Hi. Good morning, everyone. A full house here today, again, I know many more joining by the webcast. Again, good morning and welcome to the second day of Vertiv's 2026 Investor Conference. We have an exciting day ahead. Before we begin, just a quick housekeeping pause. I would like to point out that during the course of this event, we will make forward-looking statements regarding future events, including the future financial and operating performance of Vertiv. These forward-looking statements are subject to material risks and uncertainties that could cause actual results to differ materially from those in the forward-looking statements. We refer you to the cautionary language included in today's presentation, you can learn more about these risks in our annual and quarterly reports and other filings made with the SEC.
Any forward-looking statements that we make today are based on assumptions that we believe to be reasonable as of this date. We undertake no obligation to update these statements as a result of new information or future events. During our conference, we will also present both GAAP and non-GAAP financial measures. Our GAAP results and GAAP to non-GAAP reconciliations can be found in the investor presentation found on our website at investors.vertiv.com. All right. That may have been AI, Lynn. Anyway, I am thrilled. Today, we're going to talk a lot about technology and innovation, so how we are shaping the industry with our technology and innovation. It is my great pleasure to introduce our Chief Product and Technology Officer, Scott Arnold.
Tokens are helping us break new ground on a scale never attempted to empower the world.
All right. On a scale never attempted to empower the world. We're attempting it. We're in the midst of it. We're driving this. We're shaping the industry. Thrilled with the turnout. Wasn't sure how the overnight session would go. Got peppered with a multitude of incredibly good questions last night. Impressed and humbled with the technical knowledge, the awareness, the understanding of where the direction is going. I think indicative of the time we're in, the industry we're in, and the trajectory we have going. I have probably four hours worth of technical content that we will try to cram into one hour here. I will go fast. I will go deliberately fast, and I will try to leave room for ample questions at the end.
What you'll see today is a pretty significant and healthy dose of architecture evolution and changes that are happening because AI and because data centers need to evolve to achieve and meet the performance standards, the efficiency effects, the big-picture dynamic changes that need to happen. Let's dive in. I'll re-anchor by saying and reminding, Vertiv is the leading innovator with the most complete end-to-end portfolio. We talk about this in terms of power management, thermal management, IT systems, infrastructure solutions, and we wrap the whole thing with services. This will be the framework and the guide work for how we walk through some of the technology evolutions. We're talking converged infrastructure as well. The vocabulary will change. The way in which we're discussing some of the problems we are tackling is going to change and evolve.
We will move briskly through this, but this isn't an easy story that is moving to one converged answer. Maybe the seed I will plant as we're talking through the technology changes here, this will be multipath, this will be multimodal, this won't be unidirectional in terms of the end solution we arrive at. Maybe disappointing for some folks that are looking for the answer. I think it refers back to the complexity that we see in the industry today, the importance of the breadth of portfolio, and the understanding of how solutions need to be paired together to solve what are incredibly, almost inconceivable challenges that our space is undertaking today that we wouldn't have even been able to comprehend or think about 3 years ago. We hit this yesterday. Density, speed, scale, complexity, and load profile.
We are seeing AI and GPU-based racks at 140 kilowatts today, quickly approaching 300, moving to 600, and we have a megawatt rack in the time horizon and in our visibility. Incredible density problems, incredible scaling challenges that come with rack density and densification at this level. The speed at which we are driving the industry, that's time to first token, that's deployment speed, that's truly performance as well, is dramatically increasing, and we are doing this at a scale where 250-megawatt buildings and gigawatt campuses are now almost. We're numb to it. It's almost normal. We think about the scale of a data hall used to be a megawatt. We're talking about a rack that is a megawatt. We're talking about a building or entire campus that used to be 250 megawatts, we're scaling to a gigawatt.
We're doing that for cluster size, we're doing that for performance, and we're doing that for the scale at which AI needs to be deployed. We're seeing dramatic change. With that scale comes greater complexity. The amount of piping, the amount of heat rejection on a site, the amount of interconnects over the course of what is potentially a small city becomes immensely complicated. We start talking about how these different subsystems, different products, and different solutions stitch together. Those seams and that area of complexity is increasing dramatically.
That complexity favors a company like Vertiv that can think about solutions and problems end to end and can understand how some of those issues at the seams need to be scaled, need to be rectified, need to be understood and solved for in order to make the whole system operate better and more efficiently. We talk about the load profile, and I will hit this quite a bit. The AI load profile is dynamic and synchronous. Dynamic meaning it ramps up and down. It is very somewhat volatile, and it is synchronous. We talk about the unit of compute. We are scaling GPUs to pods. We're building clusters that operate as single computers at a 1 megawatt, 6 megawatt, 10-megawatt building level.
We're talking about buildings that operate now as single computers that has dramatic impact to how the upstream infrastructure, how the grid, and how the downstream GPUs and silicon itself needs to operate and behave. Dramatic implications. Solving for these challenges simply is what AI depends on. The evolution of AI depends on us solving these challenges robustly and vigorously. We're also using different nomenclature. We introduced this a little bit yesterday in terms of how we think about an AI factory, the reason for a data center, how we unlock performance and efficiency. We start talking in terms of tokens per second is the new unit of performance. How many tokens, how many units of AI compute can I generate in a second? The push towards the latest, the newest, the frontier generation GPUs.
How many tokens can I generate in that watts or megawatts envelope is the new measure of efficiency. How many tokens I can generate per dollar of my total spend, of my total infrastructure, becomes the new way in which we're thinking about cost, and deployment speed is no longer my time for construction. It is my time to first token. From conception, from power availability to idea of a site, how do I go from the go button to I have an optimized, turned up, ready-to-produce GPU producing tokens in that space as fast as possible. Vertiv's uniquely positioned to influence and drive all four of those levers. We think about how. Tokens per second, latest generation GPUs that don't have to throttle because of electrical or thermal performance. We think about tokens per watt.
How do I take the allocated or interconnected amount of power at a site and ensure that the maximum amount of power isn't dedicated to support equipment, to cooling, to other ancillary gear, and the vast majority of that or the maximum optimized amount of that power goes towards the actual GPUs themselves. In terms of cost, how do I right-size the equipment? How do I get the design right from end to end so that I don't have to oversize equipment, I don't have to deploy more than is necessary for capacity for redundancy? How do I ensure from a life cycle perspective we have the end-to-end site operating with the maximum amount of TCO to enable that performance and that efficiency in the most effective way possible? We do that thinking differently about speed.
