Best of LinkedIn: Sustainability & Green ICT CW 24/ 25

00:00:00: This episode is provided by Thomas Allgaier and Frennus based on the most relevant LinkedIn posts about sustainability in green ICT, in CW-TwentyFour and TwentyFive.

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

00:00:19: you can find more info.

00:00:22: So, you know when we talk about digital transformation these days there is this really persistent almost comforting illusion of weightlessness.

00:00:30: right yeah We use all these words like virtual or wireless, or the cloud.

00:00:35: Exactly we just kind of imagine our data floating up there in the sky somewhere which

00:00:39: is funny because The reality is that the Cloud Is incredibly heavy.

00:00:42: I mean it's constructed out of millions Of tons of concrete and steel And Just massive amounts of physical electricity In water.

00:00:49: It is!

00:00:50: Today were really looking at the sustainability And green ict trends That are just lighting Up linkedin right now Because for anyone working in tech The conversation has officially shifted.

00:00:59: Oh, completely.

00:01:00: We are way past the era of you know those abstract corporate sustainability claims.

00:01:06: we're deeply into the hard execution phase now right?

00:01:09: We're talking rigorous uncompromising discipline around energy water infrastructure and software efficiency.

00:01:17: so welcome to The Deep Dive.

00:01:19: today were giving a smart totally no fluff breakdown on actual hard engineering choices in governance realities shaping the future tech.

00:01:29: Because if you really want to understand how your digital supply chain impacts the physical world, You have to look at where code hardware and policy actually intersect.

00:01:38: Yeah And let's start with the Code because the rapid adoption of GenAI is just inflating IT budgets in emissions everywhere

00:01:45: It really is.

00:01:46: What caught my eye recently was that the easiest most immediate fixes for this massive energy spike.

00:01:51: They aren't about building entirely new futuristic data centers.

00:01:55: Right, the low-hanging fruit is actually right there at the data and software level?

00:01:58: The way we format our inputs.

00:02:00: it's just incredibly wasteful.

00:02:02: Yeah Mark Butcher shared this really practical insight that perfectly illustrates this kind of invisible waist.

00:02:07: Oh!

00:02:07: The Microsoft report test he did

00:02:09: Exactly So.

00:02:09: he wanted to test how much unnecessary energy we consume Just by formatting data poorly for AI models.

00:02:17: He took a ninety page Microsoft sustainability report and he fed it into an AI in two different ways

00:02:23: just to measure the difference in computational cost, right?

00:02:26: Yeah

00:02:26: So first.

00:02:27: He uploaded as a plain text file And then he uploaded it as the original PDF

00:02:32: and the difference was massive wasn't huge.

00:02:35: The text version consumed about fifty six thousand tokens But that original pdf consumed over two hundred and ten thousand tokens.

00:02:43: Wow!

00:02:43: So it's nearly four times as heavy for the exact same underlying information?

00:02:47: Exactly, And The reason why is fascinating...the AI model doesn't just like Scan the hidden text layer of a PDF.

00:02:54: It essentially renders A high-resolution picture Of every single page

00:02:57: right.

00:02:57: so you're Forcing this incredibly complex neural network to process ninety pages of pixels and colors And formatting rather than just reading The words

00:03:05: which is so inefficient.

00:03:06: and mark pointed out the terrifying math behind them.

00:03:08: Yeah,

00:03:09: because if You scale that invisible bloat up To a midsize organization let's say ten thousand employees Right?

00:03:15: Right and they're each uploading Just say two pdfs today to an internal ai assistant.

00:03:21: You are suddenly looking at ninety-two billion wasted tokens a year.

00:03:26: That is wild, I mean that four times token bloat costs massively in computing power and energy drawn from the grid not to mention actual operational dollars.

00:03:35: It's basically digital equivalent of shipping heavy wooden crate via next day air freight.

00:03:41: just deliver single postcard.

00:03:43: That is a perfect analogy.

00:03:45: You're paying enormous fees for packaging you don't even need, but fixing our file formats only solves the input side of the equation.

00:03:52: Right?

00:03:52: what about AI models themselves?

00:03:54: Uploading a clean text file was great But we are still throwing massive amounts of compute at relatively simple problems.

00:04:01: Yeah and Aiden Meir Mohammadi shared a brilliant metaphor He's really pushing for what he calls lean agentic AI.

00:04:07: Oh I love his posts!

