Best of LinkedIn: Green ICT & Sustainable AI CW 38/ 39

00:00:00: This episode is provided by Thomas Allgaier and Frenties, based on the most relevant linked in-posts about green ICT and sustainable AI from CW-Thirty-Eight & Thirty-Nine.

00:00:11: Frentie supports ICT enterprises in the form of delivering precise ICT market and pricing intelligence that analyst subscriptions an existing databases cannot provide.

00:00:21: You can find more info in description.

00:00:23: So, uh did you know that processing just a handful of really complex AI prompts could You know consume the exact same amount of energy as boiling a kettle of water?

00:00:33: Wow.

00:00:34: Just a kettle-of-water for a few prompts?

00:00:36: Yeah But well here is the catch.

00:00:39: That's actually only true if your request gets routed to a server sitting in a very specific Geographic region.

00:00:45: right

00:00:45: If it gets routeds somewhere else like a different grid entirely that carbon footprint might drop by.

00:00:51: I mean a factor of twelve.

00:00:52: That is just wild, a factor of twelve.

00:00:55: Well welcome to today's deep dive everyone!

00:00:57: We are looking at the physical collision between our digital ambitions and quite frankly...our ecological reality.

00:01:04: Exactly because

00:01:05: it's huge bottleneck right now

00:01:07: It really is

00:01:08: So for you, The Digital Transformation & Tech Professionals navigating all this we're pulling together the absolute most critical insights on green ICT and sustainable AI trends that are circulating across LinkedIn Right Now.

00:01:21: Yeah, we're trying to figure out how the industry is actually dealing with this because you know The cloud has this illusion of being

00:01:27: completely invisible

00:01:27: right?

00:01:28: Invisible.

00:01:28: You write some code you hit enter and it just feels totally weightless.

00:01:32: But I think you all know that behind every single API call there was just this massive heavily constrained physical infrastructure.

00:01:40: yeah We are rapidly moving away from this world where software was treated as just you know abstract logic.

00:01:45: Now it's firmly tethered to the ground.

00:01:47: It is pulling just unbelievable amounts of water and power from local communities

00:01:52: And, um...the entire industry is waking up this reality at basically exact same time.

00:01:57: The limits aren't about processor speed anymore

00:02:01: Right!

00:02:01: Its not just memory limits.

00:02:02: No its literally gigawatts of electricity Millions gallons of water.

00:02:07: So let us start by looking how the industry measuring footprint right now Because logically, you can't optimize a system if your baseline metrics are lying to right?

00:02:16: Right exactly.

00:02:17: If you measure the wrong thing You optimize the wrong things

00:02:22: Exactly and we're seeing this major shift away from isolated endpoint metrics.

00:02:27: Pavel Kamiaku brought up a really fantastic point about this.

00:02:31: Oh yeah his post was great.

00:02:33: Yeah He argues that software efficiency needs to be measured per unit of business work not just by looking at A single endpoint in a complete vacuum

00:02:42: because that can be super misleading.

00:02:44: Totally, like he gives this example of a .NET API endpoint let's say a developer makes that end point I know thirty percent faster by trading CPU processing time for some extra database queries.

00:02:57: right.

00:02:57: so on paper that benchmark looks absolutely incredible!

00:03:01: Right the developer probably gets a bonus for making things faster.

00:03:05: But what's actually happening to the backend infrastructure?

00:03:08: Well,

00:03:09: The cloud bill goes up for one and so does the carbon footprint.

00:03:12: Exactly

00:03:13: Because by shifting that computational burden away from the web server And just dumping it onto the database cluster...the database load totally spikes.

00:03:21: So this system as a whole is working harder.

00:03:24: exactly It's consuming way more total resources To complete exact same business transaction.

00:03:30: Its like if you make a conveyor belt thirty percent faster Just by pushing the load to the packaging department?

00:03:36: The factory itself isn't actually more efficient.