This idea of prefabrication, this idea of manufactured converged infrastructure that brings what can be thousands of skilled trade or labor on-site back to a factory-controlled environment to drive better quality, to ease commissioning, to deploy and test as systems before it ever gets to a site, to dramatically compress what can be very chaotic and somewhat uncertain deployment schedules in a traditional construction environment. We make that a manufacturing environment through converged infrastructure. We have levers to pull here that help all four of these things go faster. The key piece to unlocking and enabling these levers is the foundation of a broad product portfolio. Having all of the pieces, being able to see this end to end, and be understanding how all of those systems interconnect is a key piece to unlocking and understanding where some of these challenges actually lie.
Normally, this would be where I'd flash the broad Vertiv product portfolio slide. We'd look at it in a static fashion. What I want to do is in a digital fashion, in an animated fashion, take you through what a Vertiv OneCore data center looks like and walk the powertrain, the thermal chain, and give a sense of the visibility of the site. Let's dive in. We're looking at, call this a 12-50 MW data center. We have the powertrain, the thermal chain, the IT white space, and we wrap this with services. When we dive into power, talking about a battery energy storage system, the incoming MV switchgear to a low-voltage switchgear at an MV power board or a low-voltage power board.
We go through the gray space into a large power converter UPS device with lithium-ion batteries into an auxiliary UPS that's typically supporting CDUs through a power distribution or an STS type of a switching cabinet from the gray space into the white space, through our busbar, into our tap off boxes, and down into what is a high-density top-of-rack distribution device. That gets AC into the rack, DC power busbars from a rack power system and power shelf into the chip itself. From the chip, we're going through liquid cooling manifolds in the side of the rack, up into the secondary fluid network through valves, down into our row-based CDU. That CDU provides liquid cooling back to the chiller.
That chiller loop feeds air handling units that are providing air to the cold aisle, through the hot aisle, back through that primary fluid loop, out to buffer tanks, and what is a Vertiv CoolLoop for Vertiv. That represents the power infrastructure and the end-to-end story of the equipment on a site. We talk about services and control. The control dashboard of the chiller to the CDU, to an EPMS approach to all of the switchgear and the power management, as well as things like optimization, NextPredict, predictive analytics for us to understand how all of that works, with the next lever of something like NextResponse to dispatch services and look at this in a holistic fashion. Where those systems abut is where you understand system design. It's where you understand how converged infrastructure can finally come together.
It's where, as we talked about yesterday, the idea that point products have an asymptote or a ceiling in terms of how much we can optimize and how much efficiency we can drive out of a point solution. There's a new horizon or a new frontier when we think of a data center, the white space, or a holistic facility as the product or as the system itself. We can think about efficiency, performance, tokens per watt, all of those levers, not as individual things to push on or exercise, but as an end-to-end process that can be optimized and solved for in a different fashion. It unlocks a hidden amount of efficiency, of performance, of improvement that otherwise remains hidden. How does Vertiv do this? We're looking at evolving the powertrain for AI scale loads. They are different. They require different thought, different thinking, different approach.
We're talking about transforming a data center into a truly grid-connected asset. Not something that is just a grid consumer, but something that interacts with, participates with, and ultimately supports both the grid and the load in a robust and dynamic fashion. We start talking about orchestrating and controlling power, cooling, and the IT workloads as one system. We get asked this question a lot: does it actually matter that you have all of the different pieces in the portfolio? The short answer is yes. Power affects GPU performance, thermal affects GPU performance. Visibility to power ultimately is a leading indicator of the thermal need of that GPU in the white space.
Small examples of how knowledge and telemetry and interconnectedness of these devices and these systems help us to unlock, better predict, better enable performance that remains hidden to folks that are approaching this in a one-off, discrete product type of a fashion. We talk about Vertiv SmartRun and we talk about Vertiv OneCore. I think that's the embodiment. That's the easy way to think about converged infrastructure because it is the whole site. It's a good example, we take that approach and we apply it subsystem, system to site level. We're uniquely positioned to be engaged with customers to help solve some of the problems that are not yet visible to the broader industry. We will walk this somewhat discreetly. I will dive into a power management story. There will be some discussion on architectures.
There will be some discussion on evolution and ultimately where this is going. The spoiler here is there's no one discrete path. There will be multiple flavors, there will be multiple instances, but we are seeing things that maybe allow us, somewhat agnostically, to navigate where we think the technology is going and where it should go, in a way in which, somewhat altruistically, we can help the industry navigate and solve for the problems in a way that doesn't force us to push one angle of technology. I think that's very important. Traditional power architecture. Sorry for the size. It's both oversimplified and maybe perhaps difficult to comprehend, but we go from utility to MV switchgear, to LV switchgear, to a traditional AC power UPS on the source side. That's gray space out to mechanical yard.
On the load side, we're into the white space, in kind of reference back to the video we just walked. Distribution into the white space. Current GPUs today are deployed with AC power fed in, conversion at the rack level to 50 volts DC, and that's how we get to operation at a GB200, GB300, choose your GPU type of a performance today. Three major disruptors, influences, challenges are impacting the way in which we're thinking about that traditional AC power architecture today. First, grid interaction is becoming critical. We're hearing about, in the news, restrictions on interconnect, long delays on interconnect timing, limited grid capacity in certain regions, and a significant increase in what I will call grid compliance requirements, low voltage ride through, how we handle faults, how a data center needs to stay connected, if and when there is an issue upstream or downstream.
The size of data centers and the criticality of data centers to the local grid is amplified and magnified tremendously. That has significant implications for how the data centers need to design and behave, and the performance that we need to enable for those assets to not just be effectively in an island mode, but can actually be active participants in the grid. We talked about the dynamic and synchronous loads. Those large-scale AI loads behave dynamically and synchronously, and when you have 5, 10, 50 megawatts behaving in that fashion, there are dramatic implications for what is potentially influenced and reflected back upstream to the grid. There's dramatic implication for impact to all of the infrastructure seeing that volatility or seeing that impact. There's potentially dramatic effect on the ultimate performance of the GPU if that volatility and that dynamic profile isn't managed appropriately. We're talking about throttling.
From a rack density perspective, we highlighted this. We're at 140 kW today on the frontier. We are quickly moving to 200 and 240. We see 600 on the horizon. We are talking about racks that are over 1 MW. Whether that's a single wide or a double wide, whether you choose your kind of profile for a rack, the short answer is we are seeing density that is now approaching what I will call the physical and physics limits of delivering power into a space and delivering that amount of connectors, busbars, copper, wire capacity into a constrained physical space. That density is happening to unlock performance. That performance needs to happen to enable tokens per second. This isn't a "Oh, we're driving density because it's fun.