00:04:08: He basically noted that you wouldn't deliver pizza in Ferrari.

00:04:11: Exactly yet that is exactly our default behavior in the tech industry right now.

00:04:15: We use these massive energy-hungry frontier models with trillions of parameters just to do basic text summarization

00:04:22: or simple coding tasks, it's just overkill.

00:04:26: Asan Porjavad actually expanded on this arguing that the fundamental goal for AI engineering teams has to change

00:04:32: because It no longer about chasing the highest possible accuracy at literally any cost.

00:04:38: Because squeezing out that last one percent of accuracy usually requires doubling the energy costs, so Asan says.

00:04:45: The new goal is finding the Pareto Frontier.

00:04:48: for those who might not know in economics the Pareta frontier Is basically that absolute optimal point where you can't improve One variable without sacrificing another right

00:04:58: and an AI.

00:04:59: That means finding this specific lightweight model that perfectly balances business performance with actual energy efficiency

00:05:06: And there's concrete proof that this discipline works.

00:05:08: Right?

00:05:08: I was reading this great breakdown from Gudena Guzelay who recently optimized a healthcare machine learning model.

00:05:14: Oh yeah, didn't her team decide to basically reject the default computational setting?

00:05:18: Yeah they dug right into math and switched the data format from float sixty four down to flow thirty two

00:05:23: Which, for anyone outside of data science just means reducing the level of mathematical precision that a computer is forced to hold in its active memory.

00:05:31: Exactly!

00:05:32: A float sixty-four requires the computer process sixty four bits—of ones and zeros—for every single decimal point

00:05:39: And for vast majority applications you simply do not need sixty four bit's of decimal precision.

00:05:45: get an accurate result.

00:05:47: Thirty-two bits is plenty.

00:05:48: It really is, and by making that simple switch and pruning some unnecessary features from their data set her team reduced the model's energy consumption And its execution time by over forty eight percent.

00:05:59: That's incredible.

00:06:00: They cut the energy footprint nearly in half with zero meaningful loss in predictive performance.

00:06:06: Yeah Furthermore, where you choose to run those optimized models matters just as much as the code itself.

00:06:11: Oh absolutely!

00:06:12: Will Nordberg highlighted this really well.

00:06:14: he pointed out that if you self-host open weight coding models on renewable infrastructure in Finland it actually costs fifteen to twenty times less than relying on endless API calls to US cloud providers.

00:06:26: Fifteen to Twenty Times Less.

00:06:27: That's huge.

00:06:28: And in Finland, the waste heat generated by those servers is actually captured and used to warm local residential homes.

00:06:35: That's the dream right?

00:06:37: But here is the hard truth for anyone planning a digital transformation strategy No matter how much you optimize your code or compress your PDFs, that data eventually has to be processed on physical silicon.

00:06:51: Right and That hardware layer is currently slamming into a physical wall but no software update can magically fix.

00:06:58: Yeah for years engineers viewed electrical power as just like an optimization variable.

00:07:03: today Power is the primary immovable constraint.

00:07:07: Michael Friedman made an incredible observation about this regarding the hyper-stalers, you know?

00:07:12: The massive cloud providers running the internet.

00:07:13: What did he see?

00:07:15: He noted that they are actively denying server expansion requests from their own internal teams right now.

00:07:20: and it's not because of a silicon chip shortage or lack warehouse space.

00:07:24: Let me guess... It is regional power limits.

00:07:26: Exactly!

00:07:26: They're denying request simply because of power constraints.

00:07:30: The compute layer & physical plant are completely inseverable.

00:07:33: You literally cannot just drop a new server rack into a facility if the local grid physically can not deliver the megawatts to turn it on.

00:07:42: You cant draw blood from stone and you cant draw gigawatts form a maxed out grid.

00:07:47: This physical constraint is forcing massive rapid shift in national policy especially for regions that traditionally host these data hubs

00:07:55: Like Ireland, for example.

00:07:57: Stacy Keough shared a really interesting update on their infrastructure crisis.

00:08:01: Right because Ireland is a major global hub for tech giants.

00:08:05: Data centers there already consume around twenty percent of the nation's total electricity.

00:08:09: Twenty

00:08:10: percent is staggering.

00:08:11: so to protect the national grid from basically collapsing under the weight of AI Ireland is instituting a strict Bring your own power condition for new data center developments.

00:08:22: Bring you're own Power BYOP.