00:03:39: Right, you're just moving the bottleneck around and basically playing a shell game with your emissions.

00:03:43: Efficient code does not always equal an efficient system...the system boundary really matters here

00:03:49: Which kind of brings us directly into end user side of this equation.

00:03:53: Allison Freeman shared his interesting development.

00:03:56: Oh, the calculator.

00:03:57: Yeah there is now a free AI impact calculator out there that estimates energy use for your daily prompts.

00:04:04: That's pretty cool.

00:04:05: It breaks down energy, carbon emissions and water consumption based on whether you're just drafting text emails or generating those heavy high resolution images.

00:04:14: it's brilliant awareness tool For sure.

00:04:17: But she makes this really crucial point.

00:04:19: We can't blame end users.

00:04:21: prompting habits.

00:04:23: Absolutely not

00:04:24: Right, like we can't solve a massive infrastructure crisis by just telling everyday people to you know prompt less.

00:04:30: No because putting the burden of sustainability on every day users I mean it's just a distraction from the actual structural problems.

00:04:37: Exactly!

00:04:38: It is systemic issue.

00:04:39: Yeah, technology leaders have to architect solutions.

00:04:42: They need to engineer grid-aware infrastructure from the ground up like The system should be smart enough to route say non urgent batch processing two regions where renewable energy is currently peaking

00:04:55: instead of relying on the user To decide if their prompt is like environmentally justified today

00:05:00: exactly and that structural shift Is already hitting corporate budgets really hard?

00:05:06: Oh Definitely.

00:05:07: Micah Conrad's actually gave a really good look inside her company Newland regarding this.

00:05:12: Yeah,

00:05:12: they rely heavily on agentic coding right?

00:05:14: Right and They actually built their own token dashboard to track there AI usage across all these different cloud providers.

00:05:21: And what did they find?

00:05:22: well they discovered something absolutely terrifying for any CFO.

00:05:27: They realized that moving entirely to a token-based LLM billing model could actually raise their costs by a factor of ten.

00:05:34: Ten, wow.

00:05:35: Yeah

00:05:35: a factor of ten.

00:05:36: compared to traditional cloud hosting cost and carbon footprint are becoming well deeply intertwined.

00:05:43: Let's actually break down why that happens because it's fascinating.

00:05:46: you know Traditional Cloud pricing is largely based on reserved instances or just server uptime.

00:05:51: You're paying for the machine did?

00:05:53: Just be there

00:05:54: exactly.

00:05:54: you pay for availability.

00:05:56: But generative AI brings in this token-based billing.

00:05:59: you are paying per inference literally per word generated.

00:06:04: So compute intensive bursts become incredibly expensive.

00:06:08: Right, and Newland actually expanded their dashboard.

00:06:10: they worked with a green IT expert to estimate the CO-II footprint right alongside the financial cost

00:06:16: Which makes sense because every token processed requires A specific fraction of compute which draws a fraction Of energy which emits a fraction of carbon.

00:06:25: Precisely so when you reduce unnecessary token consumption just to save your budget You're simultaneously pulling down your estimated emissions.

00:06:33: The financial incentive and the ecological incentive are just perfectly aligned for once.

00:06:38: Yeah,

00:06:39: they finally line up.

00:06:40: But you know I want to throw a wrench into that logic for a second.

00:06:43: uh-oh okay well

00:06:45: You might assume that to save tokens in energy?

00:06:47: You should just automatically use a smaller leaner AI model right?

00:06:51: yeah That's the standard thinking right now Right.

00:06:54: but Anita sure points out this really critical nuance that i think most developers Are completely missing.

00:06:59: Smaller ai models are not Automatically greener.

00:07:03: See that sounds completely counterintuitive to me, right?

00:07:06: I mean a smaller parameter model requires less memory.

00:07:10: It takes less processing power.

00:07:12: how on earth could it possibly burn more energy?

