Latest / Best of LinkedIn: Digital Transformation & Tech / Best of LinkedIn: Sustainability & Green ICT CW 28/ 29

Transcript

Best of LinkedIn: Sustainability & Green ICT CW 28/ 29 · Best of LinkedIn: Digital Transformation & Tech

• 0:00
  Speaker 0: This episode is provided by Thomas Allgeier and Frennis, based on the most relevant LinkedIn posts about sustainability in green ICT from CW-Twenty-Eight and Twenty-Nine.
• 0:10
  Speaker 0: Frennes supports ICT enterprises in the form of delivering precise ICT market and pricing intelligence that analyst subscriptions and existing databases cannot provide.
• 0:19
  Speaker 0: you can find more info in the description.
• 0:21
  Speaker 0: imagine uh...you're standing next to a pristine crystal clear lick.
• 0:26
  Speaker 0: Now watch as someone casually walks up scoops out a liter of fresh water and just dumps it directly onto the dry ground, right?
• 0:31
  Speaker 0: And they do it.
• 0:32
  Speaker 0: Just to say thank you to a robot.
• 0:34
  Speaker 1: Welcome to our deep dive today.
• 0:37
  Speaker 1: Our mission is to cut through the corporate fluff and explore the reality of sustainability in green ICT.
• 0:42
  Speaker 1: We are basically synthesizing a massive wave of intelligence across LinkedIn Curated specifically for digital transformation professionals like you which brings us to this central paradox of our time.
• 0:53
  Speaker 1: I mean artificial intelligence is making software infinitely more capable, which is amazing.
• 0:58
  Speaker 1: It's unlocking unprecedented value but beneath that digital magic the physical and environmental footprint of that Intelligence is spiraling completely out-of control.
• 1:09
  Speaker 1: We are trying to build the infrastructure of the future But the real question is how do we do that without burning down?
• 1:15
  Speaker 1: The resources at present?
• 1:16
  Speaker 0: Yeah, let's unpack this by starting at the very foundation which is the software layer.
• 1:20
  Speaker 0: because before we even look at massive power grids or you know concrete data centers.
• 1:25
  Speaker 0: The most immediate lever that you as a tech professional actually have Is the code itself?
• 1:31
  Speaker 0: Absolutely and the data We are seeing here is staggering.
• 1:34
  Speaker 0: Let's start with that lake analogy And talk about the physical cost of simply being polite to a chatbot.
• 1:39
  Speaker 1: yeah.
• 1:39
  Speaker 1: So Yassin Kizirga posted This really fascinating breakdown Of this exact phenomenon.
• 1:44
  Speaker 1: The baseline metric is that a simple prompt to a model like Gemini uses roughly point two four watt hours of energy and Point two six milliliters water,
• 1:52
  Speaker 0: which I mean That sounds tiny in isolation.
• 1:55
  Speaker 1: It does.
• 1:56
  Speaker 1: but large language models do not have a memory.
• 1:58
  Speaker 1: In the human sense.
• 2:00
  Speaker 1: you know They're completely stateless.
• 2:02
  Speaker 1: Every single time you hit enter, the model has to computationally re-read the entire context window!
• 2:07
  Speaker 1: Oh wow... So if your deep into a hundred thousand token conversation and add a polite thankyou at the very end You are forcing the models attention mechanism To process that entire massive history from scratch.
• 2:20
  Speaker 1: Just generate an welcome.
• 2:22
  Speaker 0: That is wild.
• 2:24
  Speaker 1: Casierga calculates that this simple courtesy costs about two point four watt hours and two point six milliliters of water.
• 2:30
  Speaker 1: It's like
• 2:30
  Speaker 0: forcing an athlete to run a marathon from the starting line.
• 2:33
  Speaker 0: every single time you say bless your after a sneeze exactly at A global scale.
• 2:37
  Speaker 0: considering billions of queries open AI in Google are spending tens Of millions of dollars, and literally evaporating millions gallons of water just to process users politeness.
• 2:48
  Speaker 0: But wait, let me push back on this a bit.
• 2:49
  Speaker 0: Sure if polite prompts cost that much computational power shouldn't the financial costs of every single token naturally disincentivize?
• 2:57
  Speaker 0: This like why isn't the raw price tag forcing developers into greener coding practices automatically?
