Best of LinkedIn: Sustainability & Green ICT CW 50 - 01

00:00:00: This episode is provided by Thomas Olguyer and Frennus based on the most relevant LinkedIn posts about sustainability and green ICT in CW-Fifty-Zero-One.

00:00:10: Frennus supports ICT enterprises with market and competitive intelligence decoding green software developments, benchmarking emerging standards, tracking regulatory shifts, and analyzing competitor strategies.

00:00:22: Welcome to the deep dive.

00:00:24: So this time we've gone through a whole batch of fresh insights from LinkedIn all focused on sustainability and green ICT right at the start of the year and Our mission really is to cut through all the abstract pledges.

00:00:35: You know the hopeful

00:00:37: goals right the noise

00:00:38: Exactly.

00:00:39: We want to find the measurable actionable steps that technical leaders and engineers are actually putting into practice right now.

00:00:44: And that shift from pledge to proof is I think the defining theme in all these sources.

00:00:49: If there's one central tension framing this whole deep dive, it is absolutely AI.

00:00:54: Oh, for sure.

00:00:54: We see it positioned as, you know, this huge environmental risk because of its sheer frankly, fighting energy hunger.

00:01:01: And at the same time, it's also the critical enabler for efficiency gains everywhere else.

00:01:05: And what's so fascinating is how the discussion has really moved beyond just the tech itself.

00:01:10: It's about organizational structure.

00:01:11: It's about new standards.

00:01:12: And community

00:01:13: resistance, actually.

00:01:15: That's shaping what actually gets built and what scales in twenty twenty six.

00:01:19: OK, let's unpack this.

00:01:20: Let's start right at the code level.

00:01:21: Theme one.

00:01:22: Green AI and software.

00:01:24: the efficiency imperative

00:01:26: right and we have to start by Quantifying that tension you just mentioned.

00:01:30: we hear all the time about AI's potential to optimize everything But the cost of building it is well, it's enormous

00:01:37: right.

00:01:37: Joseph Nogle have this great really tangible comparison generating a single five second AI video It uses as much electricity as running a microwave oven for over an hour.

00:01:48: Wow.

00:01:49: An hour of microwave time just for a five second clip.

00:01:51: That really, that puts it in perspective.

00:01:53: It does.

00:01:53: And then you scale that up.

00:01:55: Dirk Rodetzky cited that training a model like GPT-IV required an estimated fifty gigawatt hours of energy.

00:02:01: Fifty GWOH.

00:02:02: I mean, to put that in perspective, that's enough to power tens of thousands of homes for a year, right?

00:02:07: For

00:02:07: a whole year.

00:02:08: And then looking even further out, an analysis Bogdan Sapopsky shared reminds us that if this goes unchecked, ICT emissions could potentially reach a staggering fourteen percent of all global greenhouse gas emissions by twenty forty.

00:02:23: Fourteen percent.

00:02:24: The stakes just couldn't be higher.

00:02:26: No.

00:02:27: So, okay, the solution can't just be building more power plants.

00:02:30: It has to start with smarter code.

00:02:32: It feels like green coding is finally moving from like a niche thing to a core engineering discipline.

00:02:38: A

00:02:38: new minimum standard.

00:02:40: Exactly.

00:02:40: Walk of Samad highlighted this perfectly.

00:02:42: He argued that energy efficiency is now a critical non-functional requirement in NFR.

00:02:48: It has to be treated with the same rigor as security or performance.

00:02:52: It's not an afterthought.

00:02:53: And the games you can get, they're not just marginal, are they?

00:02:55: Not at all.

00:02:55: They're transformational.

00:02:56: Yeah.

00:02:57: Bogdan Sapofsky provided the numbers on this.

00:02:59: Optimizing your code, for instance, just by strategically favoring compiled languages like C and Rust can use up to forty-eight times less energy than interpreted languages like Python or Ruby.

00:03:13: That is a huge multiplier.

00:03:16: So for listeners who aren't, you know, deep in code architecture, why is that difference so massive?

00:03:21: It really comes down to overhead.

00:03:22: I mean, Python and Ruby, they need an interpreter.

00:03:25: That's another layer of software that reads and translates the code every time it runs.

00:03:29: Which is computationally expensive.

00:03:30: Super expensive.

00:03:32: C and Rust are compiled directly into machine code once, so they execute way more efficiently.

00:03:37: Minimal memory, minimal processing.

00:03:39: And when you apply that different across millions of transactions, the carbon savings just become massive.

00:03:45: That is powerful.

00:03:45: And the sufficiency thing, it isn't just about cutting costs for hyperscalers in the West.

00:03:49: No, not at all.

