Best of LinkedIn: Sustainability & Green ICT CW 26/ 27

00:00:00: This episode is provided by Thomas Allgeier and Frennus, based on the most relevant LinkedIn posts about sustainability in green ICT in CW-twenty six and twenty seven.

00:00:09: Frenness supports ICT enterprises in the form of delivering precise ICT market and pricing intelligence that analysts subscriptions and existing databases cannot provide.

00:00:19: you can find more info in the description.

00:00:21: Thank You it's really great to be here.

00:00:23: So what if i told Asking chat GPT to check a simple network route uses about as much computational energy as say chartering a seven forty-seven To go pick up your weekly groceries.

00:00:34: Yeah,

00:00:34: that's seven forty.

00:00:35: seven analogy is.

00:00:36: It's pretty spot on for what we're looking at today.

00:00:38: Right

00:00:39: welcome to the deep dive everyone.

00:00:40: Today We are cutting through all those glossy PR pledges to look At the very real Very urgent sustainability and green ICT trends surfacing across The industry exactly

00:00:50: because if you're building or investing in tech You hear corporate net zero promises absolutely everywhere.

00:00:56: But we want to look at the measurable reality, things like the massive physical footprint of AI ,the brutal physics data center cooling and how green coding is quietly merging with financial operations.

00:01:10: That is our mission today.

00:01:11: So let's just start with how the conversation around artificial intelligence has completely matured, I mean a year ago everyone was arguing over whether AI Was inherently good or bad for humanity.

00:01:21: right and now.

00:01:22: Now it is strictly an accounting in governance problem.

00:01:25: Yeah We've basically stopped philosophizing And you know started measuring which

00:01:29: Is a huge shift?

00:01:31: It really is.

00:01:32: Anna Lerner Nezbit recently highlighted some vital research by Jennifer Turlik and The core takeaway there is pretty sobering Right now, the climate harms caused by AI are highly measurable and immediate.

00:01:43: But the benefits aren't?

00:01:44: Exactly!

00:01:45: The climate benefits remain largely speculative.

00:01:48: at this point It is a deficit from day one

00:01:50: Which means it's definitely not sustainable by default.

00:01:53: Turlix suggests that every single time of business decides to use an AI model They need to calculate the net climate impact

00:01:59: Literally subtracting the speculative Climate Benefits From the guaranteed climate harms.

00:02:05: But the underlying problem there is that the industry's dramatically undercounting those harms to begin with, right?

00:02:10: Oh vastly undercount them.

00:02:12: Ludovic, Subran and Patrick H both pointed to this recent data from Alliance Research.

00:02:17: It shows that global data center emissions are actually fifty-seven percent higher than the International Energy Agency's official estimates.

00:02:25: I saw that figure And my jaw just dropped!

00:02:28: I mean fifty seven percent is not a rounding error.

00:02:30: How does industry have a blind spot?

00:02:32: That massive?

00:02:33: Well

00:02:33: it all comes down where you draw boundary on your spreadsheet.

00:02:36: Historically The Industry only measured scope two emissions

00:02:40: Meaning just the operational power.

00:02:42: Yeah,

00:02:43: The actual electricity sucked out of a wall when servers are running.

00:02:47: but Alliance research expanded that boundary to include full value chain especially embodied carbon.

00:02:52: Oh right.

00:02:53: so emissions generated from mining silicon manufacturing hardware

00:02:58: Pouring concrete for steel racks all it

00:03:02: Wow, so if you ignore the building in the box... ...you're missing more than half of the picture.

00:03:06: Exactly!

00:03:07: But okay let me play Devil's Advocate here for a second because Bernhard Lorenz just published a book called Green AI.

00:03:12: Right

00:03:13: I'm familiar with that one

00:03:14: And his argument is that AI Is The Ultimate Optimization Engine.

00:03:19: Like, if we unleash it on smart grids or global shipping logistics he estimates that could reduce global emissions by one point four gigatons.

00:03:28: That's

00:03:28: a huge number!

00:03:29: Right...that would easily wipe out its own infrastructure footprint.

00:03:32: You spend little electricity to save an ocean of oil basically

00:03:35: And Lorenz makes really powerful case for potential absolutely.