00:07:15: think of it kind like Hiring an interned to save money.

00:07:18: okay

00:07:19: the interns hourly rate is way lower than a senior engineers rate.

00:07:22: Right

00:07:22: sure obviously.

00:07:23: but

00:07:24: if that intern struggles with a complex task and Takes ten hours to do it loops through five different revisions and needs constant oversight.

00:07:33: Ah,

00:07:33: you haven't actually saved any money?

00:07:35: Exactly!

00:07:36: You've spent way more... And it's the exactly mechanism with AI.

00:07:40: If a smaller model struggles with complex reasoning tasks It requires way more retries.

00:07:45: Or if takes heavier like reasoning effort to get equality results

00:07:49: Yes..if have to force into loop until he gets right it actually burns massive amounts of energy.

00:07:56: So basically just spends time spending its wheels.

00:07:59: A larger, more capable model might consume more power per second.

00:08:03: Sure but it does the job right on their very first try in a fraction of time.

00:08:08: That makes total sense.

00:08:09: And plus if developers are constantly switching models back and forth to optimize things they have to continually refill the cash which is just another huge energy drain.

00:08:19: Right!

00:08:19: The Total Energy To Reach The Destination Is What Actually Matters Not Just The Size Of The Engine.

00:08:24: Exactly

00:08:25: And honestly, that is a massive blind spot in how organizations are forecasting their AI emissions right now.

00:08:31: It really is!

00:08:32: Those blind spots get even more severe when we start looking at geography.

00:08:36: Lilabeth Bustos-Lanaris actually shared some cool findings from SOMA AI.

00:08:40: Oh

00:08:41: they're a carbon footprint tool?

00:08:42: Yeah exactly They found.

00:08:44: just misclassifying an AI model's tier can throw your emission estimates off by four to five times

00:08:50: Four to Five Times Just From Tier Misclassification.

00:08:53: But the real shocker is the geography part.

00:08:56: Running the exact same model with the exactsame query in a different geographic region can swing your emissions by a factor of twelve.

00:09:03: A twelve-time difference just based on where server physically sets?

00:09:07: I mean, how's that even

00:09:08: possible?!

00:09:09: Is it really just the difference in local power grids?

00:09:12: It's all about the base load power of that specific local grid.

00:09:15: Okay, like if you route a workload to a data center in a region That heavily relies on coal or natural gas The carbon intensity per kilowatt hour is just massive

00:09:24: right?

00:09:25: They make sense

00:09:25: but If you route that exact same work load To a facility and say Quebec Or the Pacific Northwest where they run almost entirely On hydroelectric power

00:09:34: the carbon footprint completely plummets.

00:09:36: Exactly,

00:09:37: and this is exactly why having sourced versioned and tier data is so incredibly critical for enterprise buyers.

00:09:43: You can't just use a generic global average.

00:09:46: No using a global average multiplier for carbon accounting Just renders the data completely useless.

00:09:51: that

00:09:51: Is fascinating?

00:09:53: Before we zoom out from the software layer in really dive into the physical infrastructure side of these data centers Just a quick reminder for y'all listening.

00:10:01: Yeah, subscribe exactly.

00:10:03: hit that subscribe button if you want to make sure You don't miss future deep dives into these curated industry insights.

00:10:08: We curate all these trends so you don't have to spend hours scrolling linked in yourself.

00:10:12: Well we do the heavy lifting for you.

00:10:14: So let's talk about the physical site now because If The software measurement is zooming out from the endpoint To the whole business system?

00:10:21: The hardware layer Is doing exact same thing.

00:10:23: yeah

00:10:23: data centers are definitely no longer just isolated buildings

00:10:27: right.

00:10:28: Paul Chernoch framed this brilliantly.

00:10:30: He argues that the physical facility itself is no longer the right boundary for optimization.

00:10:35: It's way bigger than the building

00:10:36: now.

