S2:E2: Building Tomorrow’s AI‑ready datacenters

00:00:02: Welcome to the ALSO AI podcast!

00:00:06: Discover the latest advancements in artificial intelligence through insightful discussions with

00:00:11: industry experts.

00:00:13: Each episode will bring a fresh perspective on AI and addresses their real-world challenges, opportunities

00:00:19: & interests.

00:00:21: you as an IT reseller might have or your customers could ask about.

00:00:26: Featuring experts from

00:00:27: ALSO and leading

00:00:28: Vandors,

00:00:29: we explore diverse insights on AI.

00:00:35: Welcome to also AI Podcasts!

00:00:38: My name is Henrik Asserheve.

00:00:39: I'm your host in the podcast.

00:00:42: i am working for ALSO as a European Group Sales Lead For Commercial Computing And Also Leading The Lenovo Tui-Sixty Covering Lenovo Product Lines Across Europe.

00:00:52: Together We Will Have An Exciting Journey Into The Future.

00:00:56: Each season of podcasts will be differentiated from different aspects within AI showing the real-world applications and their impact.

00:01:06: Every episode offers a new insight into tackle challenges, questions that might come to your customers – maybe they have them already or may not quite soon.

00:01:18: So topic for podcast today is build the AI-rated data centers tomorrow.

00:01:24: Where are we today?

00:01:25: What are we looking into, how will do data centers in a correct way to scale up the organizations ready for AI.

00:01:36: So everyone talks about AI models and we're focusing on how AI works in production.

00:01:43: but where should be start?

00:01:44: How Should We Scale And How Do We Do It Sustainable?

00:01:48: That would be the topic of today.

00:01:50: First I'd like to introduce our audience that are joining today, so we have four guests.

00:01:58: You will present yourself in a few minutes here.

00:02:01: I'll start with.

00:02:01: you again.

00:02:02: can tell me where your coming from?

00:02:05: what is the background?

00:02:07: Hi Enrik thanks for inviting me.

00:02:09: my name is Jan Baumann and field application engineer maybe more known as pre-sales to many of customers under technical backbone of our sales organization On one hand, our main customers with their investments.

00:02:27: And on the other hand I'm also doing a lot of trainings in the channel for these small medium business general customers and I am with NDE since almost nine years now.

00:02:38: So you're from Silicon vendor?

00:02:41: Yes Delivering data or delivering components into infrastructure that we are talking about with Lenovo.

00:02:49: That would be good.

00:02:50: handover to Simone.

00:02:52: Hi everyone I am Simone Larson.

00:02:56: I'm based in EMEA, and I lead enterprise adoption of infrastructure solutions within Lenovo.

00:03:04: so i work primarily with the ESMB segment And it's really about how do we get them on that path of scaling AI?

00:03:14: Most organizations start experimenting in the cloud but as AI maturity and adoption increases

00:03:21: they are

00:03:22: Looking at a hybrid report or data center approach and that's where I come in.

00:03:25: Looks really good, so thank you And welcome Simone as well.

00:03:30: hand over to you Alexandra.

00:03:31: Hello

00:03:32: Henrik and Thank You for the invitation.

00:03:36: i'm alexandra sanake.

00:03:37: i am a sustainability consultant here at Lenovo leading IT decarbonizations engagements across EMEA And what I do is basically help enterprises reduce their environmental footprint of technology and building practical roadmaps that cut energy use, water consumption waste and emissions across data centers.

00:04:05: That's a topic becoming more relevant.

00:04:08: some have maybe not faced it yet but be a higher interest within the next one to five, ten years from now.

00:04:18: So really interesting view what we're looking at there.

00:04:21: and finally my good colleague Tim will you introduce yourself?

00:04:27: Hi!

00:04:27: My name is Tim.

00:04:28: thank-you for inviting me Henrik.

00:04:30: I work with also as a pre sales specialist so i worked mainly seeing how we can help them on their infrastructure solutions, pretty much supporting as a technical consultant for AI projects or partners and customers.

00:04:49: So you're starting to see a lot of excitement around AI in our.

00:04:53: SMBs are asking How Can They Join In On This Market?

00:04:57: Thank you team.

00:05:01: Let's warm up a little bit here.

00:05:03: So where are we today?

