Okay, I think we are getting started. Welcome, everyone. Good morning, good afternoon, good evening, wherever you are. If you're curious about fuel cells, you've come to the right place. And if you're new to fuel cells or know quite a lot about it, it doesn't matter; we have something for everyone. Thanks for joining us today. My name is Katja Gagen. I run communications at Bloom and will be your host today. And we'll be joined by my colleague, Akhil Batheja, in a minute, who brings about a decade of experience in fuel cells and has some exciting content to share. But we also have great news today since we announced a partnership with Oracle to power the AI data centers with our fuel cell technology. Check out our website, Newswires. You'll hear more about this.
But as we know, data centers are power-hungry, and they're looking for solutions to solve their power challenges. And here's why Oracle and many other companies choose Bloom and our fuel cell technology. One, it's speed. It's time to power, which is critical for these data centers. And as you can see in our Oracle release, we're saying that we are powering an entire data center within 90 days. But that's not all. We're also ultra-reliable. Our fuel cells deliver 99% uptime with uninterrupted power, and they can follow the critical workloads and the demanding AI workloads with fuel cells in a very, very reliable manner. And lastly, we deliver clean power with fuel cells, with virtually no air pollution and no use of water. So stay tuned for more details about fuel cells. And with that, I'm going to hand it over to my colleague, Akhil.
We're doing a Q&A session at the end of the presentation, so hold your questions, put them in the chat, in the Q&A chat, and we'll address as many as we can. And if we don't, we'll follow up with you as well. And now, without further ado, over to you, Akhil.
Thank you, Katja. Hi, everyone. Just a little bit of color, a little more color on my background. As Katja said, I've been in Bloom for about 10 years now, and I've got to wear a number of different hats over my time here, from sales to international business development to product management. Most recently, I'm part of our technology strategy team, where I get to commercialize some of our new technologies. Throughout my time here, I've got to interact with the data center industry and learn about the various opportunities and challenges. I'm really excited to talk you through how our technology can help suit those and help power data centers. I thought I'd start by showing a picture of one of our larger installations. We've got a site here in Korea that's tens of MW and beyond.
I just wanted to show the scale and the ease of installation. It's a pretty scenic site, so I thought I would kick off there. Since this is a fuel cell 101, I thought I'd level set by some basic facts about fuel cells. I know people come in with various different levels of knowledge. First of all, we're a type of energy generation. I often hear, "Aren't you a battery?" No, a battery is a form of energy storage while we produce energy from a chemical fuel source, typically natural gas. We're also not a backup solution. That's another myth that I often hear. We are a form of primary power, always running, producing power for our customers. We produce uninterrupted power. In fact, our output power is of the highest quality, similar to what you get out of a UPS system.
Fuel cells are often mistakenly thought to only run on hydrogen. It's true. It can run on hydrogen, but fuel cells are pretty fuel flexible. Most of our installations today run on pipeline natural gas. It's ultra-reliable, as Katja said. We've got some clever designs and modularity that allow us to go even beyond five nines of availability if our customers require it. Because we're a non-combustion way of making energy from natural gas, we also have no NOx, no SOx, no particulate matter, and we're highly efficient. That means less CO2 per unit of energy produced. Highly scalable. We could easily reach sizes well beyond hundreds of MW. You can see a picture on the right of another installation that's in the tens of MW, and they're like Lego blocks. You can place them on and on and put as many as you need to meet your power needs.
We're also now price-competitive compared to other solutions. This was not always the case. Five to ten years ago, we definitely had a—we were rightly perceived as being more expensive. Bloom has been on an aggressive cost-reduction journey, and as we're going to see later today, we are price-competitive with other forms of energy generation. So I often hear a lot of myths, and I put them on the screen just because it's things that people hear. I'm going to draw your attention to the last one. If they're so good, why aren't people using them? I think that's a really fair question. My response to that is, they are. I point to the 1.5 GW of fuel cells we have deployed across the world, across 1,200 sites. In fact, let me draw your attention to the left, where I've got the logo of AEP.