It's a unique challenge." We are driving the density to unlock that performance, and that has significant effect and implication for how the power system and the powertrain needs to be defined and designed. I want to take a step back. Why is Vertiv winning in AI deployments for power today? Those dynamic and synchronous loads, I don't want to say snuck up on the industry, but maybe wasn't well understood the dramatic effect that those dynamic and synchronous loads downstream have on performance, as well as have on the upstream infrastructure. Vertiv is solving for this challenge today in a relatively unique and robust way, something we call power smoothing. Our Trinergy UPS paired with lithium-ion batteries in our Vertiv EnergyCore system, paired together acting as a system with its overload capability and with some patented control logic between the two, can take that dynamic load profile.
It can ensure that the equipment doesn't need to be oversized in order to meet that dynamic nature, and it smooths what the grid sees and what the input or the upstream input power sees in terms of that load profile. It does it differently, and it enables, with that UPS intelligence and that capability, the infrastructure to operate in a much more stable and much more coordinated and a much more effective way. It's subtle. It's not really well understood, but it's something that as we engage with our end customers, this is a significant problem that has existed in terms of enabling the optimum performance of GPUs without massively disrupting infrastructure or without having significant impact to the grid participants and to the upstream utility operator.
We have gotten tremendous feedback from our customer set as we're engaging in factory witness tests and performance validation and things of that nature. What I want to plant the seed here is that intelligence and that control at the UPS level is going to be very critical as we navigate some of the changes and some of the architecture shifts as we dive in. The other piece of this that I think is more of a general trend that we should expect to see, energy storage will proliferate throughout the data center. Energy storage used to be thought of as, okay, a UPS needs 5 minutes of backup so that I can handle an outage so that I can transfer over to a generator and the result of something happening at the utility level. Energy storage now is acting as much more of a shock absorber for the system.
It's at the UPS level, but it's increasingly downstream at the rack level in order to handle some of the spikes and the fluctuations that happen with GPU performance. It is happening much more upstream, connected to the utility in the form of battery energy storage systems that allow and unlock and enable grid participation, taking excess load or giving it back so that you don't have to disconnect, using it as a mechanism to replace traditional on-site backup generators and the like. The important aspect of this is in order to understand and in order to solve for things like power smoothing and the dynamic nature of load profiles, those three elements of energy storage operate at different timescales. Very fast, rapid fluctuation on the right-hand side, very long duration, 4 hours plus on the BESS side, with the UPS in the middle.
The coordination across all three of those timescales and time domains is critically important to think about the powertrain as an entire system and unlocking some of that hidden level of performance capability when it comes to power smoothing, fault tolerance, grid interaction, and enablement, and also ensuring that the downstream load, the GPUs themselves, don't have to throttle because I can't achieve overload or because I can't achieve the performance profile we're talking about. We come back to the traditional AC power architecture. I think a lot of folks are wondering, where does this ultimately go? What is the influence? Okay, what we just talked about, a lot of grid interaction and other things that need to happen.
Our opinion, Vertiv's view of this is there's an intermediate step here in order to unlock all of those grid challenges, grid enablements, and tie that involves moving upstream to medium voltage. The intelligence, the performance, the capability of a UPS moves upstream, tying to the medium voltage ring for better enablement and a shift of battery energy storage to that medium voltage level, as opposed to tied in parallel at the utility level. It brings with it an integration of larger block sizes tied to medium voltage, integration of battery energy storage. We see much more on-site power generation, which is inherently more sensitive in terms of load ramp and decrease than a traditional connection to a utility.
The logical move is in order to unlock and enable the grid interactivity, is we see a lot of the capability of the traditional architecture, the traditional UPS and energy storage move upstream to the larger block sizes with an MV AC UPS architecture. What does this ultimately look like? We're talking about 3 discrete systems that are all pulled together into the MV space. Vertiv is launching a new product category. I will call this the MV BESS UPS. It is bringing together MV switchgear, battery energy storage, the intelligence and the capability of a UPS in a 3 to 4 MW block size. It allows us to scale more efficiently. It allows us to integrate those time domains across energy storage in a pretty dramatically improved fashion. It allows us to do this more intelligently.
Instead of 3 discrete systems and 3 discrete domains, we bring it together into 1 managed system. It unlocks a different level of control and capability, and it truly allows the UPS to act at a grid level to better manage grid fluctuation, voltage ride through, fault tolerance, islanding and disconnect mode, economic engagement, and dispatch to unlock and enable the data center to truly act as a grid connected and grid interactive asset. This won't be for everyone. This won't be kind of the universal architecture that everyone rolls to, but this solves a significant challenge that the industry is facing in terms of how to make sure that data centers and especially large loads become better grid citizens and unlock a lot of the performance and capability we have been talking about in terms of more megawatts deployed to GPUs based on an architecture.
We now have two kind of predominant source level AC architectures. You have your traditional with a UPS, which I will remind, does not go away. I don't believe this goes away. There will be a long tail and many customer types that will depend on this architecture and will rely on this architecture. For those that need the higher large load level and grid interactivity, we move to this MV architecture. The key piece of this is both of those architectures feed low voltage AC into the data center and both support the current load profile, white space profile of what needs to be deployed for AI data centers today. Based on the questions I received last night, I think this is probably the flip that everyone has been waiting for, is the what happens at the load side now.
This is a story of DC power in the rack moving to DC power in the pod. We talked about the capacity and the density. We are approaching physical limits now. Once you achieve or arrive at about 350-400 kilowatts in a rack level, connector sizes, busbars, interconnect, and the physical amount of power that needs to be deployed and delivered over the copper, we run out of space. Further, we have been deploying rack level power systems and power shelves that are taking up valuable use space in that rack. As we move to racks that are going from 72 GPUs in them to 576 taking up the entirety of that rack space, we physically don't have room for power in that shelf. Two things happen.
Those racks move out, the power shelves move out of the rack into a more standalone power center, and the voltage increases. We have aligned around 800 volt DC as well as plus minus 400 volt DC to enable and unlock that higher level of voltage that will be tied to the IT to enable this densification to happen. The beauty of this approach is it allows all of the upstream AC infrastructure that we just talked about as vital and critical and evolving and changing to still be fed into the IT. It keeps this power discussion, which is inherently new, unique, requires discussion around compliance and serviceability and capability in a robust way. It keeps that power discussion contained to the IT.