00:05:04: What I'll be looking into?

00:05:06: so let me first of all start with like a super simple question, so Do you believe that the power and cooling on our bigger constraints in to AI scaling more than the compute itself And it's free just to break in now?

00:05:22: Absolutely do.

00:05:23: thank you for that question.

00:05:25: I think we all hear of these big deals and AMD recently announced some big deals with major customers investing into something like six gigawatts of installed total power.

00:05:39: That's what six nuclear plant blocks can deliver, absolutely.

00:05:44: that will be a constraint more than other resources needed to build up in the certain countries where this data centers are being built-up And it's also important to make sure that the cooling does not eat like thirty-to forty percent of their power.

00:06:03: That is where liquid cooling comes into play.

00:06:06: Okay, good!

00:06:07: Let us see what we are going there for a harder introduction and go in big problems already by now.

00:06:13: but this something I want you talk about.

00:06:15: how can find solutions?

00:06:17: so jumping into sustainability from.

00:06:20: So do you believe that sustainability decisions are being made too late in the planning from where we are today?

00:06:30: My opinion is, Eric.

00:06:31: We fail at planning for sustainability since the beginning.

00:06:37: yes so sustainability comes a little bit later.

00:06:42: And then if we talk about powering cooling technology specifically, there's also a lack of internal alignment between different departments in the company.

00:06:54: So let me just give you very simple example.

00:06:59: If We Think About Liquid Cooling Technology Now procurement department might be happy because They will use less energy, so they will pay less for their energy bill.

00:07:12: The sustainability team will be happy because they reuse the heated water from a system in their facilities.

00:07:21: So there is circularity and resources that are used by IT department.

00:07:26: who'll be happy Because it has an interesting technology built-in.

00:07:32: But then to make this happen just might need break through few walls.

00:07:37: So the facilities team will have something to say about that.

00:07:42: When we talk about cooling technologies and sustainability, there's also a need to look across-the-board at what impact that specific solution has inside of company throughout different departments and start making alignment for sustainability all together from the start.

00:08:07: So good thoughts about how to plan in time and now we are delivering the next generation of AI structure.

00:08:14: Do you believe that organizations are ready for a hybrid AI solution from where they are today, in the way that they're building?

00:08:22: AI facilities within the organization... Where do I think they started?

00:08:26: Do i think there already is a hybrid solutions meaning like cloud on premise?

00:08:33: I can weigh in on that, it's Simone.

00:08:35: So what i'm seeing particularly in EMEA is exactly what I mentioned when I introduced myself.

00:08:42: so they're starting with POVs and poof of values and minimum viable product experiments in the cloud right?

00:08:50: The challenge we are often getting to Is When They Decide To Scale Those Workloads Those AI workloads into production And they come up against a number of issues.

00:09:00: It might be data sovereignty, the EU AI Act or regulation within their own industry.

00:09:07: and that's where the aha moment comes kind of like sustainability topic.

00:09:12: They have to take step back an assess.

00:09:14: A are there ready?

00:09:16: B what it actually means to be hybrid C is it a capex or opex discussion?

00:09:21: right And how do they break through that?

00:09:25: So it's not like, They're not ready.

00:09:27: I don't think there really.

00:09:28: thinking about it from the start

00:09:30: Let see where we end up.

00:09:31: jump to a two.

00:09:32: The first topic and i would like to cover here.

00:09:35: so power on permitting.

00:09:36: How are scaling in In the world behind?

00:09:40: so We cannot change what is outside of the organization or how the behavior Is at the market behind.

00:09:45: but how can be we Do So?

00:09:48: maybe to use Simone again, so when power availability and permitting is delayed.

00:09:55: And shows a slow expansion how should organization rethink the capacity?

00:10:00: When they're planning AI development

00:10:03: I think we should have a view of where the compute needs to reside right To gain those efficiencies, it's all about bringing the compute close to the data.

00:10:18: That's where they really see the gains in latency and total cost of ownership for that AI algorithm and that model.

00:10:26: I think it's also just having a holistic look at their AI workload estate then trying to figure out how do you bring your computer as close as possible with workloads needed?

00:10:39: Some do not!

00:10:41: Some can be a bit latent, so I think it's holistic strategic look at how they can flex the levers to maximize power and cooling for certain workloads because that will bring costs down.