This was a landmark deal that was signed just a few months ago, where they've agreed to take up to 1 GW of power from us. Below that, I've got some of our data center customers, and we're well in excess of 400 MW today. In fact, Equinix, one of the logos there, has over 100 MW alone just in their data centers across the globe. On the right, I've got a small sample of some of the other customers who we serve. You can see it's a number of different industries. Broadly speaking, though, we saw four main problems for our customers. One, time to power. We're able to deliver our systems very quickly on site and get customers power as they need it. Businesses are growing quickly, power needs are urgent, and we're able to meet that demand. Second, reliable.
We can configure our system to be as reliable as the customer needs. We can be very smart and go all the way up to even beyond five nines. Third, cost of power. Very often, we're far more economic than other choices customers have. Finally, sustainability. Because of the non-combustion and the no-NOx, SOx, and particulate matter I mentioned before, we often help customers with their sustainability stories. Now we can talk a little bit about how a fuel cell works. On the top, I've got a little graphic of a small cell. This is the fundamental building block of our technology. I'm going to age myself with this reference a little bit. It's a side-hot view of a cell that's roughly the size of a floppy disk. It's very, very thin, a couple of hundred microns.
We take a natural gas on one side and oxygen from the air on the other one. Through a non-combustion process, we electrochemically convert the chemical energy in the natural gas to electricity. As you can see, a couple of key points here. One, this is a very high-temperature reaction, and that allows us to be extremely efficient. Two, it's a direct conversion. Other ways of producing energy from natural gas involve combusting the technology, converting that to heat, using that heat to move some sort of moving part, and that in turn creates energy. There's multiple steps along the way, while ours is direct. That too helps with the efficiency. On the bottom, you can see how we kind of take that cell and build it up into the large-scale solutions.
The same cells are copy-pasted into stacks, and that stack is our fundamental technology building block. There are no moving parts there, which really helps with reliability. We take multiple stacks and move it into power modules. The power modules scale into energy servers. Within an energy server, if a single power module needs to be repaired, needs to be looked at for whatever reason, the rest continue to operate. We have a very high availability because of that. That energy server can be copied and pasted over and over again to form the energy farm in the 100 MW+ solutions that I was talking about earlier. Now that we understand a little bit about fuel cells, how do they stack up with regards to data centers? Over my time working with you in the industry, a couple of things I've heard.
One, more and more data centers are exploring on-site power. You know, the utility cannot meet the need of data centers today. The demand for AI is growing ever so quickly, and data centers need power now. Because of that, data centers are often looking at on-site technologies. On the left, I've got a couple of different on-site technologies we're going to compare against. We've got large-scale gas turbines, we've got reciprocating engines, and fuel cells. Some of the attributes that I've often heard are important to the data center industry are reliability, right? You contract with complicated SLAs with your customers, and reliability is the most important. Cost. You can't have an over-expensive power. It's one of the biggest contributions to the operating cost of a data center. Time to power, as we said earlier, it's growing very quickly, and we need to meet the demand now. Load volume.
As data centers become more and more AI-driven, this is something we're hearing more and more, where GPUs compared to CPUs fluctuate in the power needs a lot more. This means the form of on-site generation technology needs to be able to ramp up and down to meet that need. Power density. Real estate's at the premium. Any square foot that's not used for GPUs and CPUs is lost revenue. The more dense your solution can be, the better it is for the data center. Of course, sustainability. Every company signed up for ambitious goals, and we need to make sure that we're tracking towards them. With that, let's compare these technologies across these different parameters. We'll start talking a little bit about reliability with fuel cells. With us, we're on the sixth generation of our technology.
A lot of investments got in, and as you can see, our systems are lasting longer and longer. That 2.3 million hours of. Fleet runtime is something we're extremely proud of, and there's been a lot of learning, and the technology has benefited greatly from that. In sites where we are operating in an islanded form, that is microgrids, we have the availability of 99.997%. That's an incredible statistic. Let's compare that with some of the other technologies. Now, with the next few slides, I'm going to dive in a little bit on how reliability works. There's two really inputs that go into how you can use energy generation to reach a reliable solution. One is the building block size, right? The larger the size, when it goes down, the larger percentage it is relative to the data center's total need. Second is how available that technology is.