00:03:50: Tiamatka Prosper and Kwazama made a really strong argument that for African developers, green coding is essential.

00:03:56: It's not a luxury.

00:03:58: Right.

00:03:58: It's a necessity.

00:04:00: Where power is scarce, unreliable and expensive.

00:04:03: Writing leaner code is just fundamental to building systems that actually work and are affordable.

00:04:07: It proves that green code is just inherently better, more sustainable code, period.

00:04:12: Which brings us to this AI assistance paradox.

00:04:16: If AI is the problem, can AI also be the

00:04:20: fix?

00:04:21: That's a good question.

00:04:22: David Scott Bernstein shared a pretty positive story about using tools like Claude for collaborative refactoring.

00:04:27: He described it as a partnership.

00:04:29: A partnership?

00:04:30: Yeah, where the AI helps spot really complex code patterns and allow them to, you know, jointly discover cleaner architectural solutions.

00:04:39: It wasn't write the code for me, it was help me see where the inefficiency is.

00:04:42: That sounds like the ideal way to use it, you know, accelerating optimization.

00:04:46: But that ideal sits right next to a massive warning we saw in the sources.

00:04:49: It does.

00:04:50: Lord Asterwest cautioned very strongly against relying too heavily on predictive AI.

00:04:54: He called it vibe coding.

00:04:55: Vibe coding.

00:04:56: I like that.

00:04:57: It's where developers just blindly accept AI-generated snippets without understanding the logic or the architectural implications.

00:05:04: And he said, this just systematically degrades code quality.

00:05:07: And it creates unsustainable technical debt right

00:05:11: away.

00:05:11: Immediately.

00:05:12: And the risk isn't just architectural.

00:05:14: It's security-based.

00:05:15: That was the shocking part.

00:05:16: The

00:05:16: statistic, yeah.

00:05:17: The report he cited showed that AI-generated code samples introduced security vulnerabilities in an alarming forty-five percent of cases.

00:05:24: Forty-five percent.

00:05:25: You have to ask, who would accept a nearly fifty percent chance of security flaws just for a temporary speed bump in development?

00:05:32: It's insane.

00:05:33: It totally wipes out any initial productivity gain.

00:05:36: So it really raises a strategic question for leadership then.

00:05:39: You have to actively manage this.

00:05:41: You

00:05:41: do.

00:05:41: And we are seeing proactive steps being explored.

00:05:45: Max Missy and Paul Young share that GitHub is looking into AI agents for continuous efficiency.

00:05:50: So embedding sustainability checks directly into the CI CD pipeline.

00:05:55: Exactly.

00:05:55: Yeah.

00:05:56: Manaphital summarized it best.

00:05:58: He said, sustainability is now an architectural concern that technical leaders must embed in their strategy.

00:06:04: It has to be central to their roadmaps, not some add-on at the end.

00:06:08: That is a great transition.

00:06:10: So if efficiency is now an architectural requirement for software, what happens when we scale that software onto the massive physical infrastructure beneath it?

00:06:20: Let's move to theme two.

00:06:21: Green Cloud and data centers, the physical footprint.

00:06:25: Right.

00:06:25: If green AI is about the invisible footprint of code, data centers are the visible massive.

00:06:31: physical consumption challenge.

00:06:32: No question.

00:06:33: And Bosena Salaminsky's projection really underlines this.

00:06:36: Global data center electricity use could nearly double, hitting nine hundred and eighty terawatt hours by twenty thirty.

00:06:42: And that surge is driven almost entirely by AI's growth.

00:06:45: Almost entirely.

00:06:45: And that doubling put immediate palpable pressure on resources beyond just electricity.

00:06:50: Prasad Anumila highlighted the benefits of green cloud, but that all depends on having hyper efficient infrastructure.

00:06:55: And David Sturkenberg emphasized one of the most forgotten resources.

00:06:58: He just said data equals water.

00:07:00: That analogy is so crucial because we always forget about cooling.

00:07:04: Stuckenberg cited reports that a single one megawatt facility can consume over twenty five million liters of water

00:07:12: a year.

00:07:13: Twenty five million liters.

00:07:14: That's a substantial claim on local resources.

00:07:16: It

00:07:16: is.

00:07:17: And the problem is compounded by poor planning.

00:07:20: Graham Plaster pointed out that nearly seven thousand global data centers are located outside the temperature range that's recommended for efficient operation.

00:07:28: Wait, wait.

00:07:29: They are deliberately placed in suboptimal climates.

00:07:31: Why would they do that?

00:07:32: Well, largely for economic or political reasons, you know, proximity to fiber hubs, tax incentives, political stability.