00:10:55: And then we'll bring sustainability lever up and perhaps minimize some work loads accordingly.

00:11:03: There is play that could be made with infrastructure solutions group solution and it can be a case of trying to figure out, is it best to leverage the CAPEX model through our true scale offering or isn't really retrofitting their data center?

00:11:19: Sorry.

00:11:20: The OPEX is the true-scale and then the CAPEx looking at that data centre refresh.

00:11:26: So just make reference from what we had in the podcast before.

00:11:30: this Is when you talk about PCs how they develop AI models on PC.

00:11:36: Now we're looking into how can you do it within the organization and putting that closer to people who are actually using models.

00:11:44: So for Jan, when you look at it from your point of view with AMD... How is the AMD's efficiency per vet in the organizations to increase or optimize the way of building data centers?

00:12:00: Does this make sense?

00:12:03: Absolutely!

00:12:04: Thank-you so much for that question.

00:12:06: AMD is actually not always known by all these IT decision makers for being so sustainable.

00:12:15: There were some times in the past when we weren't focused on efficiency, but in the meantime... We totally shifted our focus also into design phase of our products.

00:12:28: So from the beginning on really designed our product to be as performing and as efficient ever possible.

00:12:37: And yeah, that's also can be seen even in the top five hundred supercomputer list where AMD really offers number one of two supercomputers.

00:12:50: more importantly you see if we look up data there we consume less than fifty percent power per flop As the number three, which is for more competitor.

00:13:04: So we can really show even in the biggest scale you can get today on the planet that we are more than two times more efficient then others using the GPUs and CPU as a combination of these supercomputers.

00:13:19: Okay so mixing... The AI's quite

00:13:21: same workload

00:13:22: I would say?

00:13:22: Yeah!

00:13:22: Mixing it with your doing this also taking down power consumption from data centers to be most stable in the way that we are developing AI data centers.

00:13:34: That is of course something you need to be aware when doing investment, not after it's done but consider from beginning then.

00:13:44: So Tim your facing customers on a daily level how they're doing so with customers hitting power limits what can be practical ways balance products instead of just stalling and take other decisions?

00:14:00: The

00:14:01: focus shifts to, well how much performance per vat.

00:14:04: So that's where AMD helps because they can focus on getting better equipment for it and we're trying to scale what they have Because We Can't Just Get More Power When There Isn't And... ...We Have A Lot Of Issues Coming.

00:14:18: How Much Rumors Is Gonna Take And The Cooling Capacities And Rack Density That This is the Real Limits We Are Hitting With Small Customers.

00:14:26: They Don't Have The Room For It Or The Power.

00:14:29: That's why we need to scale till the needed levels.

00:14:35: Does that also mean remodifying the data center when they have something already?

00:14:41: Maybe you should convert a lot of it just to see if you can get more power and more operating, but with less power consumption or maybe less cooling...

00:14:53: There's a lot of decisions to be made, so they are trying to move it into a hybrid setup more.

00:14:59: So that can have little bit with both liquid cooling and hybrid cooling for air-cooling And then try to see if there is something where the cloud has to move some of them while still having on premise For sensitive things.

00:15:15: Great!

00:15:15: It already looks quite interesting Alexander I would like you say because You had from service point of view as well And I know yet when you look into this, You are more into the right sizing.

00:15:26: So how can we say that?

00:15:27: This is done in a rightsizing way.

00:15:29: How Can We Help The Customers To Do That?

00:15:32: Exactly Henrik.

00:15:34: so what i see quite often especially In The Early AI Project Is That Sustainability Challenges Don't Start At Scale.

00:15:45: They Actually Start Much Earlier In The Experimentation Phase Because Teams don't know yet what they need, so they tend to over-provision.

00:15:57: More compute more capacity just be safe and that creates cost but also unnecessary carbon before the project even proves value.

00:16:10: So for me The AI ready infrastructure is not about scaling up.

00:16:16: The real opportunity is to design for flexibility from the beginning.

00:16:21: So start small, fast, iterate and then scale based on actual usage rather than assumptions because if your infrastructure can adapt to real workloads.

00:16:35: you basically avoid idle capacity You reduce wasted energy And don't bring in hardware that you actually need just yet.