When we look at a large-scale turbine, like a 50 MW unit, we need to add a third unit. If we think about a 100 MW data center, like a hypothetical data center, you need at least one extra turbine to make sure that you're meeting just 99.9% availability. The reason I'm taking 99.9% availability is that's roughly the availability of the grid, and that's what I've often heard from you as the ideal solution. You already have UPS and diesels, etc., on site in the data center, you just want your solution to meet the grid's level. 50% overbuild, that's quite a bit of extra CapEx, quite a bit of extra labor, quite a bit of extra land. With reciprocating engines, there's a range of them, and their size really fluctuates. The smaller the unit size, the lower the overbuild you need, right?
On the left, I've got a 3.3 MW building block, and on the right, a close to 20 MW building block. As you can see, that overbuild is really sensitive to both the availability percentage, like how much time the technology is available to produce power, and its size. Now let's compare that with fuel cells. Our energy service, as you saw earlier, is 325 kW. That availability is 99.7%, much higher than the other technology you saw. Because of that, we only need 9% overbuild to reach that 99.9% availability. This has huge implications for your site, for the economics of the project, and for various other factors. Let's look at some economics for a hypothetical 100 MW data center. On the left, I've got the total installed cost for the different generations. These numbers vary greatly in the real world. Every site's different.
You have soil conditions, you have seismic zones, etc., etc. As you can see, fuel cells, even with the ITC, are a touch more expensive on a per-kilowatt unit. However, when you build out that full 100 MW data center to reach that 99.9% availability, it's in fact much cheaper in the capital outlay. In this example, you can save up to $66 million simply because of that overbuild percentage. You can also look at the OpEx. Because we're far more efficient, we consume less natural gas to produce the same amount of energy. That means your gas bills are a lot cheaper. In this example, I've used a $5 MMBtu gas example running that 100 MW data center. You can see there's close to $15 million a year savings just from the gas. Let's take a step back and see how that looks on a five-year project.
As you can see, the total cost of ownership across the five years is 15%-25% lower. That's hugely significant. That doesn't even begin to take into account some of the other advantages that we were calling out. Load volume is being one of them. This is something I've often heard from data centers, and it's becoming more and more a conversation. I've got a pretty busy graph here, so let's take a step and look at it. This is actual test data that we've done on some chips that we were able to get running an actual LLM, specifically the Llama 2 from Meta. The green line is the server, like the GPU-level energy consumption. As you can see, it fluctuates pretty wildly. The orange line is the output of our fuel cell system. As you can see, it's able to keep up and match up naturally.
This is real-world data, something that we've actually tested on, and it's a proven technology. Footprint. We covered this a little bit, but footprint is becoming more and more important, especially in dense areas such as where I live in the Bay Area or in Korea, where we have a lot of sites. If you had all the land in the world, you could lay out our systems flat, and we'd take about 30 MW an acre. More and more, we're seeing customers ask us to go into these stacks. Our technology is built in a way that it's very easy to build them on top of each other. These skids that they mount on are factory-built. With two levels, we can reach 55 MW, and with four levels, we can reach 100 MW. 100 MW is some of the most power-dense you'll ever see.
On the right, I've got some other advantages of the technology. Because we're more efficient, we're consuming less natural gas, that means less CO2. Similarly, no NOx, no SOx, no particulate matter. These not only improve the air quality around the site, but also provide a huge advantage in permitting. Air permits are often some of the longest lead-time items in a data center project. In some jurisdictions, even here in California, which has the tightest air regulations, we're exempt from some of the air permits. Of course, this varies site to site and geography to geography. No water. If you think back to the chemistry that I showed you, our inputs are natural gas and oxygen. Once our system's running steady state, we do not consume water. This is hugely important, as energy generation can be extremely thirsty, especially in jurisdictions like Arizona and California. Zero noise.
I often host customers here at our headquarters, and we're powered by one of our fuel cell systems. When I'm down there showing the fuel cell, opening it up, customers are always amazed at how quiet it is. In fact, the freeway that's half a mile away is often louder than the system when you're standing right in front of it. Most importantly, time to power. We can deliver to site in 90 days. Katja talked about how this is part of the buy proposition for Oracle. This is something that Bloom's invested a lot in over the years to be able to get to. We have a gigawatt manufacturing facility here in California, and we do final assembly in Delaware in an equally scaled facility.