00:07:39: So those factors are prioritized over climate suitability.

00:07:42: Exactly.

00:07:43: And if you build a massive facility in a region that needs intense constant cooling, you inherently inflate its water and energy footprint.

00:07:51: It's a trade-off that is becoming less and less sustainable.

00:07:53: And that leads right into this critical issue of transparency and local pushback.

00:07:58: because And when resources get strained, the public starts asking questions.

00:08:02: And if there are no answers, resistance grows.

00:08:05: Precisely.

00:08:07: Karen van der Zenden reported that major hyperscalers in the Netherlands are actively withholding their energy consumption data.

00:08:13: And what's the reason?

00:08:14: They're labeling it as business sensitive.

00:08:17: But that lack of transparency, especially in a region facing tight energy demands, it instantly turns a technical issue into a political and environmental liability.

00:08:26: And we are seeing that liability hit the balance sheet.

00:08:29: James Martin detailed how community opposition is actively stalling a staggering one hundred and sixty two billion dollars worth of U.S.

00:08:36: data center projects.

00:08:38: One

00:08:38: hundred and sixty two billion.

00:08:40: And this isn't just, you know, not in my backyard.

00:08:43: This opposition is often driven by real concerns over energy strain, local power bill hikes, and the fact that these new centers are being powered by fossil fuels.

00:08:51: So the AI hype train is basically running headfirst into the limits of local infrastructure.

00:08:56: And public patience.

00:08:57: The only way through this is architectural innovation.

00:09:01: We have to stop seeing data centers as just energy consumers.

00:09:06: Simeon Elvis Doomley, for example, advocated strongly for cooling innovation, like waste heat reuse and dramatic water efficiency.

00:09:14: And we are seeing new hardware architectures emerge to address this.

00:09:18: Jeff Newman shared a really interesting collaboration between Siemens and Vent.

00:09:22: I saw

00:09:22: that.

00:09:23: They're developing a joint.

00:09:24: liquid cooling and power reference architecture.

00:09:27: And this isn't just general cooling, it's purpose built for AI.

00:09:30: To maximize what they call tokens per watt.

00:09:33: Exactly.

00:09:33: It's treating energy efficiency as the primary driver of hardware design from the ground up.

00:09:39: That metric tokens per watt.

00:09:41: is so critical.

00:09:42: But I think the ultimate step forward is the circular economy model.

00:09:46: Right.

00:09:46: Carlos Clausi Marin and Hannah Erpst both highlighted successful examples where data center excess heat is being harvested for district heating in cities like Helsinki.

00:09:55: So you're turning a costly waste product into a valuable community asset?

00:10:00: You are.

00:10:00: These facilities literally become digital thermal plants that significantly reduce CO-II emissions from traditional heating.

00:10:09: And this is now being mandated in some places too.

00:10:11: It is.

00:10:12: Germany's New Energy Efficiency Act requires new buildings from twenty twenty six to utilize at least ten percent of their waste heat.

00:10:20: It's a brilliant shift from waste to regulatory compliance and community value.

00:10:24: This brings us perfectly to our third and final theme.

00:10:28: Standards, metrics, and strategic advantage.

00:10:31: Because if all these efforts from green coding to liquid cooling are going to scale, We need a standardized way to measure them.

00:10:38: Absolutely.

00:10:39: And the cornerstone here is the Software Carbon Intensity Specification, or SCI.

00:10:44: Wakazamad helped break it down.

00:10:46: The formula is SCI equals operational emissions plus embodied emissions, all divided by the functional unit.

00:10:51: OK, that's a mouthful of acronyms.

00:10:53: Let's make that accessible.

00:10:54: What does that actually mean for the average developer?

00:10:56: It's about moving from measuring total carbon to measuring the rate of carbon.

00:10:59: All right,

00:11:00: OK.

00:11:00: OA is the energy used when your software is running.

00:11:03: M. is the carbon it took to make the hardware.

00:11:06: And R, the functional unit, is the work it does, like API calls or transactions.

00:11:10: So it forces you to measure the carbon footprint per unit of work.

00:11:14: Exactly.

00:11:15: Which is the only way to accurately compare different applications.

00:11:18: And the speed that this is being adopted is pretty remarkable.

00:11:22: It really is.

00:11:23: Paula Rissi, Naveen Balani, and Jacques Kale USK all confirmed a major milestone recently.

00:11:30: The Green Software Foundation ratified the SEI for AI specification.

00:11:34: So it officially extends that standard to cover the entire AI lifecycle.

00:11:38: The whole thing, from the massive training costs we talked about to the ongoing inference costs.