00:16:46: So this is where flexible models like Lenovo TrueScale become very relevant from a sustainability perspective by allowing to align that capacity with actual usage instead of locking in assumptions too early, and this is particularly important for the SMBs because they don't have the margin for trial-and-error scale

00:17:10: exactly.

00:17:11: yeah

00:17:11: so Right sizing is not just about efficiency.

00:17:15: Here it, It Is what actually makes AI adoption possible in a more controlled lower risk way In this context.

00:17:25: Okay make sense.

00:17:27: with that I would say let's jump into our next topic.

00:17:29: That is About the liquid and hybrid cooling?

00:17:42: silicon point of view.

00:17:44: when you do look at how cooling are being made so the GPU.

00:17:50: Heat or heat from the GPUs are increasing heavily, when does air-cooling stop being possible?

00:17:57: It depends a bit on how big your build these servers.

00:18:00: So there are solutions available.

00:18:02: still today we can have like an eight u six U ten new server so big huge box and that can still be air-cooled with all these up to eight big GPUs in there.

00:18:17: There are solutions also with Lenovo available using this GPU, the issue here is more coming back to sustainability thing something like thirty or forty percent of total server when it's being air cooled just consumed by fans into cooler system.

00:18:37: Okay, so it would be more sustainable to invest that additional power.

00:18:42: That's what we did with our current GPU generation.

00:18:45: There is an option of the mi- three fifty and three fifty five x Using one thousand watts per GPU And then can be accrued?

00:18:55: That's intended to be a cool.

00:18:57: and then there's another option Of one thousand four hundred watts where we actually took that budget from the fans And said, okay we consume it and then it needs to be liquid cooled in most of the cases.

00:19:10: Okay

00:19:10: yeah

00:19:11: Then you get way more performance out of your power but You need to invest into that direct liquid cooling or hybrid cooling solutions.

00:19:20: Yeah so a shift being made there from your side as well?

00:19:24: Okay Simone From a little point-of-view So how do you help your customers choose their right design To balance all that?

00:19:34: So I think it's the right design.

00:19:36: comes back, especially for AI to again their workloads.

00:19:41: Again too having a bigger view of.

00:19:45: is that they want to transform a particular business unit?

00:19:49: Do they wanna transform our business process or do you wanna transform your organization as whole?

00:19:56: so with Lenovo we have scalable solutions leveraging Neptune liquid cooling technology that's built three point five times better to deliver thermal efficiency with one hundred percent heat removal.

00:20:12: So, with liquid cool they are able to start small.

00:20:16: provision the right infrastructure in a manner of that will reduce power and cooling costs Right?

00:20:23: The Neptune solution uses forty per cent less energy And it reduces that energy with a fanless direct liquid cooling or liquid assisted cooling.

00:20:33: So, its trying to figure out where they want start and what is the right size infrastructure roadmap in path leveraging the liquid cooling technology.

00:20:47: With that said, I would do a handover to Tim because you have the face-to-face with customers on daily level.

00:20:54: What concerns are facing when talking about liquid cooling or hybrid cooling?

00:21:01: What dialogue is there today for

00:21:04: them?

00:21:04: The dialogue here is mainly pricing as we start looking into AI.

00:21:11: A lot of small customers want to get in this AI and not sure where to start.

00:21:16: So we have to look into the whole setup on what they have of room and their price ranging, and sizing is so.

00:21:24: that's why We tend to stop by asking What do you need it for?

00:21:28: And a lot people actually don't notice yet They want get in on the AI market The whole wave of this AI data centers but they actually Don't know what they wants use It For!

00:21:38: So we Have A Lot Of Communication Inside.

00:21:41: Where Do You Want To Start?

00:21:42: How Much Do You Actually Need?

00:21:44: That's where we are at the moment.

00:22:06: Yes, absolutely.

00:22:07: And I would actually also go a little bit further than that because yes in core environments AI is increasing compute density significantly so technologies like Lenovo Neptune liquid cooling become essential to sustain performance.

00:22:26: you know more energy efficient way now.

00:22:29: if we look at how AI We can talk about hybrid and it not being a choice anymore, but the default nowadays.

00:22:41: So in a hybrid world what we see is workloads that are moving between cloud on-premise an increasingly the edge depending on where the data is And It means looking at what latency you need and how much control you want?