As you can see on the right, we've got some potential capacity from our factories, and Bloom's ready to meet this need in this moment. Finally, and what I get the most excited about this technology is how future-ready it is. There are a number of things we can do as the energy landscape evolves. You can take some of the waste heat that comes out of our system and boost efficiencies all the way up to 90% using CHP. That waste heat can be used for producing steam in industrial processes. It can be produced to heat up district air, or it can even be used to run a chiller and provide cooling for data centers. Carbon capture. Because we're non-combustion, our exhaust stream is a much higher concentration of CO2 compared to any other technology.
This means there's a lot less capital and a lot less energy required to purify that CO2 stream, to use it in some way or sequester it. DC power. We're a native DC source of power. As data centers more and more explore the idea of DC power, for us, it's just removing components as opposed to adding it. This allows the data center to be a lot more capital-efficient and a lot more energy-efficient. Finally, we're fuel-flexible. We can run on hydrogen, we can run on biogas, we can run on natural gas, and we can run on blends thereof. As hydrogen becomes commercially available, we're happy to run it through our systems. Last but not least, I'd just like to shout out to our data center report, the Mid-Year Pulse. This has some actual data from data centers, and a lot of work's gone into putting it together.
It's a great read. Highly recommend downloading it. I think we'll now open it up to questions.
That's absolutely right. We're seeing a lot of them coming in. Just a quick word before I go into the first questions. We are officially in a quiet period now, as some of you may know. Bloom has a press release out. We're announcing earnings on July 31st. Some of the questions we will not be able to provide, but dial into our earnings call next week and follow up afterwards so we can potentially address some of your questions. With that, there's a question about how many megawatts per month of fuel cell production can you provide? I don't think we're breaking it down that specifically, but maybe that's a question around capacity and production, Akhil, that you could answer probably in more general terms.
Yeah, sure.
Roughly speaking, we can do about a gigawatt a year. We use this to serve customers all around the world, between Europe, Asia, and here in the U.S.
That sounds great. Another question, how much CO2 per megawatt, again, do you generate with fuel cells compared to other solutions? I would say probably gas turbines with SIP engines, I would include here as well. Talk to us a bit more detail about this CO2.
Bloom's roughly about 800 pounds per MW hour of power produced. It depends a little bit on the technology we're comparing against. Usually, with springing engines, we see like the 37% efficiency. They're about 30%-40% more. Similar to single-cycle or open-cycle gas turbines. Combined-cycle gas turbines are actually doing internal CHP, as I like to call it. Their efficiency is a little higher and kind of matches ours.
You'd get a similar number for them as us.
Right. Is there a minimum load, Akhil, to enable full ramp-up? Talk to us about the capabilities in connection with AI workloads.
Yeah. We're able to ramp between 15% of the input to all the way to 100% seamlessly. Those are kind of the boundary conditions I tell customers to think through. We can do it extremely quickly. We're talking we can respond to those microsecond responses that you saw in some of the chips. We do it often by partnering with other technologies such as ultra-capacitors, which are really, really good at responding on the real quick while the fuel cell can respond on the second level. Combined, we can put together solutions that meet our customers' needs.
Right. I see a number of congratulations and questions about the Oracle announcement.
At this point, we can't comment beyond what we have in the release. That will be it for now. I'm moving on to the next question. Thanks for the congrats. Akhil, can you clarify the overbuild concept? What does that mean?
Sure. If you think about what a data center needs, they've already invested in UPS diesel. They want their energy generation technology just to match the reliability of the grid, so 99.9%. Any given technology has maintenance downtime, has unplanned downtime, and can't reach that on its own. The way you account for that is you build extra of them to try and achieve that goal. Now, the bigger the unit and the larger the percentage of time it's not available, the higher that percentage of overbuild it needs to be.
It's a statistical analysis that goes into that to try and figure out what percentage of asset relative to that 100 MW load you need to try and achieve that. As we saw with the turbine, which is like the 50 MW building block, if it's available 95% of the time, that means 5% of the time it's not available, you need an extra one to cover for that 5%. That's why you have that 50% overbuild. Now, because Bloom's much smaller at 325 kW and we're far more available, far more reliable on the unit level, 99.7%, that percentage of time we're not available is smaller. When it does go down, the amount that goes down is much smaller. Because of that, you only need to build 9%. That's 109 MW relative to your 100 MW need. Hopefully, that makes sense.
That sounds great.
