater solution.
• 7:00
  Speaker 0: If we solve the cooling problem by throwing our servers into the ocean, aren't we just relocating the environmental strain from atmosphere directly to marine ecosystems?
• 7:09
  Speaker 0: It's a very
• 7:09
  Speaker 1: valid concern.
• 7:10
  Speaker 0: We're essentially turning the ocean in to literal heat sink for generative AI!
• 7:15
  Speaker 1: It is exactly that kind of holistic accounting industry you are struggling with but form a pure engineering standpoint.
• 7:20
  Speaker 1: it aligns with strategy.
• 7:22
  Speaker 1: Luchia Li highlighted China's East Data Western Computing Initiative
• 7:26
  Speaker 0: Routing the workloads to where renewable energy and natural cooling naturally exist, like Nixia or Offshore.
• 7:34
  Speaker 1: We're
• 7:35
  Speaker 0: even seeing structural material innovations on land!
• 7:39
  Speaker 0: Carl Raib pointed out that push toward engineered timber & wooden data centers as a lower carbon construction alternative which
• 7:46
  Speaker 1: locks in carbon rather than emitting it during concrete production.
• 7:51
  Speaker 1: The physical innovations are remarkable but they raise much deeper structural issues which is accountability.
• 7:57
  Speaker 1: We have all these physical marvels, but how do we actually know they're delivering sustainable outcomes if no one reporting their telemetry.
• 8:07
  Speaker 0: Oh, speaking of huge gaps in accountability if you're tracking these industry shifts right alongside us definitely make sure to hit subscribe to the deep dive so you don't miss our upcoming breakdowns.
• 8:16
  Speaker 0: but um looking at that reporting gap James Martin noted that only thirty six percent of European data centers actually meet their ED sustainability reporting obligations
• 8:25
  Speaker 1: and there are zero penalties for the sixty four percent that simply ignore it.
• 8:30
  Speaker 1: The internal operational tracking isn't much better.
• 8:32
  Speaker 1: Mathias Hamos cited Uptime Institute data showing a massive disconnect.
• 8:37
  Speaker 1: What
• 8:37
  Speaker 0: kind of disconnect?
• 8:38
  Speaker 0: Well,
• 8:38
  Speaker 1: eighty-seven percent of operators track their overall facility energy consumption but only thirty one percent track the actual carbon emissions tied to that electricity.
• 8:49
  Speaker 1: and for AI workloads tracking for server utilization is stalled at a dismal thirty seven percent.
• 8:54
  Speaker 0: Let me make sure I understand that.
• 8:56
  Speaker 0: They know what the building pulls from the grid, but they have absolutely no idea if the servers inside are actually doing useful computational work or just spinning idle?
• 9:06
  Speaker 1: Exactly!
• 9:06
  Speaker 1: It's a blind spot and it brings up what David Lithicum called blatant industry hypocrisy.
• 9:12
  Speaker 1: Five years ago, Enterprise Tech was all about the green cloud.
• 9:14
  Speaker 0: Oh yeah!
• 9:15
  Speaker 0: Every keynote is about carbon reduction.
• 9:17
  Speaker 1: But that movement vanished because the race for AI infrastructure dominance took priority.
• 9:22
  Speaker 1: And you
• 9:22
  Speaker 0: see this hypocrisy in the Carbon accounting too.
• 9:25
  Speaker 0: Carl Koster noted Meta's exit from the RU- one hundred initiative.
• 9:28
  Speaker 0: It exposes this massive loophole where hyperscalars use annual clean energy matching, which
• 9:33
  Speaker 1: is essentially grainwashing
• 9:34
  Speaker 0: right?
• 9:35
  Speaker 0: They buy a massive amount of solar power during the day claim they are a hundred percent renewable on paper while relying on gigawatts of localized gaspower all night when the sun is down
• 9:44
  Speaker 1: and When clients press them on.
• 9:45
  Speaker 1: why there disclosing specific AI model emissions?
• 9:49
  Speaker 1: The excuses were thin.
• 9:51
  Speaker 1: No When Godard highlighted Google's defense here.
• 9:54
  Speaker 1: They claimed a lack of standard measurement and limited hardware telemetry
• 9:58
  Speaker 0: Which no one pushed back on right saying directional data is absolutely enough
• 10:02
  Speaker 1: to start.
• 10:02
  Speaker 1: Yes, you don't need millimeter precision To know your driving in the wrong direction.
• 10:06
  Speaker 0: I mean wait.
• 10:08
  Speaker 0: These companies map the entire human genome and indexed the whole internet in milliseconds, but they're claiming that can't put a smart meter on server rack.
• 10:16
  Speaker 0: It's
• 10:16
  Speaker 1: absurd!
• 10:18
  Speaker 0: Like no one said lacking telemetry is a priority problem not capability problems.
• 10:24
  Speaker 1: And regulators are rapidly losing patience.
• 10:27
  Speaker 1: Petteri Niki pointed out Finland actually delayed a thirty million Euro data center tax refund scheme to bundle it with strict mandatory sustainability registry law.
• 10:37
  Speaker 1: If you want a massive tax break, prove your metrics with hard telemetry.
• 10:40
  Speaker 0: So if tracking the hardware is this opaque what about software running on top of it?
• 10:45
  Speaker 0: Because software choices offer massive immediate levers for efficiency right?
• 10:49
  Speaker 1: Huge levers!
• 10:50
  Speaker 1: Mark Bradley and Vasuki Sharam Anavilla highlighted that there's now one hundred billion dollar.
• 10:55
  Speaker 1: cloud-waste problem Gardner predicts combining fine ops traditional IT asset management can cut waste by sixty percent.