00:22:59: That has a direct impact on both cost and energy.

00:23:03: So for example, pushing everything to centralized environments is not always efficient.

00:23:10: In some cases running inference closer to where the data is generated reduces data movement and improves responsiveness in lowers overall energy demand.

00:23:23: so this SMBs can really participate in AI by using edge and distributed models instead of trying to replicate large-scale data centers.

00:23:36: Yeah, but to really manage that sustainability layer organizations will need visibility.

00:23:43: so this is where tools like Lisa Lenovo Intelligent Sustainability Solutions Advisor can help translate infrastructure choices into lifecycle carbon impact at the endpoint level, where many organizations still lack visibility today.

00:24:05: So what this does is basically it allows fleet management decisions to be guided by real data and at the end sustainability's not just about optimizing infrastructure and energy in isolation.

00:24:22: It is about making adapted and informed decisions across the entire system, depending on the business size and business

00:24:31: model.".

00:24:33: So a great aspect to look into then.

00:24:36: so you talked a little bit about Edge as well.

00:24:39: that will lead us into going through next topic where we say okay from edge-to-data center which would call it hybrid AI in practice.

00:24:48: Let's go into that topic and see how we can dive a little bit deeper.

00:24:53: Simona, when you're looking at hybrid AI why is Hybrid AI becoming the default architecture rather than an optimization?

00:25:04: I think it's to defaults.

00:25:06: so what i'm seeing with customers who are truly scaling AI and either or conversation, right?

00:25:15: So it's not either cloud or on-prem.

00:25:18: Right.

00:25:18: so its having a look at where the data is collected.

00:25:22: how do they reduce that latency And How Do They Bring The Compute And The Processing Closer To The Data For example if you are an SMB in retail Or even On The Manufacturing Floor You're Collecting Data Either About Customer throughput the operations of your shop floor, it's prudent to have a think about how do you localize that data collection?

00:25:50: Do the processing onsite and then perhaps send it to the datacenter or send it into the cloud.

00:25:56: Organizations are really having a relook as to what does it mean to have our datacentre fast approach in todays AI fast-moving environment.

00:26:11: The architecture is there to support hybrid, the infrastructure's there for a support hybrid.

00:26:17: And it's about how they contextualize that and their organization in their workloads.

00:26:22: Okay!

00:26:24: Jan from an AMD point of view so... How do AMD platforms enable AI workloads?

00:26:30: To move across from edge into co-environments?

00:26:34: Yes we have all different ingredients needed for AI The supplier having the broadest portfolio in the market.

00:26:44: So there's one that always puts a big GPU on every AI problem or challenge, There is another one That always put CPU and we have both words.

00:26:56: We are heavily working on broadening our portfolios.

00:27:00: so In the meantime acquired companies Having programmable logic for building accelerators.

00:27:07: That's also the technology that goes into our client devices called an NPU, New Europe Processing Unit.

00:27:15: And we also work heavily and invest in two open standards for putting all of those together.

00:27:22: so on one hand a new product supporting OpenUltra Ethernet which is bit off own approach to connect these servers well together no matter how many are or how big they.

00:27:37: And we also have an open standard called Ultra Accelerator Link, so that's the counterpart to NVLink.

00:27:44: To connect together all these big GPUs in the super data centers.

00:27:49: but it is not very relevant for small and medium business customers you are talking about today.

00:27:55: So here its more important to have a right product from the Big Data Center even to the Edge, where there are also edge servers offered by Lenovo.

00:28:07: Where you can put in some small accelerator cards maybe PCI Express GPUs things like that and AMD has these GPU's as well.

00:28:17: We're coming up with more.

00:28:19: so please stay tuned for PCI express solutions here.

00:28:24: And we talked about NPU just to mention that The audience is not really aware of the NPU, then you can just go back and listen to other podcasts.

00:28:35: Then they will have more deep dive into that.

00:28:38: Tim as being so close with customers on a daily level do we have some examples where HAI has delivered faster decisions?

00:28:49: We had case with local factory actually sometime ago when they put in these edge computers for monitoring the belts and what product was going across.

00:29:02: So this is a point where they actually take it all the way to the edge point, and process that data out here instead of moving back into a data center.