00:10:36: Exactly AI infrastructure must be optimized as this deeply coupled system of power, compute and economics

00:10:44: Which is a huge shift.

00:10:46: Yeah

00:10:46: Because for twenty years The industry basically just optimized the building

00:10:50: Right?

00:10:50: They minimized electrical losses they bought slightly more efficient air conditioners

00:10:55: Right, and he treated the actual useful work produced by servers as just somebody else's problem.

00:11:00: Totally!

00:11:01: And AI absolutely shatters that separation.

00:11:04: Chernot calls this new paradigm fuel-to-token efficiency.

00:11:08: Fuel to token?

00:11:09: Yeah I like it.

00:11:10: It is no longer about how efficiently power reaches server rack but effectively the entire ecosystem converts raw energy into economically useful output.

00:11:20: Factoring in things like grid constraints

00:11:22: right?!

00:11:23: Yes, grid constraints water availability even community acceptance.

00:11:28: But you know here is the paradox with just optimizing a facility You can have the absolute most efficient power delivery in the world but if the servers inside are running garbage code... ...you're just efficiently wasting energy.

00:11:42: Exactly!

00:11:43: Your'e doing really good job at wasting power Right.

00:11:47: And that's why industry veterans like Greg Bodenheimer and Vinod Balani Are arguing PUE.

00:11:53: Oh, power usage effectiveness?

00:11:55: Yeah.

00:11:56: Power usage effectiveness which has been the golden metric for data centers for twenty years They're saying it's no longer enough.

00:12:03: The industry really needs to transition toward metrics like watts per token

00:12:08: Or a useful AI output per megawatt.

00:12:10: I think they called it

00:12:11: Exactly!

00:12:12: Have to play devil's advocate here.

00:12:13: for second though Go

00:12:14: for it

00:12:15: Because we can't throw PUE out window just yet.

00:12:18: Like Watts Per Token is definitely Holy Grail but base facility efficiency still moves massive markets.

00:12:25: Oh, for sure!

00:12:26: Vinod Bajlani actually ran the math on this to show why PUE still matters at hyperscale?

00:12:31: What did The Math Show?

00:12:32: Well if you take a one gigawatt AI factory which is the crazy scale we're moving toward right now and you cut its PUE from the industry average Which is about one point five four okay And You get it down To A highly Efficient One Point O nine The physical Savings Are Just Staggering.

00:12:48: Okay, wait.

00:12:48: A gigawatt is roughly the power draw of like a medium-sized city.

00:12:52: So what does it drop?

00:12:53: Of zero point four five and pue actually yield.

00:12:57: in reality

00:12:58: you save about three thousand nine hundred forty gigawatts hours of energy every single year.

00:13:02: Wow

00:13:03: Yeah And in financial terms that translates to roughly three hundred fifteen million dollars in lower annual energy costs.

00:13:10: Three hundred and fifteen million Dollars.

00:13:13: yep

00:13:14: And that is before you even generate a single AI token or run a single line of code.

00:13:19: Facility efficiency is absolutely still the bedrock of economics.

00:13:23: Okay, yeah Saving three hundred and fifteen million dollars Is pretty compelling reason to keep tracking PUE.

00:13:29: We'll give ya That

00:13:30: Just A little bit.

00:13:31: But driving that number down actually brings us To probably The most contentious physically demanding part Of data center design right now.

00:13:39: Which is cooling?

00:13:40: Oh Yeah The heat is a massive problem.

00:13:43: Anil Velecha and Ravi Shankar Bhashbhai weighed in on this, they noted that cooling is just a brutal trade-off.

00:13:49: there's literally no single best cooling method out there.

00:13:52: Right

00:13:52: it's the tight rope walk between water use and energy use.

00:13:55: Exactly!

00:13:56: It's entirely dependent upon local Water Energy Nexus.

00:13:59: Like if you want to use less energy to cool your facility You often have evaporate millions of gallons through these massive cooling towers

00:14:07: Which local communities hate?