• 3:03
  Speaker 1: I hear that logic but The Financial Cost completely obscures the physical reality.
• 3:09
  Speaker 1: Mark Bradley addresses this with a concept he calls sustainable tokenomics.
• 3:14
  Speaker 0: Okay, what does that mean?
• 3:15
  Speaker 1: Well He points out the cloud providers artificially smoothed out their pricing to maintain market share.
• 3:21
  Speaker 1: There's a massive hidden carbon shadow beneath every AI prompts price tag that you aren't paying for in dollars, but society is paying for it in resources.
• 3:30
  Speaker 1: Right!
• 3:31
  Speaker 1: And we have to tread carefully here too.
• 3:33
  Speaker 1: Wilco Bergraff highlighted in his research that green coding and sustainable coding are not interchangeable terms...
• 3:39
  Speaker 0: Oh so?
• 3:40
  Speaker 1: Well an optimization that reduces one type of impact say tweaking an algorithm to lower local CPU energy use might quietly require drastically more memory bandwidth, which shifts the environmental impact.
• 3:52
  Speaker 1: To the manufacturing side of the supply chain.
• 3:55
  Speaker 1: you really need a holistic view not just a financial optimization.
• 3:58
  Speaker 0: make that makes sense for a developer though how do you actually measure?
• 4:01
  Speaker 0: A holistic view because traditionally measuring software energy consumption is incredibly noise.
• 4:06
  Speaker 0: Oh
• 4:06
  Speaker 1: absolutely
• 4:07
  Speaker 0: right like if I run a script on my laptop the energy spikes depending on my background apps, or I don't know...the ambient room temperature.
• 4:15
  Speaker 1: Yeah!
• 4:15
  Speaker 1: Or if you just plugged into wall versus running out of battery?
• 4:18
  Speaker 1: It is a terrible metric for his standardized ESG report which is why Francesco Scala's recent release an open source Python library called Aflopi is such a huge breakthrough.
• 4:30
  Speaker 0: I felt floppy!
• 4:31
  Speaker 1: Yeah, that floppy completely abandons those noisy hardware metrics.
• 4:35
  Speaker 1: instead it tracks the computational workload purely at the algorithm level by counting floating point operations or FLOPs.
• 4:42
  Speaker 1: It's one hundred percent deterministic
• 4:44
  Speaker 0: meaning you get the exact same computational cost matrix.
• 4:47
  Speaker 0: whether run an algorithm on five-year old laptop Or massive state of art GPU cluster That's nicely.
• 4:53
  Speaker 0: That gives digital transformation leaders a rock solid, auditable number to actually put in their sustainability reports.
• 4:59
  Speaker 1: Yeah and building on that need for better measurement Naveen Balani just released the suite of forty-nine open source MIT licensed lean agentic AI skills.
• 5:08
  Speaker 0: Hang on wait I need stop.
• 5:09
  Speaker 0: you there Are saying we should use AI agents too?
• 5:12
  Speaker 0: To
• 5:12
  Speaker 1: find out where AI is wasting energy?
• 5:14
  Speaker 0: Yeah isn't it?
• 5:15
  Speaker 0: fighting fire with fire burning more compute
• 5:17
  Speaker 1: On surface yeah sounds totally contradictory but these are essentially lightweight design patterns Like, installable expertise for AI agents designed specifically to spot systemic waste and cut carbon across your cloud stack.
• 5:31
  Speaker 0: Oh I see!
• 5:32
  Speaker 1: The compute spent running a lean agent to audit your architecture is just fraction of the fraction it saves.
• 5:39
  Speaker 1: by identifying massive redundancies in data pipelines.
• 5:42
  Speaker 1: It basically shifts efficiency from being an afterthought To automated first class skill.
• 5:47
  Speaker 0: Okay i see ROI on that.
• 5:49
  Speaker 0: Another architectural fix that really blew my mind came from Teranjeet.
• 5:53
  Speaker 0: They highlighted a new caching engine called Infinity by GAQX.
• 5:57
  Speaker 1: Oh yeah, this is the great one.
• 5:59
  Speaker 0: The underlying logic is so simple.
• 6:00
  Speaker 0: you know Why should massive GPU cluster burn power to answer a prompt?
• 6:05
  Speaker 0: someone else has already asked?
• 6:06
  Speaker 0: Right!