00:03:39: But potential doesn't automatically equal reality.

00:03:42: Yeah Fair point.

00:03:44: An AI model optimizing the logistics route doesn't automatically decommission a coal plant, you know.

00:03:50: To actually materialize those net savings an organization needs ruthless

00:03:55: governance.".

00:03:56: Which brings us to Aparna Kaye right?

00:03:58: She brought this up recently...

00:03:59: Yes she argued that carbon intensity has to become a board level non-functional requirement.

00:04:05: Let's break down for the listen really quick!

00:04:07: A functional requirement is like um.. The app needs process of payment.

00:04:12: A non-functional requirement is usually about speed or security, like it must process the payment in under one second safely.

00:04:19: Right and a partner saying carbon efficiency needs to be the third pillar right there alongside them?

00:04:24: It has

00:04:25: to be mandated!

00:04:26: Exactly if a board signs off on a brilliant new product roadmap but doesn't demand an auditable software carbon intensity score along side it...the company's net zero pledge is just an illusion.

00:04:38: It never actually reaches the engineers building,

00:04:41: which puts us on a massive collision course.

00:04:43: because if boards Actually start mandating these limits The engineers have to confront.

00:04:48: The physical reality of where this tech lives

00:04:51: right?

00:04:51: These models aren't magically floating in a cloud somewhere.

00:04:53: No

00:04:54: They're sitting inside massive industrial buildings connected to stressed power grids.

00:04:58: the physical constraints You know energy water and geography those are becoming the defining bottlenecks for the whole sector.

00:05:06: You're seeing intense friction between the energy grids required to power these chips, and the water supplies required to keep them from literally melting.

00:05:14: And suddenly geography is the most important tech decision you can make!

00:05:18: Shabab Kohi proved this beautifully... Oh with

00:05:20: his open source tool right?

00:05:22: Yes

00:05:22: he built a tool.

00:05:26: When he ran the numbers, He found that simply taking a workload and deploying it in a data center in Stockholm instead of Tokyo resulted In a ninety-eight percent reduction in carbon emissions.

00:05:37: That is just wild!

00:05:39: Ninety eight percent...

00:05:40: And It's The Exact Same Code, the exact same processing time.

00:05:43: but the underlying grid in Sweden Is heavily powered by hydro and nuclear whereas Tokyo relies heavily on fossil fuels.

00:05:50: So that figure really proves that choosing your availability zone is A high leverage climate decision.

00:05:56: But what happens when those green grids simply run out of capacity?

00:06:01: That's the real issue, isn't it.

00:06:03: Yeah Robert Mueller highlighted this situation unfolding in Texas.

00:06:06: right now The grid operators there are essentially telling data center developers sure you can build here but we Can't power you for years.

00:06:16: bring your own electricity.

00:06:18: Wait

00:06:18: how do you just bring your on power to a hyperscale data Center?

00:06:22: Are they building their own nuclear plants

00:06:25: eventually?

00:06:25: maybe.

00:06:27: But right now, it's sparking this huge boom in behind-the-meter generation.

00:06:31: Facilities are bypassing the grid entirely by installing massive onsite gas turbines

00:06:37: like the solid oxide fuel cells

00:06:39: exactly Like those manufactured by bluen energy.

00:06:42: They're securing their own base load power because they simply cannot afford to wait for public utility upgrades

00:06:47: but even if you generate your own power You still have to deal with a heat.

00:06:50: Louis Tossi laid out the brutal engineering trade-off between power and water when it comes to cooling these server farms.

00:06:56: It's a fascinating dilemma!

00:06:58: It really

00:06:58: is, if you use an air cooled chiller You need a tremendous amount of electricity To run those massive fans but you used almost zero water

00:07:05: right?

00:07:06: And If you switch to a water cool chiller

00:07:08: its vastly more energy efficient But it evaporates roughly twenty two thousand cubic meters Of water every single year.

00:07:15: so your basically forced to pick your poison based on local scarcity Like, is the local community more strapped for electricity or for drinking water?

00:07:23: Exactly.

00:07:23: And sometimes public perception forces the absolutely wrong choice.

00:07:28: Profile care shared this wild story about a Google facility down in Chile.