00:14:09: But if you want to save water by using closed-loop mechanical air cooling, your energy consumption spikes dramatically.

00:14:16: Especially in really hot climates?

00:14:18: Yes because you have run these massive compressors!

00:14:21: There are some breakthroughs bypassing this entirely

00:14:23: right?!

00:14:23: And

00:14:24: they'll actually share the wild success story from a company called Viridian.

00:14:28: Oh The Immersion Cooling one

00:14:29: Exactly.

00:14:30: They took this legacy air cooled data hall and entirely retrofitted it with immersion cooling.

00:14:36: literally dunk the servers into these specialized non-conductive liquid tanks.

00:14:40: Which sounds crazy, but let's explain why that actually works because liquid conducts heat significantly better than air does

00:14:47: way better.

00:14:47: so when you submerge The hardware You completely eliminate the need for those tiny high rpm mechanical fans inside the servers themselves.

00:14:54: and Those fans alone can consume what?

00:14:57: Ten to fifteen percent of the racks total power

00:14:59: exactly.

00:15:01: Plus you eliminate them massive air handlers that are just pushing cold air across the room.

00:15:06: And, the transformation for Viridian was staggering!

00:15:09: They went from over forty-six hundred legacy servers down to just three hundred and fifty seven liquid cooled ones.

00:15:15: Wow

00:15:16: that's a huge footprint reduction.

00:15:18: It is.

00:15:19: they swapped traditional CPUs for GPUs and managed get four times of total compute power.

00:15:25: Four Times The Compute.

00:15:27: Yes

00:15:27: but...the kicker Is the energy draw..they actually cut their total power used eight hundred and forty-eight kilowatts down to six hundred and fourty five kilowattes.

00:15:37: So wait, more than quadruple the compute in exact same footprint?

00:15:42: And they used less energy overall?

00:15:44: Yes

00:15:45: that is exactly what happens when you rethink infrastructure from physics up rather then bolting bigger air conditioners onto legacy designs.

00:15:53: That's incredible!

00:15:55: It isn't just cooling physics changing right now The entire geographical strategy of industry shifting.

00:16:01: How so

00:16:01: Well, Meg's and Cheng highlighted this fascinating pivot happening with Chinese renewable energy giants.

00:16:07: Oh!

00:16:07: With the solar farms?

00:16:08: Yeah they are currently dealing with this massive solar-and-wind overcapacity in really remote regions.

00:16:14: so their generating all of these green power but face severe grid curtailment

00:16:18: Meaning transmission lines were just too congested to actually transport that electricity back into big cities.

00:16:24: right.

00:16:25: Exactly So the power is essentially stranded.

00:16:27: it was wasted.

00:16:29: What s there solution?

00:16:30: do build more lines?

00:16:31: No, instead of waiting a decade to build new high voltage transmission lines.

00:16:36: They are literally building AI data centers directly next to the massive desert solar and wind farms.

00:16:41: Wow!

00:16:42: Yeah it's a strategy that is known as East Data West Compute.

00:16:45: It basically the modern digital equivalent of Building The Steel Mill right Next To Iron Mine

00:16:50: Exactly its such brilliant business mechanism.

00:16:53: they completely bypassed grid bottlenecks.

00:16:55: Right

00:16:55: because they don't need to transport power anymore

00:16:58: exactly.

00:16:58: They're pivoting from being pure power generators to becoming these fully integrated compute and energy providers because selling green electricity in the form of cloud compute yields vastly higher margins than trying to sell really cheap, stranded electricity back to a congested grid.

00:17:13: That makes total sense!

00:17:15: At this level of integration isn't just about power grids either – it's also extending into local ecology….

00:17:21: Oh like bio-mimicry stuff?

00:17:23: Yeah Joris Shonis and Ard Bleeker shared updates on Microsoft's biomimicry pilot in Middenmere over the Netherlands.