00:08:24: essentially the government is saying You cannot just plug a new gigawatt facility into The public utility grid anymore.

00:08:30: right if you want to build A data Center there, you have To secure your own on-site power generation or fund brand New renewable energy contracts that add fresh power to the grid

00:08:39: instead of just siphoning off what the citizens need.

00:08:42: But you know, the friction isn't just happening at the macro national grid level.

00:08:47: it's happening street by street.

00:08:48: Yeah!

00:08:48: It is about actual people living next door to these facilities.

00:08:51: Right.

00:08:52: Mark P highlighted this truly historic vote recently in Monterey Park California.

00:08:57: The local residents there voted eighty six percent In favor permanently banning any new data centers from being built into their city.

00:09:06: I mean, that is not a narrow political margin.

00:09:08: That's a resounding unified rejection by an entire community!

00:09:12: Exactly.

00:09:13: And the residents were citing severe concerns over rising local electricity bills massive drawdowns on municipal water supply and air quality impacts.

00:09:21: Oh because of heavy backup diesel generators these facilities require to ensure uptime?

00:09:27: Yeah

00:09:27: people are finally realizing data centers aren't just quiet invisible office buildings storing emails.

00:09:33: they're resource-intensive industrial facilities.

00:09:36: They really are, and to help communities navigate these massive zoning fights Ashley Anderson built something incredibly clever using an AI coding tool called Claude Code.

00:09:46: Oh I saw this.

00:09:47: she created an interactive dashboard for a proposed data center site at a former coal plant in Pittsburgh right?

00:09:53: Yes!

00:09:54: And instead of a town hall just devolving into a yelling ratch residents could actually look.

00:10:02: So

00:10:04: it democratizes the impact analysis.

00:10:06: Exactly, It allows residents to compare the data center's real footprint Its carbon emissions its exact water use The tax revenue generated against seven alternative futures for that exact same plot of land.

00:10:17: That

00:10:18: is so smart.

00:10:19: They could compare the data center's impact side by side against building, say a restored public park or a battery storage facility.

00:10:27: It completely changes the conversation from this highly polarized binary of build it-or don't to helping communities make deeply informed comparative choices about their own backyards.

00:10:37: and one Of the biggest reasons communities are pushing back so hard brings us to probably The most difficult environmental challenge in the tech sector right now cooling.

00:10:45: oh

00:10:46: yeah AI computation runs unbelievably hot.

00:10:50: Racks of GPUs generate staggering amounts of thermal energy, and cooling those servers is a massive bottleneck.

00:10:57: WillR actually posted this hilarious but scientifically sound takedown Of the classic sci-fi dream of putting data centers in space.

00:11:05: Oh I love this one.

00:11:07: The idea always sounds so elegant, right?

00:11:09: Space is freezing cold.

00:11:10: So let's just orbit our servers and let the universe cool them.

00:11:13: Right.

00:11:13: but Will points out what physicists call the vacuum paradox.

00:11:17: Here on Earth we use air or water to carry heat away from the servers.

00:11:21: through convection

00:11:22: The fluid or gas moves over hot metal And pulls the heat away

00:11:26: Exactly.

00:11:27: But space a vacuum.

00:11:29: There was no air or Water To absorb that thermal energy

00:11:32: Without physical mediums to transfer the heat into.

00:11:35: The thermal energy just gets trapped right next to the hardware.

00:11:38: Space essentially acts like a giant thermos, so instead of freezing an orbiting data center would just rapidly suffocate and melt in its own generated heat

00:11:46: Which means we are stuck cooling our hardware Right here on Earth.

00:11:50: Yep And the engineering solutions for that Are getting wild.

00:11:54: Larissa Stetsenko posted about An incredible project where China launched A commercial underwater data center off the coast Of Shanghai.

00:12:02: Wait Underwater?

00:12:04: They literally sank the server modules into the ocean.

00:12:07: Ten meters below the surface.

00:12:09: Larissa described this approach as a classic tree-IZ solution.

00:12:14: T or I Z, that's the Russian acronym for that highly structured engineering framework focused on inventive problem solving right?

00:12:21: That is one.

00:12:22: And one of the core principles of TRD-IZ is that instead fighting a contradiction with brute force like using expensive mechanical chillers to fight heat, you resolve it by changing the baseline environment itself.

00:12:33: You place this system where cold already naturally exists?

00:12:36: Exactly!