00:07:32: Oh I saw this one.

00:07:33: There was intense activist pressure protesting The data center's water usage which apparently based on some exaggerated math.

00:07:41: So Google caved and switched facilities design from water cooling to air-cooling.

00:07:47: Let me guess the outcome The local grid had to supply way more power.

00:07:51: Much worse than that!

00:07:52: By switching to air cooling in THAT specific climate, the energy demands COMPLETELY skyrocketed.

00:07:58: That meant the local power plants had burned significantly more fuel.

00:08:01: Oh

00:08:01: no...

00:08:02: Yeah it created a massive spike in CO² and local air pollution.

00:08:06: The activists celebrated a win on water conservation but the holistic climate impact was total disaster.

00:08:12: That really underscores the danger of single metric optimization.

00:08:16: Yeah, you squeeze the balloon in one place and it pops somewhere else.

00:08:19: entirely

00:08:20: right

00:08:20: but You know we also can't let the tech giants off The hook here because their holistic metrics aren't looking great either.

00:08:27: James Martin and Will Nordberg pointed out the stark numbers in Google's latest environmental report.

00:08:32: And what did that show

00:08:33: since?

00:08:34: Their overall emissions have nearly doubled and they're water withdrawal has almost tripled.

00:08:40: But wait, Google constantly publishes research showing how much more efficient their specific AI models are getting.

00:08:46: I think they just claimed that their median text prompt uses thirty-three times less energy than it used to.

00:08:52: It's a classic case of Jevon's paradox The rebound effect.

00:08:55: Yes the individual prompt is drastically more efficient but because its cheaper and faster we're integrating into absolutely everything.

00:09:03: So this sheer volume usage scales up exponentially

00:09:06: Exactly And completely wipes out efficiency gains.

00:09:09: This is exactly why the debate over what constitutes greenwashing, so fierce right now.

00:09:14: So if hyperscalars are hitting a wall with grid capacity and cooling them as this constant environmental trade-off The industry's being forced to rethink the physical building itself Right?

00:09:26: We're moving from just Building data centers To practicing industrial symbiosis.

00:09:31: Industrial symbiosis Is perfect term for it.

00:09:33: It about looking at a data center Not as parasite on city grid But as potential donor.

00:09:40: Emanuella Grandi shared a brilliant case study out of Milan.

00:09:43: This is the Equinix partnership, correct?

00:09:45: Yeah.

00:09:45: Equinics partnered with The Local Energy Company A-to-A.

00:09:48: instead just using chillers to vent all that server heat uselessly into the atmosphere they're actually capturing it

00:09:53: and They pump it directly Into this city's district heating network.

00:09:57: exactly Using large scale heat pumps they redirect at thermal ways To heat over twenty one thousand homes.

00:10:03: Wow yeah It avoids Over three hundred forty five thousand tons Of CO two annually.

00:10:08: The data center goes from being a mass of liability to an active thermal contributor.

00:10:13: To the city's infrastructure.

00:10:14: that is such a smart way to solve the operational waste.

00:10:17: But what about the embodied carbon we talked?

00:10:19: About earlier, the concrete and steel?

00:10:22: well Carl rave has been tracking the rising interest in wooden data centers.

00:10:26: Wooden

00:10:26: data centers really yeah.

00:10:28: architects are turning to cross-laminated timber or CLT.

00:10:33: By swapping out traditional concrete and steel for engineered wood, they can radically slash the upfront carbon footprint of the building itself.

00:10:40: That's incredible!

00:10:42: And that mindset is tripling down to a hardware life cycle too.

00:10:45: Daniel Bukley pointed out that the concept of circular IT is finally maturing.

00:10:49: It used to just mean ethical disposal like making sure a server didn't end up in landfill.

00:10:54: Right...and now?

00:10:55: Now it's about strategic value creation

00:10:58: meaning they are actually refurbishing and redeploying these chips instead of just shredding them.

00:11:03: Yes, exactly.

00:11:04: through secure data erasure and refurbishment organizations are realizing that turning e-waste back into functional technology provides measurable environmental impact.

00:11:13: And you know it looks fantastic.

00:11:15: on an ESG report.