00:17:30: Right, where they're designing data center campuses to actually function in harmony with native habitats Which

00:17:36: sounds like PR but it is real!

00:17:39: It means stepping away from that traditional model of just paving a massive plot of land into this sterile concrete fortress.

00:17:47: So what does it look like?

00:17:49: They are planting native trees, they're creating pollinator corridors and their actively restoring wetlands right around the facility.

00:17:56: That's amazing!

00:17:58: But it isn't just for aesthetics...right?

00:17:59: Not at all.

00:18:00: Restoring those wetlands provides natural groundwater retention which actually actively lowers the ambient microclimate temperature around the entire facility.

00:18:10: Oh wow so cooler ambient air means less mechanical cooling is required for servers inside.

00:18:15: Exactly its this incredible feedback loop.

00:18:18: It's been so successful from a pure operational standpoint that Microsoft is now expanding this biomimicry approach to over twenty sites across the US and Germany.

00:18:28: It really represents this profound shift in engineering mindset.

00:18:32: Infrastructure is finally blending with ecology!

00:18:35: Yes, but let us be realistic here – not every operator is taking this pro-active beautifully integrated approach….

00:18:43: Oh definitely

00:18:44: And with gigawatts of power and massive water resources at stake, regulators in corporate buyers are completely losing patience.

00:18:52: Yeah we're working on it is just no longer an acceptable answer for these regulators.

00:18:57: No transparency is shifting from this nice to have corporate PR talking point To a really strict operational requirement

00:19:05: And the policy landscape is hardening much, much faster than the tech sector anticipated.

00:19:10: Like Matthias Heimoth pointed out that Scotland has effectively drawn a line in the sand regarding data center development.

00:19:16: What did they do?

00:19:17: Their parliament officially mandated full environmental impact assessments for any data center over fifty megawatts and That has to be done before planning permission can even be considered.

00:19:28: Wow I mean a fifty-megawatt facility.

00:19:30: it's pretty substantial.

00:19:32: Why are they putting up so much friction upfront?

00:19:34: Because a facility that size places a massive, sudden strain on the local grid queue and the local water rights.

00:19:41: Regulators need to see the forecasted impact up front... ...to ensure grid stability for actual residents.

00:19:47: Proof of sustainability is now literally the strict price of admission to build!

00:19:52: And thats happening across Europe too right?

00:19:54: Christoph Kluzzo and Michael Kaye noted that EU Commission has officially adopted A-G ratings for data centers energy & water use.

00:20:03: Oh right, just like the energy efficiency labels you see on household appliances.

00:20:07: Exactly

00:20:07: Like that!

00:20:08: And very first public labels are expected by twenty-twenty seven.

00:20:12: Think about operational risk for a facility operator though You will not be able to hide an inefficient water guzzling facility anymore

00:20:20: Because data is totally public.

00:20:22: Right

00:20:23: But I do have ask Will G rated datacenter actually lose customers?

00:20:28: I mean, with AI compute demand being as high it is right now.

00:20:32: Will buyers even care or will they just take whatever service space?

00:20:35: They can possibly get?

00:20:37: no they will definitely care mostly because of scope.

00:20:40: three emissions compliance

00:20:41: ah Right

00:20:42: and that actually makes this a next example from Mark Butcher highly relevant.

00:20:46: he reviewed AWS's recent submission to the Australian Senate regarding their data centers down there.

00:20:51: okay what was in?

00:20:53: Well, AWS has been very vocal about investing over twenty billion dollars into Australian facilities right?

00:20:58: Yeah that was a huge announcement.

00:21:00: But in their thirty-four page submission to the Senate they completely omitted there absolute energy and water use.

00:21:07: Wait...they left out raw consumption numbers entirely?

00:21:10: Entirely They provided regional averages for PUE & WUE which is Water Usage Effectiveness.

00:21:17: Right The efficiency ratios.