00:12:37: By dropping the data center into fifteen degrees celsius seawater they achieve near perfect passive energy efficiency.

00:12:44: Okay, it sounds brilliant on a spreadsheet but you really have to look at the physical reality.

00:12:48: Doesn't that just boil the ocean locally?

00:12:50: Yeah!

00:12:50: That's the catch.

00:12:51: all of that concentrated thermal energy has to go somewhere into this surrounding marine ecosystem

00:12:56: And environmentalists are already sounding the alarm about potential for massive marine ecological dead zones forming around these submerged facilities.

00:13:05: Because of constant localized heat plumes disrupting water temperature.

00:13:10: Right, but then again terrestrial cooling has its own dark side that corporate reports rarely mention.

00:13:16: We constantly hear the buzzword zero water cooling and it sounds so perfectly eco-friendly to a consumer.

00:13:22: It does, but Mexica Allgood posted a really stark warning about what is actually running through those closed loop pipes.

00:13:30: Yeah these Zero Water systems often rely on hundreds of thousands of gallons Of highly toxic chemical coolants.

00:13:36: we're talking industrial antifreeze motor oils And forever chemicals.

00:13:40: Wow So if there's a catastrophic leak at one of These facilities The damage To local soil municipal water tables could just devastate a community for decades.

00:13:49: Exactly, we are essentially trading the risk of municipal water consumption for the risks of toxic chemical disaster.

00:13:55: It is a terrifying trade-off but the discipline creating this massive demand data science also proving to be our sharpest tool for mitigating it.

00:14:04: How so?

00:14:05: Well

00:14:05: Kirk Metler highlighted how reinforcement learning agents are being deployed to analyze sensor data inside hyperscale facilities.

00:14:12: Oh, like AI managing the building's climate control!

00:14:15: Yeah

00:14:16: a reinforcement learning agent analyzes thousands of temperature and airflow sensors simultaneously constantly making micro adjustments to fan speeds in valve pressures in real time.

00:14:26: So by letting the software teach the building how to cool itself dynamically, what kind of savings are they seeing?

00:14:31: They're

00:14:32: cutting cooling energy by thirty-to forty percent.

00:14:34: That

00:14:34: is massive!

00:14:35: And on hardware manufacturing side, Curious Woody highlighted a company called XNRGY that has developed specific zero water loss chillers designed to actively avoid those environmental spill risks while still maintaining efficiency.

00:14:51: Luis Timodio also shared a highly impressive metric, noting that AWS achieved the global water efficiency average of furo point one two liters per kilowatt hour through smarter operations.

00:15:02: But you know this brings us to a massive overarching problem with everything we've discussed today.

00:15:07: AWS claims nero-point one.

00:15:09: two leaders X and RGY claim zero water loss.

00:15:13: various cloud providers claim hundred percent renewable energy.

00:15:17: Right, how do you actually verify who is telling the truth?

00:15:19: Exactly.

00:15:20: The digital supply chain is incredibly opaque.

00:15:23: Sean O'Connell shared a really painful real-world lesson about this.

00:15:27: Oh with his website hosting Yeah

00:15:29: For years Sean proudly used a verified green hosting provider for his website.

00:15:34: He even displayed a green credential badge right on his home page.

00:15:37: And let me guess that boutique hosting provider was inevitably acquired by A massive corporate giant

00:15:44: Yep And the new parent company quietly ignored all verification requests from The Green Web Foundation.

00:15:49: So Sean lost his green credentials without ever touching a single line of his own code.

00:15:55: Because his upstream supplier changed their internal policies, His own sustainability claims suddenly became false.

00:16:01: It just proves that a company's sustainability is only as strong as its weakest vendor.

00:16:05: You have to constantly audit your digital supply chain

00:16:08: And when you actually perform those audits, they reveal some pretty surprising things about the platforms we trust most.

00:16:15: Julie Schiller posted results of a technical sustainability audit she performed on LinkedIn's own website.

00:16:21: Oh!

00:16:21: The irony...the very platform where all these corporate leaders are posting their ESG commitments?

00:16:27: What did she find?

00:16:28: Well She found that sixty-one point six percent every page load was just heavy JavaScript.

00:16:34: And heavy JavaScript is a massive carbon problem because it offloads the computational burden onto the user, right?

00:16:40: Exactly.

00:16:41: The platform server isn't doing that heavy lifting... ...the users' local smartphone or laptop browser has to download, parse and execute all of this bloated code which

00:16:49: drains billions of device batteries globally pushing energy costs straight on into the local power grids of their users.