• 6:07
  Speaker 0: Infinity sits in front of models like Claude and Gemini intercepts redundant queries And delivers globally cast response.
• 6:14
  Speaker 0: they are seeing an eighty-three percent reduction in expensive gpu cycles.
• 6:18
  Speaker 1: It forces us to stop treating generative AI like a search engine that needs to generate every single answer from scratch.
• 6:26
  Speaker 1: And, you know we can also optimize the architecture of the neural networks themselves.
• 6:31
  Speaker 1: Oh so?
• 6:32
  Speaker 1: Isha Hebarr ,a student researcher published some incredible findings recently.
• 6:37
  Speaker 1: She trained fifteen different convolutional neural networks or CNNs on the KineImageNet dataset and she artificially constrained her compute to test environmental efficiency.
• 6:48
  Speaker 1: She proved that lighter, plain CNN architectures consistently beat heavier more complex ResNet style architectures when measured purely on CO-II emitted per accuracy point.
• 6:59
  Speaker 0: Let's break down the mechanics of why this happens though.
• 7:01
  Speaker 0: because ResNets are famous for skip connections which allow data to bypass certain layers and intuitively it should be faster and cheaper.
• 7:10
  Speaker 1: It aids learning process but every time data takes a structural shortcut in a neural network, it requires significant memory bandwidth to hold and move that data around.
• 7:20
  Speaker 1: Moving electrons in an out of memory physically generates heat which then requires power to cool.
• 7:27
  Speaker 1: Heber's research shows that at a certain scale the environmental cost of those complex memory movements completely negates the slight bump in model accuracy you get.
• 7:36
  Speaker 0: So even if we perfectly optimize the code with tools like FL Hoppy or Infinity and we use plain CNNs, We have eventually hit a physics wall.
• 7:45
  Speaker 0: A perfectly efficient algorithm still generates heat.
• 7:48
  Speaker 0: it's still needs of physical home which forces us to look at the physical layer where the abstract cloud becomes very real heavy iron?
• 7:56
  Speaker 1: For decades The engineering behind data centers was basically just brute force.
• 8:01
  Speaker 1: you build a warehouse filled with servers And blacked industrial air conditioning.
• 8:05
  Speaker 1: Zach Ross Miller shared a perfect example of this legacy mindset from the University of Montana.
• 8:10
  Speaker 1: They had a data center operating out of a converted in nineteen seventies bathroom,
• 8:13
  Speaker 0: A bathroom like literally a bathroom
• 8:15
  Speaker 1: and because of The outdated air cooling design it literally drank eight point five million gallons Of water year through evaporation just to keep the servers From melting.
• 8:25
  Speaker 0: A bathroom drinking eight point five million gallons.
• 8:28
  Speaker 0: I mean, that is the definition of unsustainable
• 8:31
  Speaker 1: but The engineering is evolving rapidly.
• 8:33
  Speaker 1: thankfully in twenty-twenty They decommissioned that room and replaced it with a modern water free closed loop system.
• 8:40
  Speaker 0: Oh That's great.
• 8:41
  Speaker 1: Yeah It captures the hot exhaust air cools it with refrigerants In a sealed cycle And recirculates without consuming a single drop Of local Water.
• 8:52
  Speaker 0: But when you scale that up to the bleeding edge of global compute, air cooling—even closed-loop air cooling is physically reaching its limit.
• 9:01
  Speaker 1: For sure!
• 9:02
  Speaker 0: George Tritsano's highlighted Motiverr in Buffalo New York are manufacturing closed-loop liquid cooling systems for the United States massive Xascale super computers like Aurora, Frontier and El Capitan.
• 9:15
  Speaker 1: The density of those chips is so high that air physically cannot carry the heat away fast enough.
• 9:20
  Speaker 0: Exactly!
• 9:21
  Speaker 0: Traditional data center air conditioning is trying to cool down a boiling hot cup of coffee by turning up the AC in your entire house.
• 9:28
  Speaker 1: That's great way.
• 9:28
  Speaker 1: put it
• 9:29
  Speaker 0: Direct to chip.
• 9:30
  Speaker 0: liquid cooling is dropping an ice cube directly into the mug.
• 9:33
  Speaker 0: They are piping chilled liquid straight to a cold plate sitting on top of the processor.