00:29:10: so you reduce all these latency and avoid the data movement.

00:29:15: They can reduce cost in networking as well.

00:29:19: That's an actual case when we use the Edge computing on the Edge of our whole process.

00:29:26: So if the customer is facing a high cost or high time consumption, then it might be interesting to look into how its computing and on-premise data center can maybe solve some of this with AI solutions.

00:29:38: so they take the production.

00:29:40: Or whatever topics that are looking at from their customer point.

00:29:46: For

00:29:47: example, some of the larger factories maybe don't have a room for a data center but they need this AI computing on-premise.

00:29:57: So instead of having these offsite and having to move all this data They just get their edge computing all the way out till factory flow instead.

00:30:05: so that's huge benefit.

00:30:07: you'd reduce cost for networking and latency.

00:30:11: where Edge Computing can actually help customers

00:30:14: Put it closer too.

00:30:15: things are really happening.

00:30:17: So based on services, Alexander coming from edge to data center like what problems can we face here?

00:30:26: What I'm hearing all of my co-speakers today is the need for resource management.

00:30:35: Whether it's space energy and compute power.

00:30:40: so one thing that is still underestimated in the AI discussion, it's that impact of hardware lifecycle.

00:30:51: Because new workloads and new performance requirements they often give the instinct to replace infrastructure faster.

00:31:01: so AI is basically accelerating refresh cycles But that has a direct impact on embodied carbon, especially in servers and endpoints.

00:31:15: So if we talk about those AI ready data center?

00:31:19: We also need to talk about how long the infrastructure actually lives And what happens to it afterwards.

00:31:28: extending the usable life of an equipment is one of the most immediate and practical ways to reduce cold three emissions.

00:31:38: And decrease total cost of ownership, so you get carbon out and cost out altogether.

00:31:46: now thinking about life cycle from the beginning?

00:31:51: You remember we were talking about planning From start.

00:31:55: that means planning not only for budget from a procurement perspective but also for the maintenance, redeployment recovery refurbishment and end of life management.

00:32:08: This is where for example services like Lenovo Asset Recovery allow organizations to securely recover assets at End Of Life And also capture residual value through financial credits.

00:32:24: So sustainability and financial return start to align but we need to plan for it at each refresh cycle.

00:32:35: And when these elements are structured, sustainability stops being a reporting exercise and becomes something that you can actually manage operationally.

00:32:48: So I think introducing the circularity concept here is key as part of how you design AI infrastructure end-to-end.

00:33:00: Great.

00:33:01: Let's go over to scaling and I will keep the conversation with you, Alexander because you are into this as well here.

00:33:08: so for scaling without losing control.

00:33:11: So how do circular models like asset recovery and CO two reporting support responsible AI growth?

00:33:19: Absolutely The circular management of infrastructure is key here Because it allows to dispose off end-of life assets with security, so you get the certification that your data hasn't been lost and everything is secure.

00:33:41: And then there are also environmental certificates for the conscious dispositioning of the assets cannot be given a second life.

00:33:51: So circularity in this case When we talk about endpoints, for example.

00:33:59: We can also talk about refurbishment and a second life or those assets.

00:34:05: so Refurbishment service that basically gives the new life to The machines that you already know And you have used any already trust?

00:34:16: So it kind of takes out the unknown from the equation of using those second hand assets.

00:34:24: They're not really a secondhand, it's just the Second Life of the assets that you already owned and used.

00:34:32: And let's look into what Simone to you where do two scale fit for customers who want to have flexibility but still retain applications in security control?

00:34:43: Security is becoming more relevant.

00:34:45: so security.

00:34:47: it might be number one topic, security.

00:34:50: But how do true scale fit for customers to bond that flexibility and application control together with the security?

00:34:59: Yeah so in my conversation with customers what I'm seeing is that their aspiration to scale AI isn't necessarily grounded the current state of your data center, right?

00:35:14: So it's oftentimes all.

00:35:16: It's a need for modernization and that where not only AI-ready infrastructure comes in but those security considerations are for AI.

00:35:27: so when they're looking at which route to go oftentimes true scale comes up because it is a flexible, resilient secure solution that they don't necessarily have to wait for CAPEX for.