• 11:02
  Speaker 0: Sixty
• 11:02
  Speaker 0: percent?!
• 11:03
  Speaker 0: That's massive.
• 11:04
  Speaker 0: Naveen Balani provided a really stark example of what this waste looks like.
• 11:09
  Speaker 0: In one retail AI deployment, the autoscale floor was set so high that two-thirds their standing AI capacity sat warm and completely idle for an entire month
• 11:18
  Speaker 1: Just waiting for peak queries never arrived but burning power all time.
• 11:22
  Speaker 1: This gets dangerous as enterprise tech moves toward agentic AI.
• 11:26
  Speaker 0: Oh!
• 11:26
  Speaker 0: The red AI vs green AI framing
• 11:28
  Speaker 1: Exactly Nina Benoy introduced it.
• 11:30
  Speaker 1: Red AI agents maximize performance at all costs and are essentially always on, because they operate autonomously.
• 11:36
  Speaker 1: They create a massive blast radius of emissions.
• 11:39
  Speaker 1: that
• 11:39
  Speaker 0: red AI model is basically like leaving your car engine idling in the driveway.
• 11:43
  Speaker 0: twenty four seven just incase you suddenly need to drive to the store.
• 11:46
  Speaker 1: It's terrible default setting.
• 11:48
  Speaker 1: Nina advocates for green AI Like requiring hard expiration dates For agent since strict confidence thresholds.
• 11:54
  Speaker 1: so The AI stops computing an asks human for help instead Of entering endless retry loops.
• 11:59
  Speaker 0: And we are also seeing efficiency gains at the model architecture level.
• 12:04
  Speaker 0: Robert Kias noted that performance of the DeepSeq V-Four flash model, it scores a fifty on intelligence indexes.
• 12:11
  Speaker 0: but here's the key... It only uses around thirteen billion active parameters meaning
• 12:16
  Speaker 1: it activates The mathematical operations drop astronomically.
• 12:21
  Speaker 1: It's up to ninety-five percent cheaper and vastly more energy efficient than frontier models,
• 12:26
  Speaker 0: And we can't forget traditional web engineering either.
• 12:29
  Speaker 0: Jane Cagliola study found that simply implementing proper browser caching cuts front end energy use by thirty percent
• 12:35
  Speaker 1: because it reduces network transit.
• 12:36
  Speaker 1: energy in the programming language you choose matters too.
• 12:40
  Speaker 1: Vincent Volbon highlighted a benchmark study where Java ranked fifth overall as an Energy Efficient Virtual Machine Language
• 12:46
  Speaker 0: which really challenges the assumption that running on a JVM makes your software a massive energy drain.
• 12:52
  Speaker 0: So we have all these software labors available today, but zooming out there is a parallel track where emerging markets are building this right from the ground.
• 13:00
  Speaker 0: up rate Yes
• 13:01
  Speaker 1: and This Is Where It Gets Optimistic.
• 13:02
  Speaker 1: John Kamara Framed Africa's Lack of Data Centers Currently Less Than One Percent Of Global Capacity Not As A Deficit But as A Blank Page.
• 13:11
  Speaker 0: That'S A Powerful Pivot building a decentralized, renewable-powered intelligence economy from scratch rather than retrofitting the broken hyperscale model.
• 13:19
  Speaker 1: Precisely.
• 13:20
  Speaker 1: and Devesh Sharma pointed out how in India green hydrogen and AI could add fifteen to twenty gigawatts of annual solar demand by fiscal year.
• 13:29
  Speaker 0: The demand for compute is effectively subsidizing the massive scale-up of And Aldrich Zappellini showed that credit-agricles data lab achieved certification by strictly prioritizing frugal AI, rightsizing the compute to the task instead of just using the biggest model.
• 13:56
  Speaker 1: Which brings us right back to the core structural issue?
• 13:58
  Speaker 0: Right so let me ask you a structural question based on all this.
• 14:01
  Speaker 0: we started talking about communities blocking one hundred and thirty billion dollars of mega campuses.
• 14:07
  Speaker 0: does the hyper scale megacampus model even make sense anymore?
• 14:11
  Speaker 0: Or is this decentralized, renewable first approach in emerging markets actually the blueprint that West needs to copy?
• 14:17
  Speaker 1: The billion dollar question.
• 14:19
  Speaker 1: The physics of power transmission suggests the centralized mega-campus is hitting a hard wall.
• 14:24
  Speaker 1: The future architecture likely has to be distributed placing smaller compute nodes directly adjacent to stranded renewable energy.
• 14:31
  Speaker 1: rather than breaking the national grid.
• 14:33
  Speaker 0: It fundamentally changes how we think about the footprint of intelligence which leaves you with final thought to mull over.
• 14:39
  Speaker 0: We've seen the intense pushback against physical buildings, but we have also seen massive waste sitting in idle software.
• 14:46
  Speaker 0: What happens when communities currently protesting those billion-dollar data centers realize that facility draining their local aquifer is largely running completely idle?
• 14:57
  Speaker 0: poorly optimized red AI agents?
• 15:00
  Speaker 0: Will the next wave of climate activism target specific software workloads rather than just physical buildings?
• 15:06
  Speaker 0: It's
• 15:06
  Speaker 1: totally different kind.
• 15:07
  Speaker 1: accountability.
• 15:07
  Speaker 0: Exactly, if you enjoyed this episode.
• 15:09
  Speaker 0: new episodes drop every two weeks.
• 15:11
  Speaker 0: also check out our other additions on cloud insights and sovereignty digital products in services AI and agentic systems health tech ICT and Tech Insights defense tech and HealthTech.