• 9:39
  Speaker 0: It is completely different thermal paradigm that drastically putts facilities overall energy footprint.
• 9:46
  Speaker 1: We're also seeing shift where physical environment itself becoming primary design tool rather than an obstacle overcome.
• 9:54
  Speaker 1: Cleo Rocier shared a brilliant example from Switzerland.
• 9:57
  Speaker 0: Oh, the mountain one!
• 9:58
  Speaker 1: Yes they built a data center inside of former military command center sitting under fifteen hundred meters of solid mountain rock and adding housing
• 10:06
  Speaker 0: Using the Mountain as heat sink?
• 10:08
  Speaker 1: Exactly The thermal mass of fifteen-hundred meters of stone means that ambient temperature stays naturally stable year round.
• 10:15
  Speaker 1: It cuts their active cooling needs down to almost zero.
• 10:18
  Speaker 1: They aren't fighting in this environment but are literally leveraging it.
• 10:21
  Speaker 0: That is so cool.
• 10:22
  Speaker 0: And Carl Raib is taking a similar approach, but looking at the construction materials themselves he's pushing for wooden data centers?
• 10:28
  Speaker 0: Oh boy!
• 10:29
  Speaker 1: Wood.
• 10:29
  Speaker 0: Yeah by using engineered mass timber instead of traditional concrete and steel you can significantly cut construction time through prefabrication.
• 10:38
  Speaker 0: But more importantly, Concrete is one of the worst emitters of CO² on the planet.
• 10:43
  Speaker 0: Wood actually sequesters carbon Building the shell out of timber massively boosts the ESG credentials before you even plug a server in.
• 10:52
  Speaker 0: That's
• 10:52
  Speaker 1: fascinating and geography is no longer just about where land is cheap, it's hard design constraint now.
• 11:00
  Speaker 1: GiveawayPanicker points that India's data center capacity is projected to jump from one gigawatt today over two gigawatts by twenty-thirty.
• 11:08
  Speaker 1: Currently this massive portion infrastructure is concentrated on IT hubs like Mumbai or Bengaluru.
• 11:14
  Speaker 1: The problem is, those cities are heavily water-stressed and wet bulb temperatures are rising.
• 11:18
  Speaker 1: So
• 11:19
  Speaker 0: they can't handle the cooling demand?
• 11:20
  Speaker 0: Exactly!
• 11:21
  Speaker 1: Panicers strongly advocates for shifting this future infrastructure to Kerala.
• 11:25
  Speaker 1: It has some of highest rainfall in country for sustainable cooling And a much lower seismic risk profile.
• 11:30
  Speaker 1: you basically have build where resources actually.
• 11:33
  Speaker 0: By the way, if you're finding these infrastructure insights useful for your own tech strategy make sure to subscribe so you catch our future deep dives.
• 11:40
  Speaker 0: We are just scratching the surface of how this impacts enterprise architecture
• 11:45
  Speaker 1: and with all these incredible innovations in cooling and geography You would think the industry is on the right track.
• 11:50
  Speaker 1: but that brings us To The Massive Issue Of Trust And Transparency.
• 11:54
  Speaker 0: Oh boy!
• 11:55
  Speaker 1: Alexis Bateman recently announced That AWS Is Now Showing Customers Their Actual Water Withdrawal on their AWS Sustainability Console.
• 12:04
  Speaker 1: AWS states that they're data centers only use water for cooling about ten percent of the year, relying on outside air at the other ninety percent.
• 12:12
  Speaker 0: Okay I have to challenge that narrative because a dashboard is only as good as the data feeding
• 12:17
  Speaker 1: it.
• 12:17
  Speaker 1: Fair point
• 12:18
  Speaker 0: Mark Butcher highlighted a glaring discrepancy That should make every tech leader pause.
• 12:23
  Speaker 0: There's an active lawsuit from a former AWS Water Manager in Northern Virginia.
• 12:27
  Speaker 0: Oh
• 12:27
  Speaker 1: i heard about this.
• 12:28
  Speaker 0: Yeah The lawsuit alleges that AWS's public claims of a forty-two percent year over year water reduction were completely false.
• 12:36
  Speaker 0: According to local utility data cited in the court documents,
• 12:44
  Speaker 1: which is a huge difference.
• 12:45
  Speaker 1: And the discrepancy really comes down to boundary conditions, you know?