In much the same way a power shelf is delivered with a current GPU generation rack, this sidecar approach or this power center approach enables the same thing. The power can be bundled with the IT rack itself up to a megawatt. From a Vertiv perspective, maybe visually this helps. The power moves out of the rack, which can be 8 to 12, up to 16U of a 42U rack space, and frees that up for GPUs. It moves to an 800-volt power center that helps from 400 up to 900 kilowatts, and it ties directly to interconnect busbar throughput to the GPU rack itself. We free the space, we reduce the conversion, we enable this in a better fashion. I think what is different for Vertiv is we've been in development for a long time.
The picture on the right is a real lab-level validation that we have going on right now with a customer with their intended GPU rack for this level of architecture that is going through not only customer validation but the entire ecosystem validation for conversions, IT, the power distribution, and treating it as a bundled system. We are well within and well through engineering validation, and we are ready to take this into scaling. Product will be ready end of this year. Commercialization will happen in the beginning of 2027. From a timing standpoint, we expect a steady ramp throughout 2027 as we think about scaling. We think about the, I'll call it, the supply chain robustness and the build-out that needs to happen. A key point here is that this is really truly dependent on the IT.
This isn't about pushing a power architecture in order to land and be ready. This is a bundling and a very direct tie to the GPUs themselves that will be natively operating at this voltage. That's where it makes sense to drive the transition. From our readiness and enablement standpoint, we are trying to ensure that we are unlocking and enabling the power architecture so that this can land when those GPUs are ready. All right, now we have two AC power architectures. We have two load side architectures. I think, pardon the arrows and the interconnected and tangledness of this, but the key message here is that we have optionality and we have flexibility. Traditional AC architecture works with current GPUs, and through this sidecar approach with 800 volt DC power, traditional AC power architecture works with the next level of GPUs that are coming.
The MV AC architecture that we talked about for better grid enablement and grid tie doesn't preclude you from powering current generation GPUs, and it's perfectly well-suited delivering low voltage AC power into a rack that converts to 800 volt DC and unlocks that next level of densification. There is optionality here, and there will be many instances and many flavors of this depending on the customer type and the business model we're talking about. The next logical step here is we talk about native 800 volt IT and silicon. The next step to this as we talk about centralizing and moving to a more centralized DC power architecture, is one in which the sidecar is no longer tied to the IT, but it comes across and extends across the entire facility.
We see multiple paths to this, and I think there's a lot of discussion and a lot of debate in the industry as to what's the right thing and how do I optimize for the maximum amount of efficiency? How do I reduce the number of conversion points in our entire power architecture? I will say for those that are looking for the maximum optimization, for those that are building AI factories that will have almost exclusively 800-volt native GPUs, which is not the entire market, is going to be, to a certain extent, a specialized or a longer-term type of an architecture for folks that are building purpose-built AI factories. We see a couple of steps towards unlocking and enabling this architecture. You can move a sidecar into a centralized pool of systems to enable more distribution rather than at a rack level, but at the pod level.
The more meaningful transition that we would drive towards is an MV connected centralized DC architecture. From our vantage point, there's two parallel paths in order to get to that centralized DC architecture. The first has a higher degree of technology readiness level. It has a higher degree of maturity on the supply chain, and that's what we would call an MV DC UPS. Really, it is a traditional transformer paired to a rectifier that unlocks and enables the MV to DC conversion and distribution at an 800 volt level throughout the site. The parallel path to this, and it's not binary or one or the other, is investigating and driving towards solid state transformer technology. There is a maturity and a technical readiness that still needs to come along with this technology, and I think it's exciting for all of the right reasons. It promises better efficiency.
It promises smaller footprint. It promises fewer reliability points, based on conversion as opposed to more static and manual mechanical devices. The truth of the matter is there's a lot of technical maturity and readiness from a supply chain perspective that has to come along. We always like to joke that transformers have had a 100-year head start on solid state conversion technology, and I think that is apropos in terms of the readiness and the maturity for this type of an architecture to be deployed in earnest. The important takeaway here is Vertiv is developing both paths. From a customer engagement perspective, what we are trying to do is unlock and enable the best TCO, the most reliable, the most effective solution to navigate these paths for our customer types because there are a lot of different needs we are moving towards.
This isn't a one size fits all solution, and this isn't a singular approach that we will take in the market. Now we have six architectures. The takeaway to the power story is, again, I know everyone wants this singular converged story. The world is not that simple. There are different business models. There are different archetypes. If we're talking neo cloud versus co-location, if we're talking hyperscale versus an on-prem enterprise, there will be different flavors and different needs based on what is important to our customers. The generations of GPUs that they will have over the life of their 20-year facility. The purpose, whether it is inference or training, whether it intends to evolve over time, will ultimately dictate which of these architectures make the most sense to navigate to.
The short answer for us is we want to unlock and enable all of these architectures in a way that allows us to altruistically and agnostically drive the right technology for our customers to solve the problems that are relevant to them, as opposed to pushing a singular approach or a one technology narrative. I think that's the importance and the power of the portfolio, and that highlights the way in which we engage with customers differently when we talk and operate and act at a system level. That was the run through power. I want to shift gears towards thermal, and there's a similar story and a similar dialogue as we think about architectures and we think about evolution. I think there's a lot of maybe misunderstood or oversimplified views of the world, an extremely simplified view of what I would call the modern thermal chain.
We have liquid cooling that happens at the AI loop level back to a CDU through the technical water loop or the secondary fluid network loop. We have both liquid cooling and air cooling that exist together in harmony. Both of those devices are connected on a primary loop out to heat rejection. Heat rejection, heat collection, heat collection. There are 5 pretty significant challenges, disruptions, issues facing thermal designers and data center operators today. First and foremost, liquid cooling. I think we get this, but liquid cooling is mission-critical. I always like to compare a data center 5 years ago that was 100% air-cooled. We never even used to put thermal equipment on UPS backup. There was enough thermal inertia in a facility for us to be able to ride through to switch over to gens.
Liquid cooling has 1 to 2 seconds of thermal inertia to where if you are out of range or you're out of tolerance, you are throttling a GPU or you are shutting down. It is an incredibly higher degree of mission criticality on liquid cooling compared to air cooling. We are driving towards higher operating temperatures. I think there's been a lot in the news around this push towards 45 degrees Celsius to the GPUs and why we're doing that. We'll unpack that a little bit, but the short answer is we are driving to higher fluid temperatures to reduce the need for mechanical cooling, enable more free cooling for energy efficiency and performance, but most importantly, that reduction in mechanical cooling frees up some of those trapped megawatts from the site to deploy more power to the GPUs. We'll talk about how we're enabling that.