00:11:44: And Asim Hussein confirmed the recommended metric for consumers is now carbon per token.

00:11:49: That's huge, because it gives practitioners a comparable engineering focus metric for something that was, you know, totally fragmented before.

00:11:57: And we're already seeing this mindset in action through carbon-aware computing.

00:12:01: Aiden, Mir Mohamedi discussed the time-shifting approach used by companies like ScriptRunner and BlueHands.

00:12:07: The idea is elegantly simple.

00:12:09: You take jobs that aren't latency, critical big-batch process as overnight analytics, and you shift them to run only when the local grid's carbon intensity is lower.

00:12:19: And that can make a huge difference.

00:12:21: Yeah.

00:12:21: Mohamedi noted that in a place like Germany, the grid's carbon intensity can fluctuate wildly from like fifty to seven hundred grams of CO two per kilowatt hour.

00:12:31: So if you can shift your biggest computational load from a high carbon moment to a low carbon one, the reduction is immediate and significant.

00:12:39: Time shifting sounds smart.

00:12:40: a classic optimization play.

00:12:42: But it raises a critical systemic question that no one Goddard brought to the table.

00:12:47: And we really need to address this.

00:12:48: Her counter argument is vital.

00:12:51: She cautioned that on today's fossil-backed grids, shifting your load might only displace the emissions to another time or to another user.

00:12:58: So it doesn't reduce the total energy needed by the system?

00:13:00: Exactly.

00:13:01: And worse, it can risk grit stability by triggering fossil fuel backup generators to handle those peak shifts.

00:13:07: She argues the real challenge is an optimization.

00:13:10: It's constraining absolute energy demand.

00:13:12: So she's arguing for a shift to grid-aware computing that forces companies to invest in additive clean power, not just rearrange the chairs on the Titanic.

00:13:22: It's a necessary challenge to the optimization mindset that has dominated IT for so long.

00:13:27: And that systemic challenge is playing out fiercely in the policy landscape, which Antonio Vizcaia-Abdo described as, well, mixed and confusing.

00:13:35: That's

00:13:35: a good way to put it.

00:13:36: We saw Nathaniel Barola criticize a high-level UNAI resolution basically saying, it sounds good, but it lacks any real detail on energy use, life cycle analysis, or enforcement.

00:13:48: So global policy is lagging way behind the technical reality.

00:13:51: Way behind.

00:13:52: And on the flip side, domestic policy could actually be moving backward.

00:13:56: Samuel Rinse shared some concerning news that the European Commission is proposing easing environmental obligations on large data center projects.

00:14:03: Easing them?

00:14:03: Yeah.

00:14:04: Things like limited public consultation times and maybe even waiving environmental impact assessments for expansions.

00:14:10: The tension between needing infrastructure capacity now and our climate goals is just extremely acute.

00:14:16: Okay, so we've covered the software, the hardware, the standards, the policy challenges.

00:14:21: Where does this leave a business leader?

00:14:23: Is green IT still just about reputation and PR?

00:14:26: The resounding answer from the sources is no.

00:14:29: Wilco Bergraf and Arwal Owen both stated that by twenty twenty six, green IT stops being just a moral obligation.

00:14:36: Right.

00:14:37: It transforms into a demonstrable operational advantage.

00:14:40: It gives you control over resources, complexity and inherent risk.

00:14:44: By identifying and reducing that virtual waste inefficient code, inefficient architecture, you're actually improving resilience and security.

00:14:52: Efficiency is, finally, strategically profitable.

00:14:55: It's moving far beyond just compliance.

00:14:58: So to summarize this deep dive, the focus for green ICT has fundamentally shifted away from those abstract goals.

00:15:04: Sustainability is now a core engineering metric.

00:15:07: It demands precise measurement, architectural redesign and accountability across every single layer of the stack.

00:15:12: And the tension between AI's explosive demand and our climate goals is absolutely the defining strategic challenge of twenty

00:15:19: twenty six.

00:15:19: And given that tension, here is a provocative thought for you to take away and discuss with your team.

00:15:25: We've established the difference between optimization and constraint.

00:15:28: So considering the rapid growth of AI and data centers, how will you distinguish, within your own organization, between carbon aware efficiency, which just... optimizes existing usage and solutions that genuinely constrain absolute energy demand, ensuring you're not just displacing the problem.

00:15:45: That is a crucial distinction that will define winners and losers in the next few years.

00:15:49: If you enjoyed this deep dive, new episodes drop every two weeks.

00:15:53: Also check out our other editions on cloud, digital products and services, artificial intelligence and ICT and tech insights, health tech, defense tech.

00:16:00: Thank you for listening and don't forget to subscribe.