00:35:42: They can continue with their long-term plans of data center refresh and modernization but at quickly leverage the TrueScale solutions privacy and that hybrid approach built in to help them get to their aspiration.

00:36:00: In the short term, I think what companies are also having a look at is how do they localize development as well?

00:36:08: So we're seeing not only a look of true scale but a look to build and train models locally, right?

00:36:21: So circumventing their oftentimes antiquated data sent in the fast instance.

00:36:27: And then pushing those models into securely using a true scale model.

00:36:32: so there are number of considerations that are being made but I think resilience of true scale, security and flexibility.

00:36:43: And I guess option it gives customers who are on that refresh cycle is second to none?

00:36:50: Yeah i think what's interesting about TrueScale if we think about from a sustainability perspective Is that It is fully sustainable solution form the full ESG spectrum.

00:37:02: on the environmental side because you avoid over provisioning, therefore.

00:37:07: You avoid emissions for the unused hardware.

00:37:11: from a social perspective Because it comes with managed services so it allows your IT team to focus On more higher ROI initiatives and you deploy Your human resource more efficiently.

00:37:28: And then from a governance perspective because it allows you to save cost from that initial investment, be more efficient as you scale flexibly and also has that whole circularity built in.

00:37:44: So yes I think from a sustainability and full ESG perspective Lenovo TrueScale is my favorite!

00:37:52: It's your favorite?

00:37:54: Good to have a favorite.

00:37:55: so then on the wishlist for Christmas maybe yeah Jan, when you look at the scaling with AMD how do we get that?

00:38:06: So I think it's important.

00:38:07: We had that right sizing thing very often mentioned today and i think thats a huge topic And also customers using too much compute power to expensive systems.

00:38:22: so where AMD comes into play again is to help the data center to the edge or even the client on one hand.

00:38:33: And then also, use the right technology.

00:38:35: we had that example of production line there.

00:38:38: you could even use some AMD programmable technology based on FPGA Technology.

00:38:45: here and There We have reference projects with That as well.

00:38:49: when it gets through The GPUs we Have the advantage that we can provide more memory per GPU And we usually are better available than the other guy out there at somewhat the same price.

00:39:03: Okay, now we have all these memory constraints up here but AMD has a clear value proposition that you can do more per GPU models bigger models then with other solutions and sometimes it makes even sense to just consider using only this CPU as your workhorse in your server and doing some local inference.

00:39:28: So use your model locally based on a CPU only, And here AMD has the advantage that we scale much wider in the CPU space of data centers so customers can choose whether they want to use.

00:39:44: maybe then coming back Decide to use a single socket solution with AMD only being more relaxed in your rec space and power.

00:39:57: Use of your rec.

00:39:59: And then put more servers into your rack until you're at the power limit.

00:40:03: off the wreck, and Fill your existing infrastructure better than also give that Infrastructure kind for second life.

00:40:13: That's where amd helps because we can just provide up two like two acts performance per watt and per socket than other solutions in the CPU side

00:40:22: of things.

00:40:23: Okay, Tim?

00:40:25: The real customers what do they really think when doing a balance with their scaling setup... When you look at risk control and accountability how does it scale up-and down?

00:40:37: where needs are there?

00:40:40: Well.. They look into all parameters.

00:40:42: so we look at sustainability and need from customer's sides.

00:40:48: They need the right tools where TrueScale can help them to size it from the start.

00:40:53: And they try to avoid getting locked in, where a hybrid architecture could help them.

00:40:59: and then visibility into what will cost?

00:41:02: I think that's the scaling without losing control.

00:41:07: If there is one thing customers don't want to lose control.

00:41:10: So all of this we have touched today control of the data, having the option to be flexible.

00:41:20: To have a cost-effective model.

00:41:24: power consumption cooling everything has been part of the topics for today.

00:41:28: I will ask now each of you so if You can choose from what we talked about Today.

00:41:34: We are closing down the podcast within few minutes.

00:41:37: What would it take away?

00:41:42: when they're closing it down and like, okay what should I take from here?

00:41:45: What should i do just after listening to what we have talked about.

00:41:50: Who volunteers to take it first?

00:41:55: We are

00:41:58: all enthusiasts

00:42:01: here.

00:42:01: yeah that's so fine!

00:42:02: So Alexander what would you recommend the audience?