Air and liquid will coexist and will intermingle for the foreseeable future. I know Gio already claimed air cooling was dead yesterday. Long live air cooling. We are in an environment now where it's probably 80% liquid, 20% air on the latest gen, highest performing GPU and AI environments. Even in the most aggressive projections, that moves to 90% 10 or 95% 5. When you do the math on 10% air cooling on a gigawatt campus, that's still a significant amount of air cooling that needs to be thought about and needs to be considered. You pair that together with the temperature discussion we just had on rising temperatures to the liquid that can't necessarily be achieved or delivered when you have to consider air cooling as well. That creates complexity and further design challenges for us to consider and articulate in the data center environment.
The other piece of thermal is that often gets lost in the discussion is it's location dependent. Your ambient conditions, your local weather, your local temperatures on where that site is going to be deployed, and whether that's Phoenix or the UAE or Jakarta or Columbus, Ohio, those are different performance requirements, different design requirements that need to be thought about and considered. Again, there is no one size fits all. There's a tremendous amount of complexity that comes into the heat rejection and the cooling story. You wrap that all up with our discussions now with customers. Turning up, starting up, and driving scope and thermal services is a critical part of that discussion.
You can't turn up this amount of liquid cooling without the service enablement, without the ability to start clean and stay clean, without the ability to get your loops right, without the ability to actually scale and mass for some of these sites. It becomes a critical part of the story on how not only will you support test turn up and commission, but how you will actually enable the life cycle of that offering. I'll start with liquid cooling. We can talk about our CDUs, we can talk about the ability to have the secondary fluid network, but from a Vertiv perspective, I simply want to highlight that this isn't a discrete product. This needs to be thought about, managed, controlled, and operated as an entire system.
The CDU has incredible performance around flow control, leak detection, performance parameters, operating set points, filtration, and fluid chemistry that all inherently matter to the performance of that system. We often talk about a human hair is about 100 microns, and we're talking about filtration in these fluid networks that is at 25-50 microns. Particles matter. The level of cleanliness, the level of efficiency and performance of your GPU system depends on the criticality of that loop. A CDU is pumps, pipes, and heat exchangers, but it is far more when you actually think about the system and how it ties together with the entire secondary fluid network. It moves into the chip and the server cooling loop.
Our acquisition of Strategic Thermal Labs is a critical enabler for us in terms of visibility into that cooling loop at the server level to understand where the performance is going, to understand what the requirements and the needs are, and how that connects back to the full system. Absolutely mission-critical. We pair that with PurgeRite and our Vertiv service capability around the liquid cooling and the technical loop to be able to piece those things together so that we start, we flush, we fill, we operate, we maintain the fluid in a consistent and a robust manner throughout the entire life cycle. You pair that together with our Vertiv Unify platform, something we internally call our intelligent secondary fluid network, to manage valve control and buffer tanks and discrete flow to individual racks.
You have to think about this in terms of it is a system that is operating. No 1 CDU, no 1 intelligent valve, no 1 piece of equipment throughout this chain can do that unless you are thinking about this and controlling and managing as a holistic loop. The takeaway here is this is mission-critical work now. The criticality and the complexity is increasing. Vertiv is extremely well-positioned to continue to evolve and innovate in this space around all of those pieces coming together. This is the fluid temperature story. Again, why are we pushing towards higher GPU operating temperatures? When we talk about the efficiency gains and the performance we can unlock, when I consider a traditional chiller loop, for every degree Celsius of temperature I increase that fluid loop, rough numbers, I get about a 3% efficiency gain.
That's through less mechanical cooling, more free cooling, more enablement of not having to run compressors or the mechanical load of a thermal device. 1 degrees Celsius gets me 3%. If I'm operating at 24, 27, 30 degrees Celsius on a primary chiller loop today and I can push that to 45, that is a significant energy savings and a significant peak power reallocation from a data center site to the IT for better efficiency and unlocking some of those trapped MW we've talked about in terms of the tokens per watt and tokens per second. Naturally, the industry says, "Okay, well, we're moving away from mechanical cooling. Simple story. Everything's going to be free cooling and dry coolers and we're going to enable that performance and we're going to unlock that capability for the industry." It's not so fast.
There's a lot of challenges and other considerations that come into thermal design and site design that needs to practically be understood, unpacked, and unraveled. Your site type, your footprint, your air-to-liquid mix, the ambient temperature, the local temperature that we are operating in, the GPU type, and the fluid temperature it can maintain, customer preference, risk profile, all of these things factor into how do we unlock and think about the heat rejection strategy for a data center site given all of those different variables. A dry cooler is an incredibly effective, efficient, by nature, free cooling approach to heat rejection. The problem is it doesn't work in every environment, and it works in a surprisingly few number of environments based on ASHRAE design locations. How do we handle this?
If I'm at a 20-25 degrees Celsius ambient, hug the left-hand side of this chart, and I want to move my server fluid temperature up, I can move all the way up to 45 degrees Celsius and I'm operating in dry cooler land. The second that ambient temperature increases and Minneapolis, Ohio, northern states that you would think are kind of high free cooling zones all fall into the 35-40 degrees Celsius operating range on this chart. We end up in a dead zone on this chart. The gray hashed area is an area where modern chillers can't operate and perform at those elevated fluid temperatures that we would like to move to, and a dry cooler doesn't have the capability to be able to operate in all of those ambient zones. What do we do?
Vertiv has pioneered what I would call a new product category that effectively blends together the capability of a dry cooler with the performance, the assurance, the, I'll call it the backstop of a chiller and mechanical cooling together into a single package that fills in that gray space, that allows us to push to 45 degrees Celsius operating temperatures to unlock and enable the efficiency of those GPUs while still operating or enabling, I'll call it chilled water mode or mechanical cooling mode, for those days, weeks, months out of the year where we need that backstop. It also enables different flexibility and different performance parameters for us to think about and articulate the entire thermal design of the site differently. We call it the Vertiv CoolLoop Trim Cooler. The best of a dry cooler, the best of a chiller blended together.
What that does for us is it combines dry cooler efficiency. You can use that efficiency for increased capacity. You can use that efficiency for reduced footprint. You can deliver that increased capacity as peak power gains. Uses zero water and it is unlocking and enabling almost 134% more free cooling hours compared to standard designs that we have based around chilled water today. This is a game changer that allows us some flexibility and some better articulation as we go forward. This will be a lot to take in, but I want to give a practical example to how having all of the pieces, having flexibility between chilled water and dry coolers and trim coolers, having flexibility between air and liquid, and having a breadth of portfolio allows Vertiv to think about thermal challenges differently. We have what I would call the two-temperature problem now.