Tell us also about, you know, how can you reassure a data center planner who wants to power like a 1 GW campus? How does the Bloom Fusel solution work for these data centers when we think about this that you can scale power like stacking them like LEGO blocks?
Mm-hmm. Yeah. You know, it all comes down to the scalable nature of that technology, right? Like you saw, it's like I love the LEGO block analogy because I do have a little bit of LEGO addiction, but you can just add them on and on and on and get to that 10 MW level. Often with data center customers, there's a lot of due diligence to get familiar, and very often they see that the system's extremely reliable on a system level because of that smaller building block.
In the data center world, there's this term called blast radius, which means like when something goes wrong, how big of a problem it is. Because of our system size, our blast radius is really, really small. That's how data centers get really comfortable with it.
Great. Another good question is, can you review your efficiency, fuel cell efficiency, and what sets them apart from other solutions? There's specific questions around the efficiency, gigawatt hour, four gas, et cetera. I'm not sure we're going to get that specific, but talk to us about what sets fuel cells apart from other solutions.
Yeah, sure. Our lifetime average efficiency is 54%. We start somewhere in the low 60s, and over time, our systems degrade a little bit, and we average across the life of a contract 54%.
Why is that number so much higher than like the 30s and sometimes low 40s that you see from other technologies? It's a couple of reasons. One, we're a high-temperature process. If everyone thinks back to their high school chemistry class, at high temperature, all those atoms are bouncing around with a lot more energy. That means to kind of break the bonds and reform the molecules requires less energy. That allows us to be a little more efficient. Secondly, it's a direct process, right? So because it's electrochemical, we're going directly from chemical energy to electrical energy. There's no intermittent steps. Compare that to a combustion technology where you have to combust it, produce heat, which then produces motion, which then produces electricity. You've got three or four different steps there along the way.
Each step, you lose a little bit of energy, and that comes out to a much lower energy number in the end. Because of those two reasons, we're able to achieve these much higher efficiencies just natively. Now, when it gets really exciting is when you can use some of the waste heat from our system, and then you can boost that combined heat and electricity efficiency in the north of even 90%. That's where you start to see some really interesting economics.
That sounds great. Another question is, since we are talking about AI workloads and reliability, how would you generate reliability beyond the five nines? Or we talk three nines, four nines, five nines. Tell us a little bit about that.
Yeah. Again, it just becomes down to math, right? I said for 99.9%, which is three nines, I need 109% overbuild.
For four nines, I'm going to need a larger number. For five nines, I'm going to need a slightly larger number. Then depending on what the customer wants, if they want to go even beyond, not that we see many customers going beyond even five nines, even like government-sorted data centers, five nines is kind of like the limit. You could just add more and more. Now, because of those fundamental facts of our building block size and each unit being fundamentally more reliable, to any level, we're probably the lowest percentage of overbuild to get to that level of reliability.
That sounds great. Can these systems, the fuel cell systems, operate in island mode off-grid? If so, how does this work? How do the systems respond to in-rush current when they're starting? Large motors in island mode is a question from Cory. Thank you.
Great question. Absolutely, yes.
They can operate in islanded mode. We have a number of sites that are completely disconnected from the grid where the utility is just not available in a time for a factory to be built out or for a facility to get up and running. They can operate in microgrid where they run in parallel to the grid. When the grid's not available, then they switch to islanded mode. We have smart inverters that can be voltage-following current sources when the grid's available and. Voltage sources when the grid's unavailable. And there's all sorts of clever power electronic solutions that we can do to reach whatever level of reliability required for the customer. To answer your question on the in-rush current, that's a great question. We often use capacitors. So we work very carefully with our customers, understand what those load dynamics are.
More often than not, we see with compressors and motors, they often have VFDs, which limits the in-rush current. But then we complement that with the ultra-capacitors. And that helps put together a solution that's what solves for the customer's needs.
Sounds great. Also, another question. So there's obviously other fuel cell companies around. What is it that makes Bloom's fuel cell so special and competitive versus other fuel cell providers?
Great question. So it's a number of things. So on the technology basis, our fuel cell is a solid oxide type of fuel cell. And you know, I can spend hours and hours talking about the different types, but solid oxide is really, really well suited for primary power stationary generation. It's extremely reliable. Once it starts running, it's really, really, really, really a beautiful technology. It's fuel-flexible.