00:11:16: Okay so we've got wooden buildings recycled heat & refurbished servers but at the end day hardware is just a vessel.

00:11:23: The real engine driving this massive energy consumption is the software running inside those servers.

00:11:29: And, this brings us to one of the most exciting shifts happening right now.

00:11:32: Green's software is aggressively converging with FANOPS.

00:11:36: This where environmental mandate finally gets some financial teeth.

00:11:40: VJ Zende phrased it perfectly.

00:11:42: He said a bloated cloud bill Is essentially carbon in disguise.

00:11:46: That framing makes so much sense.

00:11:49: If you have an idle microservice spinning in AWS, it's burning your quarterly budget but its also quite literally burning coal.

00:11:57: The incentives are finally aligned.

00:11:59: when a Phenops team right sizes and organizations cloud infrastructure to save money they inadvertently shrink the carbon footprint.

00:12:06: Cloud cost optimization and green computing or becoming the exact same discipline.

00:12:11: Naveen Balani introduced concept that illustrates this perfectly which he calls forever data.

00:12:16: Oh, this is a great concept.

00:12:17: Yeah

00:12:17: he was analyzing how companies manage routine AI upgrades.

00:12:20: let's say a dev team upgrades their API to a newer better embedding model.

00:12:25: it sounds completely harmless but

00:12:26: nobody puts delete the old data on the deployment checklist do they?

00:12:30: Nope so these massive outdated datasets just sit there in cloud storage forever quietly drawing power and racking up hidden storage costs.

00:12:39: It's the digital equivalent of paying rent on an empty warehouse and just leaving all the lights On.

00:12:44: yeah, And the inefficiency isn't Just in storage it's In how we compute.

00:12:48: you brought up that seven forty-seven analogy early.

00:12:50: Oh

00:12:50: yes from The green i o munit conference.

00:12:52: right From a fascinating discussion Edward Dalharu started there.

00:12:56: He was analyzing modern network automation, right?

00:12:59: He saw engineers taking a very simple query like just checking a BGP session which is essentially looking at the internet's basic postal routing

00:13:07: and they were sending that query to hyperscalar large language model.

00:13:12: They were firing up hundreds of billions of parameters, spinning up massive GPUs just to parse a highly structured binary networking

00:13:19: task."

00:13:19: Which

00:13:20: is insane!

00:13:21: The daughter pointed out that if you route the task into an optimized on-prem small language model and SLM it accomplishes the exact same thing.

00:13:29: using fractions of one watt.

00:13:30: You really have right size the intelligence

00:13:33: And it goes even deeper than the model size.

00:13:36: Sud Aslan shared research showing that the underlying programming language you type in makes a staggering difference.

00:13:42: Yeah,

00:13:42: her team tested at Random Forest Machine Learning Model right?

00:13:45: Yes Just by writing it into R programming languages instead of Python and switching precision numbers from Float sixty four down to float thirty two They cut models energy consumption by seventy five percent.

00:13:58: That's a massive reduction for just tweaking a few lines of code.

00:14:02: And, for the listener who isn't writing code every day it really helps to understand why that works.

00:14:07: Walk us through it!

00:14:08: Well Float sixty-four means the computer is using sixty four bits memory to store incredibly precise decimal numbers.

00:14:15: by dropping into float thirty two you instantly have the memory bandwidth and processor strain.

00:14:20: And the kicker is, for most practical applications... ...the loss in accuracy is completely unnoticeable.

00:14:26: Exactly!

00:14:27: So hearing all this about idle compute and processor strain my brain immediately jumps to an architectural solution.

00:14:33: If idle servers are the enemy shouldn't we just make everything serverless?

00:14:37: The serverless debate?

00:14:38: Right because a serverless architecture scales down to absolute zero when no one's using the app.

00:14:45: Doesn't that magically solve energy problem?

00:14:48: It's a very logical hypothesis, and hyperscalars certainly market serverless computing as the ultimate efficiency hack.

00:14:54: But Wilco Bergraf provided a really sharp warning against treating it is a silver bullet.

00:14:59: What's the catch?

00:15:00: Green software design is incredibly contextual.