• 12:48
  Speaker 1: What a company chooses to include or exclude in itself reported metrics.
• 12:52
  Speaker 1: if You
• 12:53
  Speaker 0: are CTO listening to this and you're building your companies twenty-twenty five ESG report based on The dashboard Your cloud provider gives you might be building A house on sand.
• 13:02
  Speaker 0: absolutely.
• 13:03
  Speaker 0: If regulators allow hyperscalars To pick their own reporting methodologies How can enterprise leaders actually trust the numbers?
• 13:10
  Speaker 0: it sounds entirely like they Are grading Their Own homework holding the legal liability for those scope three emissions.
• 13:16
  Speaker 1: That tension is exactly why this stops being merely an IT problem and immediately becomes a boardroom in regulatory battlefield.
• 13:23
  Speaker 1: big tech, it's currently treating ESG compliance as an accounting puzzle rather than hard engineering problems.
• 13:29
  Speaker 1: The friction between public climate commitments And actual business operations Is reaching breaking point.
• 13:36
  Speaker 0: The raw numbers prove that James Martin shared a staggering statistic Microsoft, Google and Amazon's combined emissions are up forty two percent since twenty-twenty.
• 13:45
  Speaker 1: Forty Two Percent
• 13:46
  Speaker 0: Breaking that down!
• 13:47
  Speaker 0: Microsoft is Up Twenty Five Percent, Google Eighteen Percent And Amazon Sixteen Per Cent.
• 13:53
  Speaker 0: Martin points out the glaring hypocrisy here.
• 13:55
  Speaker 0: Hyperscalers often blame consumer demand for this spike claiming they simply can't decarbonize fast enough to meet our needs.
• 14:02
  Speaker 0: but They Are Actively Pushing Unasked For Compute Heavy AI Features On To Users.
• 14:08
  Speaker 1: AI overviews in search are the perfect example of this.
• 14:10
  Speaker 1: Exactly!
• 14:11
  Speaker 1: You can't turn them off, and generating an AI summary uses drastically more energy at the inference level than a traditional database query search.
• 14:19
  Speaker 1: They're forcing the energy spike And then blaming the consumer for using the product.
• 14:23
  Speaker 0: While physical footprint emissions grow The regulatory bite meant to control it might actually be softening.
• 14:28
  Speaker 1: Yeah Daniel Bowman posted that EU's revised European sustainability reporting standards The ESRS have actively cut mandatory data points by over sixty percent.
• 14:38
  Speaker 0: Sixty percent?
• 14:39
  Speaker 0: That's
• 14:39
  Speaker 1: huge!
• 14:40
  Speaker 1: It gets worse on the infrastructure side.
• 14:42
  Speaker 1: Nathaniel Barola noted that a planned EU traffic light rating system for data centers, which would have publicly rated facilities based.
• 14:50
  Speaker 1: their energy and water use is actively being watered down after intense lobbying by Big Tech.
• 14:55
  Speaker 1: Of
• 14:55
  Speaker 0: course it is.
• 14:56
  Speaker 1: They're argued that strict transparency will damage cost competitiveness.
• 15:00
  Speaker 1: so emissions go up And reporting requirements are systematically dismantled.
• 15:05
  Speaker 0: I want to slow down here for a second because we talk about emissions, like they're just abstract numbers on the spreadsheet.
• 15:11
  Speaker 0: But there is very real local human cost of this.
• 15:14
  Speaker 0: Right?
• 15:15
  Speaker 0: Dr.
• 15:15
  Speaker 0: Sascha Lucciini brought up a Reuters analysis of XAI's new Colossus II data center in Tennessee.
• 15:21
  Speaker 0: The
• 15:21
  Speaker 1: scale that facility was unprecedented
• 15:23
  Speaker 0: It IS!
• 15:24
  Speaker 0: And then the power grid couldn't handle it.
• 15:26
  Speaker 0: so they brought in massive gas burning turbines
• 15:31
  Speaker 1: trucked-in gas turbine.
• 15:32
  Speaker 0: Yeah, and the emissions there are far beyond the threshold that normally requires a federal permit And it is adding disproportionate smog in pollution burdens to local communities of color right there in the area.
• 15:43
  Speaker 0: Wow
• 15:44
  Speaker 1: This is critical.