00:16:56: Shiller also noted that main content took eight point two seconds to appear but crucially she found there green hosting status was completely unverified.

00:17:05: Wait, really?

00:17:06: But they run on Microsoft Azure which has very public renewable energy commitments.

00:17:11: They do but the infrastructure itself wasn't independently verified by The GreenWeb Foundation.

00:17:16: It perfectly highlights how our massive digital infrastructure mostly runs completely unexamined in the background.

00:17:24: Which is exactly why The International Community has finally stepping into force visibility!

00:17:29: We saw several posts discussing how Antonio Guterres, the UN Secretary General alongside advocates like Mathieu Francois are pushing for a major AI environmental transparency initiative

00:17:40: at because Mathieu Francois makes a crucial point.

00:17:45: Right now, every single government framework assumes we can actually measure and compare the environmental impacts of different AI

00:17:52: models.".

00:17:52: But the truth is...we can't!

00:17:54: We just do not have interoperable standards.

00:17:57: So the UN initiative is pushing to make measurement mandatory globally standardized an entirely independent of the tech provider's own self-disclosure?

00:18:05: Because if you let the tech companies mark their own homework you are never going to get the full, unvarnished picture of environmental costs.

00:18:12: We

00:18:12: absolutely need independent metrics that travel across borders just like The Cloud does!

00:18:16: And if we connect all these pieces—the software bloat and hardware constraints… …the cooling chemicals... ...and lack transparency — there is a really profound thought from Wilco Bergerraff that kind-of encapsulates this entire deep dive.

00:18:32: What was his take?

00:18:33: Well, Woko shared this deeply personal story about growing up in the nineteen eighties and The Netherlands.

00:18:38: His family literally lived inside the grounds of a highly polluting factory where his dad worked as a guard.

00:18:44: Oh wow

00:18:44: Yeah.

00:18:45: He remembers the dust settling on the fresh laundry... ...the fish dying from illegal chemical dumping.. ..the lingering metallic smell-in-the air And severe asthma his families suffered.

00:18:57: And didn't that factory eventually relocate?

00:18:59: It did.

00:19:00: In the two thousands it packed up and moved to China, but Wilco noted something vital we constantly distance ourselves from the effects of our actions But the impact remains.

00:19:10: The factory moved with environmental damage on his family were already done

00:19:15: Exactly.

00:19:16: And today, Wilco is a solution architect focusing specifically on sustainable digital operations.

00:19:23: He argues that our current obsession with just measuring AI tokens and counting carbon credits Is way too narrow.

00:19:29: So what does he say?

00:19:30: true digital sustainability actually looks like?

00:19:32: He says, it isn't just about tweaking assorting algorithms energy use in isolation or buying enough solar credits to offset a massive data center.

00:19:41: It is about fundamentally reducing overall organizational waste

00:19:45: like looking at avoided work Or reducing legacy technical debt

00:19:49: right keeling inefficient software containers and stopping pointless duplicated processes entirely.

00:19:55: if you make A software system highly energy efficient but it Is fundamentally building useless things?

00:20:02: Wilco's core question for you, the listener is this.

00:20:05: Is your tech actually solving a real necessary problem or are you just efficiently maintaining waste?

00:20:11: That is the ultimate question

00:20:12: and it brings us right back to that illusion of weightlessness we started with because its digital We assume it leaves no permanent mark.

00:20:21: But we have to ask ourselves a harder question about the legacy of this infrastructure.

00:20:25: Yeah Just as Wilco physical factory left a permanent environmental scar on his hometown

00:20:31: What happens fifty years from now if the current AI bubble shifts or the underlying technology fundamentally changes?

00:20:37: Are we currently building that digital rust belt of the twenty-first century.

00:20:41: I

00:20:41: mean, We are pouring millions and tons of concrete permanently altering local water tables and rewiring national electrical grids for these massive data centers.

00:20:51: If the algorithms move on The physical footprint our digital obsession will still be sitting right there in someone's backyard.

00:20:58: The weight is very real, and the impact remains.

00:21:01: If you enjoyed this episode new episodes drop every two weeks.

00:21:05: Also check out our other editions on cloud digital products & services artificial intelligence in ICT and tech insights health tech defense Tech.

00:21:13: Thank You so much for joining us on this deep dive And don't forget to subscribe.