00:15:03: to give you that instant zero latency experience when your code finally triggers The cloud provider has to keep massive cools of CPU and memory constantly spinning in the background Ready to allocate at a milliseconds notice.

00:15:15: Oh, I see so the orchestration overhead is immense.

00:15:18: exactly.

00:15:19: So if your workload requires steady constant computing Forcing it into a serverless architecture actually creates massive digital waste Compared to a traditional optimized server?

00:15:29: You really have to understand this specific physics of your own work load.

00:15:32: But you know maybe the ultimate green software principle is much simpler than any of us.

00:15:37: Valerius Asalis argued that the most underrated question in all of IT is simply Do we really need to build this feature at all?

00:15:44: In a tech culture obsessed with constant shipping and feature bloat, that is almost taboo question.

00:15:49: But she's totally right!

00:15:51: The most sustainable code —with a carbon footprint of absolute zero—is the code you never

00:15:56: write.".

00:15:57: Which brings us into our final hurdle today... If green software is this contextual… And if it's so easy for companies to hide behind these market-based ESG reports….

00:16:07: How do we actually hold industry

00:16:08: accountable?!

00:16:09: The answer we are seeing across the board is a massive push toward rigorous measurement in public standards.

00:16:16: You could really feel this shift at the Green IO Munich event.

00:16:19: Yeah, leaders like Gail Duvez and Baptiste Paravaccini made it very clear sustainable computing is graduating from corporate PR to a strict engineering framework...

00:16:28: ...the era of just raising awareness is over!

00:16:31: We're moving into scalable auditable implementation

00:16:34: And we see the public sector step up to enforce that which is huge.

00:16:37: Hannah Lipstein recently introduced a new handbook designed entirely for government procurement teams.

00:16:42: Right,

00:16:42: it basically trains governments to translate the environmental costs of AI—the water evaporation and grid intensity that embodied carbon into aggressive actionable questions in their request for

00:16:55: proposals.

00:16:56: That is massive lever for change!

00:16:58: governments procure billions of dollars of technology.

00:17:02: The moment they start rejecting bids because the vendor can't provide an auditable software carbon intensity score,

00:17:13: Money talks.

00:17:14: It really does, so before we wrap up today's deep dive there is one critical piece of advice from Paul Noon that every single person listening needs to internalize if you are sitting across from a vendor or evaluating your own company's impact.

00:17:27: You have to demand location-based emissions figures not market based emissions figures.

00:17:32: Yes This is perhaps the single most important distinction in carpet accounting right now.

00:17:37: walk us through why?

00:17:38: That specific gap matters so much.

00:17:40: Well, market-based accounting is an administrative sleight of hand.

00:17:44: It allows a company to claim their data center as one hundred percent renewable because they purchased a bunch of solar renewable energy certificates in one country even while the actual physical servers are plugged into coal heavy grid and completely different countries.

00:17:57: That's incredibly misleading!

00:17:59: Location based emissions however measure absolute physical reality.

00:18:04: it tells you exactly how carbon intents of electricity flowing into that specific building at exact hour.

00:18:10: Wow.

00:18:11: Yeah, as Paul Noon warned the gap between market-based claims and location based reality is exactly where all.

00:18:26: And

00:18:37: before you go, I want to leave with one final thought to Mollover.

00:18:41: Building on a really provocative point raised by Petra Heinerbrock She was analyzing those throwaway AI generated visuals we see flooding our LinkedIn feeds every day.

00:18:51: Oh like the synthetic office chaos

00:18:53: ones?

00:18:54: Yeah slightly off cartoon crowds.

00:18:56: Generating single image uses about as much energy As leaving an LED ball-bond for eight minutes.

00:19:02: That's

00:19:02: crazy!

00:19:03: It doesn't sound like a lot until you multiply it by the millions of rejected, regenerated and instantly forgotten prompts happening globally every single day.

00:19:12: So what if The Future Of Digital Sustainability isn't just about making our server hardware more efficient?

00:19:17: What If Its About Fundamentally Auditing Our Digital Waste?

00:19:20: Imagine A Future Where Companies Are Required To Report Unjustified Computations On Their Balance Sheets Sitting Right Next to Their Financial Debt.

00:19:28: Would That Change The Way You Prompt AI Tomorrow?