• 15:45
  Speaker 1: We're turning AI from a fascinating tech tool into a serious environmental justice issue.
• 15:51
  Speaker 1: We are exporting the physical cost of our digital convenience directly into local neighborhoods.
• 15:57
  Speaker 0: It is a sobering reality, The cloud isn't in the sky right?
• 16:00
  Speaker 0: it's someone's backyard.
• 16:02
  Speaker 0: But we also have to recognize enterprises that take real structural accountability rather than just dodging regulations.
• 16:10
  Speaker 0: Amir A Kangaloo shared how Deutsche Telekom is tackling this head-on.
• 16:14
  Speaker 0: Their SAP to SKY cloudification initiative successfully cut their energy consumption by a verified forty percent.
• 16:21
  Speaker 1: That's massive for a telecom giant!
• 16:23
  Speaker 1: How are they managing the AI side of things though?
• 16:26
  Speaker 0: They launched program called Detoxit.
• 16:28
  Speaker 0: Instead of using tools like Gemini to generate endless new marketing content, they're using generative AI as digital archaeologists.
• 16:35
  Speaker 0: Oh that's interesting!
• 16:36
  Speaker 0: Yeah They use it map understand and safely decommission messy thirty year old legacy infrastructure That was just sitting around burning zombie power?
• 16:43
  Speaker 0: That's
• 16:43
  Speaker 1: brilliant.
• 16:44
  Speaker 1: As Joy D denoted in a recent post green ICT is no longer an environmental side project for the Marketing team.
• 16:51
  Speaker 1: It Is A hard boardroom mandate on how To scale a business sustainably.
• 16:55
  Speaker 0: This shift toward accountability is definitely becoming a global movement.
• 16:59
  Speaker 0: At the AI for Good Summit in Geneva, researchers highlighted the mechanics of energy grid mix.
• 17:05
  Speaker 0: They noted that training a frontier model at a location like Nevada emits ninety times carbon as training an exact same model in France.
• 17:12
  Speaker 1: Ninety times just because it's located?
• 17:15
  Speaker 0: Yeah!
• 17:15
  Speaker 0: Because France relies heavily on a nuclear and hydropowered grid whereas Nevada relies heavily natural gas.
• 17:23
  Speaker 0: The exact same code, the exact same compute.
• 17:26
  Speaker 0: Ninety times the carbon based entirely on where you plug the machine in and the stakes for getting this right are astronomical.
• 17:32
  Speaker 0: Riyad Medeb at a UN event warned that by twenty thirty data centers globally could use nine hundred forty five terawatt hours of electricity.
• 17:40
  Speaker 0: just to
• 17:40
  Speaker 1: put that abstract number in perspective That is the current electricity consumption of the entire continent of Africa.
• 17:45
  Speaker 0: Yeah we're building an infrastructure literally rivals continents
• 17:49
  Speaker 1: And the developer community is waking up to this reality, too.
• 17:52
  Speaker 1: We saw a huge organic engagement at community events recently like Green IO in Munich The True Price AII event series and the Coding WaterCant Festival Up In Kiel.
• 18:02
  Speaker 1: That's
• 18:03
  Speaker 0: great to see!
• 18:03
  Speaker 1: The sentiment is shifting.
• 18:05
  Speaker 1: Developers are actively demanding open standards digital sovereignty and real sustainability criteria for the software they write.
• 18:12
  Speaker 1: They're realizing that every single line of code they push has physical thermodynamic consequence.
• 18:19
  Speaker 1: Well,
• 18:19
  Speaker 0: we've covered a massive amount of ground today pulling threads from the water cost of a polite prompt to deterministic FLOP measurement down into underground Swiss data centers all the way to boardroom battles over ESG liability and environmental justice.
• 18:35
  Speaker 1: Which leaves us with one final provocative thing to ponder as you go back to your day jobs.
• 18:39
  Speaker 1: What's that?
• 18:40
  Speaker 1: We spend almost all our time in capital figuring out how to deploy AI everywhere As fast humanly possible.
• 18:47
  Speaker 1: But when you look at the true physical cost of this technology, the strain on our grids.
• 18:52
  Speaker 1: The evaporation of water and impact in local communities... What if the ultimate competitive advantage for digital leaders tomorrow isn't knowing how to use AI but having a strategic discipline know exactly where
• 19:16
  Speaker 0: not?