00:21:19: But an efficiency ratio only tells a regulator how efficiently a resource is utilized inside the building.

00:21:26: It tells them absolutely nothing about total volume being drained from local grid or water shed,

00:21:31: which raises a huge red flag.

00:21:33: I mean are tech giants weaponizing ratios like PUE to hide their massive absolute consumption of regulators and communities?

00:21:42: it definitely looks that way for people Because when regulators are trying to manage grid capacity or, you know secure drinking water during a drought.

00:21:49: A ratio is practically useless.

00:21:52: They need the peak megawatt demand

00:21:54: Exactly and millions of gallons of water required.

00:21:57: The lack absolute disclosure creating really severe friction

00:22:02: And this lack transparency hitting software buyers just as hard too.

00:22:06: Richard Tarboton shared that right now AI products are bought, sold and deployed with almost zero environmental data attached to the specific product layer.

00:22:15: Which is crazy!

00:22:16: Imagine being a corporate procurement officer today.

00:22:19: you're choosing between two enterprise AI products that perform the exact same business function but one amidst say twenty grams of CO₂ per million tokens, and the poorly optimized one emits a thousand grams.

00:22:31: That's

00:22:31: massive difference!

00:22:33: Huge!

00:22:34: And if you're a buyer trying to meet strict corporate climate targets that difference alters your entire procurement decision.

00:22:40: But because suppliers do not actually provide this information buyers are flying completely blind Right... ...and that is exactly why major corporate buyers now demanding product level carbon disclosure.

00:22:51: They want the specific traceable footprint of the AI product before they sign.

00:23:18: So it's transitioning from the marketing department directly into operations?

00:23:22: Yes.

00:23:23: It is being embedded directly in to core operating model.

00:23:26: Executives are relying heavily on technology to close gap between their climate ambitions and reality

00:23:32: But a CIO cannot operationalize what they can not actually trace Exactly,

00:23:37: if AI tools that procure remain black box of emissions They just cant balance carbon ledgers.

00:23:44: We have covered incredible ground today.

00:23:47: I mean, we started at the micro level of software metrics and token billing.

00:23:51: We moved through the physical realities of emerging cooling in grid stranding And ended up with macroeconomic regulatory shifts.

00:23:58: It really proves how deeply interconnected this entire ecosystem is.

00:24:02: it Really does.

00:24:03: every single prompt you type sets off this physical chain reaction with thermal electrical and ecological consequences.

00:24:11: The boundary between the digital and the physical just basically no longer exists.

00:24:16: he really doesn't.

00:24:17: But I want to leave you with one final provocative thought to mull over, which was brought forward by MD Satakwad Hussein.

00:24:25: Oh this is such a good post!

00:24:27: Right?

00:24:27: We spent so much time analyzing how cool these gigawatt data centers and construct massive nuclear powered AI facilities.

00:24:36: but consider the biology.

00:24:37: we already have

00:24:39: Our own brains.

00:24:40: Exactly The human brain contains roughly eighty six billion neurons.

00:24:44: It remains the absolute most capable general intelligence we know of, adapting to complex unstructured environments instantly.

00:24:52: And it operates on about twelve watts of power.

00:24:54: Twelve Watts?

00:24:55: That is energy required for a single small LED light bulb.

00:25:00: A lightbulb!

00:25:01: Meanwhile... ...a single modern AI training system consumes one hundred megawatts which has enough to power an entire town.

00:25:09: Perhaps the ultimate future of AI isn't about building exponentially larger power-hungry models.

00:25:15: Maybe the true frontier of AI lies in discovering how biological intelligence achieves so much with so incredibly little?

00:25:23: That is something to think about!

00:25:25: Definitely, if you enjoy this episode new episodes drop every two weeks.

00:25:29: also check out our other editions on cloud insights and sovereignty digital products and services AI energetic systems health tech ICT and Tech Insights defense tech, and health tech.