And it's a number of things that actually you see it comes because of the fact we're solid oxide. The other thing, I think Bloom just has such incredible experience, right? Our first commercial site was 2008. That's still running. That's like 17, 18 years at this point. There's a lot of learning that's gone in. And it's the scale, right? I mean, how many fuel cell companies do you know that are able to have 1.5 GW deployed, learning from those systems operating in the field, and are able to manufacture the scale Bloom does, right? I think that's where Bloom really stands apart from other fuel cell technologies. And that's why you see the AEPs and the Oracles choosing to go with us.
I'm glad we're going around the world with our webinar since there's also a question from someone who's talking about Portugal.
Saying that there's some projects implemented, fed by renewable energy electricity, such as wind and solar, directly from the grid. And then a lot of questions around, how would that fuel cells change if we used a different fuel? You mentioned we are fuel agnostic or flexible. What about hydrogen? Lots of questions around what about hydrogen?
Sure. Okay, a couple of things. One, in fuel cell mode, where we're producing electricity from chemical energy, we can use hydrogen, we can use blends thereof. If hydrogen's commercially available and makes economic sense for the customer, we can run on hydrogen. We've demonstrated this in a number of projects. There are press releases on a project we did with SoCal Gas maybe a year or two ago where we ran on a blend for quite a while to demonstrate this.
Alternatively, if the grid has abundant renewable energy and excess renewable energy, you can run our device the other way around. Like we re-engineer it and we have a product called the electrolyzer. And that takes in electrical energy to produce chemical energy. It's, think of it like a fuel cell backwards. And we have an extremely efficient electrolyzer that's able to do that. And that's also commercially available.
That's right. And one question, I'm seeing a lot of questions about the slides. We're not sharing the slides. So this is your chance to see them. We do have materials on our website that are public, but the slides are really part of this exclusive webinar that you have access to by watching. One last question or two last questions. What about the resilience?
How do you maintain reliability in active tectonic areas, Akhil, like Silicon Valley, Taiwan, Japan, or other areas? Talk to us about resilience.
Great question. So, you know, I kind of glossed over it when I showed you how the cell brings up because I didn't want to go into too much detail. But there's almost no single point of failure in our system. Like everything's talked through on a highly redundant, reliable level. It's extremely oversensored. So I think our fleet on a whole generates 8 billion data points every single day, which we ingest in the data center here in a facility I'm sitting in. And we're constantly monitoring the systems, making sure that they're running in an extremely reliable way.
For the tectonics, there's a lot of site considerations that go into account on how you place the skids, what type of ground you do it on, how much clearances you use. There's a lot of code that has to be required on that. So that's how we handle the tectonic things. In terms of the robustness system, just, you know, in my eight years here, just some anecdotes. I've seen us run through typhoons in Taiwan. I've seen us run through the wildfires of California. In fact, like the wildfires were like inches away from a system that was running. One of my favorite anecdotes is there was, you know, we had a site right on the bottom of the hill and somebody left the parking brake on a forklift unlocked and it rolled down and fell into our systems.
The system that crushed obviously shut down immediately, but the rest of the systems operated. And then we got the call from the customer, "Hey, can you turn it off? We need to like pick up this forklift." So it really goes to show how reliable and robust these systems are.
Great. And then our last question. About in the process of fuel cells, obviously they operate at very high temperatures. So what is the temperature of that waste heat and also the heat that the fuel cells produce? How can that be managed?
Yeah, so great question. It's about 400 degrees Fahrenheit. We are using it in a number of places around the world to use it for district heating, to use it for cooling. In fact, our factory is powered by an absorption chiller that runs off the waste heat.
In terms of how we manage it, we have advanced control systems that handle under the valves. We can kind of direct how much heat goes into the heat exhaust versus our regular exhaust. That is how we kind of control and manage to make sure that the customer is retrieving exactly the level of heat that they need.
Thank you so much. I think that brings our webinar to a conclusion. I wish we had more time. We will take a look at all questions. As I mentioned, we are in a quiet period, but we also will take good care of looking at all the questions. Dial into our next webinar as well and follow us on our website and check into earnings next week. We are announcing earnings on July 31st, and we hope to stay in touch with you.
Thank you so much to everyone for attending today, and I wish you a great day, evening, whatever you are up to today. Thank you so much.