GPUs, I want to push to higher water temperatures, 41 degrees C, 45 degrees C to unlock that efficiency and performance gain. I still have air cooling on the site that can't push past 30 degrees C without rendering it ineffective or inoperable. How do I manage those things? I have an inherent trade-off. I have an inherent optimization discussion I have to make. I can solve that problem as Vertiv by moving to two thermal loops. I have a chiller for my air cooling tied to an air cooling loop that operates at one temperature. I have a trim cooler or a dry cooler that operates for my liquid cooling loop.
I've now prevented myself from having to make a difficult temperature trade-off and have gained some efficiency, but I've also introduced a significant amount of complexity in terms of dual loops and dual piping. The right-hand side, single loop architecture that I can drive with a trim cooler that enables that higher water temperature through innovation on the air cooling side by enabling water-cooled DX technology or trim capability within our air cooling portfolio. I can do that mechanical cooling just where I need it within the air cooling portfolio, run a single loop for improved complexity, have the higher elevated water temperatures that my GPUs want, while giving and enabling the proper air cooling and air temperature for the remainder of the air that is needed.
Having the breadth of that portfolio can help us unlock 10%, 20%, or more of efficiency for the same design compared to others that are pushing a single technology or a single approach to trying to solve these problems. Having the broad pieces matters tremendously. When you actually stack up what our thermal portfolio looks like, those pieces become apparent. We are not pushing a technology. We are agnostically looking at a broad holistic portfolio to help unlock those technology challenges differently. It is air cooling. It is liquid cooling. It is single-phase direct to chip. It is two-phase, ultimately, potentially where these things need to go. It is unlocking and enabling room to row at an air cooling level.
From a heat rejection perspective, you need all of these tools in the toolbox in order to unlock and solve those challenges in the way that we just talked about. We heard yesterday from Chris Crosby, the innovation and the approach to intermingling and interconnecting hybrid air and liquid systems gives us a different palette to work from as we think about solving for the complexity challenges that are going to be facing data centers. It requires multi-technology. It requires a broad portfolio and discipline. I'm extremely excited about the opportunity we have in front of us to engage in solving these problems differently. I'll keep going. From an IT systems and an infrastructure solutions perspective, let's dive into the white space. How we think about white space and infrastructure in this day and age is fundamentally changing.
Piece parts that would normally be assembled on site between busway and circuit protection, rack level power, distribution architectures, cable drops, thermal management, secondary fluid networks. We are now thinking holistically as a system. We introduce SmartRun. SmartRun is designed to work together. It is designed with the end GPU load or TPU load profile in mind, and it is designed to have a future-ready aspect to rack positions, power levels, flow rates, to understand where the actual cluster and pod size is going, and how we are best suited to put all of those pieces together. When we go deeper into the white space, we've robustly built out our rack portfolio and rack capability.
We are solving new and unique problems on density at the rack power distribution level and top of rack distribution level that enables and unlocks greater levels of density, whereas normally we'd hit physics-based constraints and challenges. We build out and we are reinforcing our rack level power portfolio, both on the 50 volts DC side, as well as on the 800 to 50 volts DC side. We enable row and rack-based cooling, rear door heat exchangers, in-rack CDUs, other capability that makes us a robust and compelling provider to system integrators, rack integrators, OEMs, and we bring that all together into integrated rack solutions that enable enterprise-level on-prem deployments for folks like Vertiv that are looking at and evaluating, how do I enable AI on site? How do I enable an AI and a hybrid cloud-based environment and deployment? It's those solutions that unlock and enable that capability.
For those that have the benefit of going to the tour, you will see some flavors of SmartRun. The primary GB200 architecture that we deployed in earnest and are deploying now at volume is a 1.2-1.5 MW 40-foot SmartRun that supports 115-140 kW in a rack space on an average level. We start to move through what that SmartRun looks like as a product offering, and we evaluate and we elevate from 4-6 MW of contained infrastructure that enables us to be future-ready and, to a certain extent, GPU and TPU agnostic to tailor these designs and release them as products regardless of the chip architecture, regardless of the chip type that we want to deploy into these spaces.
Enables an incredible amount of flexibility, responsiveness and thoroughness as we think about unlocking and enabling this multiple GPU generation ahead type of a discussion and an architecture. We bring this all together at a site level. We've seen multiple iterations of Vertiv OneCore. The beauty of Vertiv OneCore from our perspective is it's designed as a system. It can be tailored to the right application, we are releasing this and maintaining this as a product. We have a 10-MW Vertiv OneCore that scales to 50 MW, 125-MW Vertiv OneCore that scales to 250 MW, and a 250-MW building that I can scale out in a repeatable block size, building block, chunked style and fashion out to 1 GW. Dramatically compresses the time, significantly reinforces the throughput, the total cost of ownership and the enablement that we've talked about over the last 2 days.
This is truly converged infrastructure at scale. It unlocks a different level of performance and capability for the industry that checks the box and enables optimization across tokens per second, tokens per watt, tokens per dollar, and critically, time to first token. We talked a little bit about the wrapper that we are putting around this. I've gotten a lot of questions around our software, our digital enablement strategy. Vertiv Unify is a layer of controls that allows us to go from unit-level firmware on a CDU or a UPS to subsystems, an intelligent secondary fluid management capability, a chilled water manager that allows me to look at the entire control architecture and optimization between chiller to CDU to air handler to chip-level cooling.
We can do this in a way that allows us to better maintain performance thresholds and performance tolerances to incorporate the power view into a dashboard. We are trying to enable and unlock a better way of driving system control on things that are inherently not typically treated as systems. It's a critical and vitally important unlock for Vertiv that we are now pushing and enabling our customers to operate and actually engage with our systems as systems. That is up to and including Vertiv Unify at a Vertiv OneCore level. I'd be remiss if I didn't maybe bring this home with a discussion on services. The critical point, complexity, and all of these architecture changes are driving further importance to our service portfolio.
All of those changes we just talked about, medium voltage, 800-volt DC, mission criticality of the liquid cooling loop, the vital nature of starting with clean fluid and maintaining fluid management over the life cycle of the product, all of those things make the world more complex. They put more importance onto our service capability. They put more importance onto us getting the technology right for our service and scaling it in a dramatically important fashion. All of that makes our service content go up, and it makes that story resonate as vitally important with our customers as we go.
We drive service, and we focus on service as part of the product development and as part of the system solution, not only in terms of service enablement but customer enablement in terms of how that capability and that competence needs to scale and be there to unlock the level at which we're trying to deploy data centers today. From an engineering perspective, we have made tremendous investment in expanding the portfolio, but also the competence, the capability, and the ability to test. Testing a medium voltage UPS at 4-megawatt level is a different level of lab validation. It's a different level of lab capability. We are driving significant investments across the globe throughout our thermal, our power labs, our system integration capability.