00:42:07: After

00:42:11: thinking through our topics today and listening to my colleagues here, I think from a sustainability perspective what is clear for me?

00:42:22: that most organizations don't fail on sustainability strategy.

00:42:27: They all want the good they have their strategies nailed down but do struggle with how decisions are made every day because when it comes you cannot decarbonize through ambition alone.

00:42:44: You decarmonized to life cycle decision, and it makes me happy to realize how sustainability consulting plays a critical role today as that thinking linking layer between IT procurement facility sustainability teams helping companies translate targets into value added operational choices across the entire lifecycle.

00:43:09: so Yes, I would say for the companies who are listening in to us think about not only your sustainability strategy but then next step.

00:43:21: About how you put it into practice and link it through your process end-to-end.

00:43:29: So because AI maybe is a new AI any longer?

00:43:32: But still many companies will like start with some AI closer creating the right platform from the beginning and get some advices to that.

00:43:43: So a really good, hard statement is not an easy task you're asking people do but it's very important for them at this point in time.

00:43:51: Let us see if Simone can have also a hard one just after here?

00:43:57: It's similar vein.

00:43:58: I'm just jotting note down.

00:44:02: so what i often times see across the media urgency to over-provision infrastructure.

00:44:10: I think we touched on that, I would urge listeners just take a step back and really assess what your short medium or long term goals are for that infrastructure.

00:44:22: you have the provision in line with business units coming out within your overall AI strategy if oftentimes buy a seven-layer wedding cake when they need like, smaller cupcake.

00:44:38: So

00:44:38: assess the market to determine what modular solutions... so I'm going to put plug in here for RTX Pro Six Thousand which is scalable two use solution with multipurpose GPU that can scale agentic workloads, that can reduce the cost of tokenization which is a new form of currency and measuring AI efficiency.

00:45:01: So take a step back.

00:45:03: assess how modular you can be in your approach but also don't discredit hybrid just if it's a or statement, It can be an and statement as to how you provision this infrastructure improve sustainability.

00:45:19: And reduce the total cost of ownership in the long run?

00:45:23: I like the way that they put into cupcake.

00:45:24: i'm getting hungry already.

00:45:26: so yeah Jan what is your takeaway from an AMD point-of-view?

00:45:34: What should the audience do listening to our podcast then too To take some clear actions just Do this.

00:45:43: Yeah, so I would advise customers to think a bit out of the box.

00:45:48: it's a bit in the same way that was already stated here.

00:45:52: many still today say okay always bought that vendor and that solution And i will do that on.

00:45:59: we are in a more challenging world Today.

00:46:03: there are availability constraints price constraints A lot around.

00:46:10: People should think out of the box and also consider other more open solutions to not get into some lock-in situations, maybe their solution is quicker in their own data center.

00:46:26: Secondly I perfectly agree with what was said already.

00:46:29: try to write size your solution.

00:46:32: Try to tailor it for workload, use the right components there To best address you workload.

00:46:39: Excuse

00:46:41: me, Tim.

00:46:43: Do you know who's next?

00:46:45: Yeah I think my closing for the customers whoever listens to this podcast is that scaling AI isn't just about building the biggest systems.

00:46:55: it's about we build the right system within real world limits they have.

00:46:59: so if they're cooling their powers sustainability and what can do So They Can Keep This Control And Involve It Over Time.

00:47:08: maybe build smaller to start with so it fits the real use case and then they can evolve that, get a bigger system.

00:47:15: I think this is my closing point for today's podcast.

00:47:19: So by that i believe we are to finalize And hope everyone who is listening that you're ready to build your next data center starting from the small cupcake or building up into a bigger one.

00:47:33: Let us see what will come in.

00:47:38: reach out to ALSU, AMD and Lenovo.

00:47:42: Ask what is it that you want us do?

00:47:44: We can help you with all of this.

00:47:46: I really appreciate having the discussion.

00:47:50: That was many useful experiences for sharing good advice from customers here.

00:47:57: Thank-you so much for joining me in our next podcast!

00:48:01: Let's see if we find more topics to talk about also again in the next season as well.

00:48:09: Thank you, goodbye!

00:48:34: Don't forget to check out

00:48:35: our show notes and

00:48:36: visit

00:48:37: our podcast website.