One example I want to highlight, we are months away now from turning on what I would believe to be the world's first infrastructure AI system validation lab. A 1.5 MW AI pod with up to 600 kW TTVs or Thermal Test Vehicles for us to simulate and drive at a system level the entirety of how an AI pod performs in a real-world environment. For us to be able to drop in equipment and understand performance profiles.
For us to be able to take and understand application-level failure mode analysis and technology behavior and performance profiles, not in a live environment, but in a controlled lab environment, and allow us to test and benchmark against how things operate in the real world, a critical capability and a critical enabler that we are driving into the future and will become part of our normal course of business as we think about systems design and scaling. From an investment profile, we're accelerating. We have a great deal of pride in our end-to-end portfolio, in our competence and capability across our engineering and technology.
We are making decided and deliberate investments to ensure that our end-to-end portfolio stays end to end throughout these architecture changes, that we are pushing the frontiers of advanced technology, Horizon Two and Horizon Three efforts associated with where the industry is going and where it ultimately needs to go. We have a 20% annual increase in our plan around engineering in order to make sure that we take that lead and we stay ahead. We pair that with the ability to incorporate AI, productivity, improvements, throughput, and efficiency into how we develop and how we go. The key piece is time to market.
It's not about being alone necessarily, but it is about being ahead and driving our time to market so that our product development cycles and our portfolio is inside of the development cycle of the rapid pace that GPUs are evolving is critical to and core to our strategy, and it's how we continue to stay ahead. I'll bring this all back by saying we have a tremendous amount of work and effort and build-out and scaling happening today. We intend to accelerate our development. We intend to accelerate our investment. We want to lead the industry. We want to shape the industry with where the architecture, the innovation, and the evolution of data centers need to go.
We're going to extend and enhance the industry's broadest portfolio, and we're going to do this robustly through systems-level design, engaging with customers differently, and thinking about the end-to-end system and where we can unlock performance for this entire industry in new, unique, and differentiated ways. Incredibly excited to be in this space, incredibly excited to be in charge of leading the direction of this portfolio, and appreciate you guys giving me the time to walk through where we see the technology ultimately evolving to and going. Thank you.
All right. So much exciting innovation underway. I know there'll be a lot of questions. We'll open it up for Q&A, so please raise your hand.
Good morning, Scott. Good morning. Are there launch customers for your new products that you're rolling out?
Launch customers.
Pilot, launch customers.
Yes. Not that I would be at liberty to disclose, but the vast majority of what we just talked about are specific or tied customer engagements that we are designing around, building around, and preparing for launch.
Are they co-developed then, or?
Co-developed is probably a strong word in terms of there's co-development and engagement on specifications, requirements, understanding the performance profiles, and things of that nature. Typically, in these fashions, though, we go to great lengths to ensure that innovation within solving those problems remains core to Vertiv.
Okay, I gotcha. Thank you.
This guy over here. Hey, good morning.
Good morning.
2 questions. First, what is your thought process around cold plate? You did the Strategic Labs deal. I think that's more of a technology deal, but do you see a path to manufacturing there? Is that even interesting? I was wondering if also if you could just maybe address the broader strategic landscape. The nature of my question is, the message here was integrated solutions, kind of soup to nuts. There's a lot of companies out there that are sort of like the dog that caught the bus, right? They have a product, and they're on this growth curve. Do you see evidence of any sort of shakeout yet in that sort of mixture of companies? I could think of the 50 some odd companies that say they have CDUs or have a picture on a website anyhow.
Just how the landscape is playing out in your view?
Okay. Appreciate that. In two parts, I guess first on the cold plates, certainly the competence, the capability, the technical wherewithal, and really the, I'll call it the customer presence of Strategic Thermal Labs is really what attracted us to the company, and visibility to design profiles, customer access, understanding the nature and the mechanics of the server cooling loops themselves is paramount. Having it in the portfolio and scaling it are probably two different discussions. From our vantage point, we certainly are starting with the, I'll call it the system level understanding and engagement, understanding how the cold plates fit in with the broader thermal design, as I talked about with the liquid loop.
Certainly, as we look at specific customer engagements, in order to actually be relevant and have presence and engagement with customers, we have to look at this maybe discreetly about how you actually scale to a level of manufacturing capability that allows that engagement. That's maybe the way in which we're thinking about that. The other part of your question was more along the lines of are we seeing a shakeout? The way in which I would frame that is, I'll go back to the mission criticality of what we've been talking about, and I'll go back to as we are thinking about systems, services, the ability to scale, the ability to ramp is almost equally important to that, I'll call it program level engagement and discussion.
I do think we are seeing more and more specific questions around how do you intend to scale, how do you operate, which I don't want to infer about what other companies' engagements with customers might look like or might be, but I know that's core to the discussions we have with customers. Give me a credible story as to how do you ramp, how do you scale, how do you deliver at mass, how do you test and turn up and commission and all of these things? While the innovation is still probably the leading piece of that story, the ability to scale and really support that infrastructure is probably equally as important.
I do think that is having an effect on the industry that folks that can scale are going to be the first look or maybe more of the trusted partner as we think about how this ecosystem builds.
Thanks so much. As you described on the electrical side, you described this more traditional architecture, medium voltage DC, and relies on more existing technology, UPS, and then you also described solid state path. As you think about your customers, is there a preference as you have discussions? If you put them in buckets, hyperscalers, colos, neo clouds, is there a preference, at least marginally, for one path or another? I guess the gist of the question, do your hyperscaler customers versus neo clouds, is there a fundamental difference in how they think about their paths, technology path over the next 2 to 4 years? Thank you.
I think a lot of that is still emerging. I will say depending on the customer's business model, depending on what they're trying to accomplish, dramatically influences the, I'll say the optimization path that they want to go to. Like as an example, a wholesale data center builder, a co-location provider that may be building on behalf of a hyperscaler. They may be building a gigawatt campus in 100 MW or 250 MW buildings in chunks where They don't necessarily know the GPU, the end customer tenant, the person or the company that will be taking up that capacity. That has a dramatically different impact on the type of flexibility and optionality they would want to build into versus somebody that is driving towards a purpose-built, latest generation, AI only, large scale cluster training facility.
I think maybe that nuance of business model and uncertainty and flexibility is underappreciated in how a lot of folks are looking at the market and the breadth of solutions and the kind of the paths that these things will go down. Undoubtedly, some of the folks that are building at large scale will want to ultimately move towards that centralized, truly optimized, maximum efficiency type of an architecture, but I don't believe it will be everyone.
Thanks a lot, Julian Mitchell. Maybe 2 questions. First off, when you think about dollar per megawatt between power versus cooling, and you look at Vertiv's portfolio today, which one do you think offers the higher growth rate in the coming, say, 5 years, in terms of where your dollar per megawatt can grow most quickly? Secondly, I wanted to understand in cooling, more specifically, what do you think happens to the chiller? I think you made some comments around maybe only 10% of new data centers could be maybe the content would be air-cooled in a few years, but I think maybe I misunderstood, so maybe just clarify that, please. Thank you.
Sure. Appreciate that. I'll point back to maybe the way Gio answered the mix question yesterday in terms of power versus thermal, I think is a good proxy based on kind of the Vertiv at a glance profile we gave yesterday. In terms of growth rates, I think the architecture evolution I just walked through shows that there is more and there is growth in both of those buckets, especially as these architectures change. I think that's maybe the critical takeaway here is we are talking about incremental and we're talking about expansion, even as some of these architectures end up moving what I would consider traditional product. I'm sorry, what was the second part of your question?
Just around in cooling specifically.
The air versus liquid. Maybe let me back up. What we're talking about is when you look at the entire site or you look at a data center pod, we're moving towards more and more liquid cooling, but the GPU rack itself is going to be 100% liquid-cooled at some point. There are networking racks and support racks and other infrastructure and power equipment and other pieces of gear in that pod that still require and need air cooling. What I was talking about was the ratio of the amount of equipment or the total power that is liquid-cooled versus air-cooled. The trajectory for highest density AI sites is more and more is moving towards liquid cooling and less of that proportion is air cooling.
The point I was making is I don't think we see a world in which that air cooling completely goes away in the foreseeable future in this planning cycle and this time horizon. Especially then as sites are getting bigger, even as that proportion of air to liquid reduces, air is still growing. I think that's maybe a critical takeaway. It further increases the complexity of these sites and the way in which thermal needs to be designed around it.
I think there was another part to the question. The last part.
Please
was about chillers. We should not necessarily associate chillers to air or liquid.
I think it's more complex than that. What we see in the future, Scott can, of course, be way more eloquent about that. It is the entire kind of a spectrum of technologies in the heat rejection part of the portfolio to be alive and well for eternity, dare I say.
Yeah.
It will be a combination of the various parts.
I think that's a critical piece of it. Certainly, the performance and the capability of what I would call a traditional chiller, even within the Vertiv portfolio, needs to expand and evolve to meet and address these kind of new requirements.
Hey, Andrew Kaplowitz at Citigroup. Just on the liquid cooling side, obviously, there's been explosive growth over the last few years, but it seems like only lately you guys have made some bigger deals. We talked a little bit about CDUs and what have you, PurgeRite comes up a lot. Is there like sort of an unlock that you've had with some of these deals lately that sort of accelerated your performance? Revenue synergies are a hard thing to sort of calculate, how do you sort of think about that as you go forward?
Yeah, I think maybe the unlock is more of we are moving into a stage of deployment in the industry where we tend to talk about leased capacity and order intake and other things. For the GB200s, GB300s, AMD TPUs, like we're moving into a phase now where we're into deployment. That deployment is illuminating a lot of the inherent challenges of scaling, of flush and fill, of commissioning at this scale, of managing fluid and turning up hundreds of CDUs on a site. I think there's almost an unlock and an awareness of where some of the gaps are in capability within the industry that has enabled us to have more of a heightened level of like understanding of the relevance and the importance that a team and a capability like PurgeRite needs.
It's also highlighted, I kind of talked about it previously, like the ability to scale those things matters tremendously. Figuring out how to do the fluid management, the flush and fill on the primary side and the secondary side at a 250-megawatt site is not a walk in the park. It's a critical enablement and a critical skill that is needed holistically across the data center. Maybe the level of importance of that is being highlighted by, we're in the deployment stage now, and we're seeing that really come to fruition.
All right. We'll take this as the last question.
Oh, boy. Thanks, Scott. That was a great presentation.
No pressure.
To Julian's earlier question, one of the high growth stories here seems to be energy storage. It's now going everywhere. It's going to the rack, you've got the BESS. The question is really, how should we think about your differentiated value proposition in energy storage? How do we think about your priorities for development of the product suite? Clearly there's more content there.
How do you create products that have the margin profile that would be consistent with your targets when the lithium-ion cell, that's not something you make, right?
Correct
help us understand the priorities here.
Yeah, appreciate that question. I think certainly the system story applies here en masse. The capability that we are bringing, I guess it's managed as discrete systems today. As I think about the battery energy storage market, as I think about that's more connected in parallel at the utility level, maybe the unlock that needs to happen with all of these discrete energy storage pieces is it needs to be brought together into an ability to actually properly manage data center requirements. I think that's maybe an underappreciated lift in terms of understanding the capability, understanding how loads need to be managed, the interconnection, orchestration, and control of those devices across all three of those time domains, as we talked about.
There is truly a system-level orchestration and capability to where if you have one of those piece parts, you don't necessarily see how it needs to interact and behave across the board. Yes, it's more integrated or third-party content as we think about cells not being part of our portfolio, but in much the way we have interacted and engaged with energy storage at scale, throughout our history, the critical capability and unlock of that system is the control logic, the battery algorithms, the orchestration layer. That's where the intelligence that's inherent to a UPS today, I think applies to the entire powertrain, which gives us a pretty strong confidence in terms of our trajectory, our profile, our content as this evolves.
All right.
Cool.
Thank you.
Thanks, Lynn.
Appreciate all the great questions. It's certainly an exciting time to be part of the industry and an exciting conversation. I'm going to turn it over to Giordano Albertazzi for closing remarks.
Well, thank you, Scott. Thanks everyone for the two days, and I'll wrap it up very quickly. We're shaping the industry. We're the thought leader in the industry, and you've just seen thought leadership in action. We are accelerating. We are accelerating. We're focused on creating value for our customers. We are focused on creating value for our shareholders. We believe that the market is strong and will be strong long term. Our portfolio, our product, our system, our converged infrastructure, our services are augmenting that market opportunity, and we're invested in capacity. We are investing in R&D, innovation, inorganic growth when needed. As we have the strength and the balance sheet, we'll stay focused on accurate execution. We'll execute diligently everything we do, and we have strong confidence in our long-term financial plan that we have shared with all of you.
With that, thank you very much, and thanks for being with us.