2026.10.07 | TB | Power | Datacenter Power Generation & Bloom Energy

Client-led • 25 Sep 2026

Companies: META US, BE US, ORCL US, NVDA US, CWR GB, A000150 KR

Expert Former Executive, Supply Chain & Procurement at Meta Platforms Inc

Agenda

• Around 25 years' experience in HPC (high-performance computing) infrastructure and data centre power generation
• Well-placed to discuss hyperscaler evaluation and procurement process for behind-the-meter power generation, particularly for Bloom Energy, as well as alternative technologies such as gas turbines, reciprocating engines and renewables
• Knowledgeable on data-centre power requirements, fuel-cell performance and economics and technology selection by site, as well as practical considerations around grid access, lead times, infrastructure and 800V power delivery

Introduction

Client [00:04:16]: We are interested in knowing more about Bloom Energy. Can you tell me about your current position? What did you do at Meta?

Expert [00:04:54]: Yes. Professionally, my background is electrical and electronics engineering. The role I have been playing has been as one of the key decision-maker when it comes to data centre infrastructure management, including power, which is coming from the grid or the behind-the-meter, the thermal infrastructure and also tying that to the IT infra for AI training, inferencing and high-performance compute depending upon the specific site where Meta is looking to deploy their capacity.

Questions

00:06:57] When you were at Meta, did you evaluate Bloom? What is your overall feel about its technology vs other behind-the-meter options available?

Expert [00:07:34]: Yes. Bloom Energy, you probably know this already to some extent, is the number one company at the moment as it relates to being the manufacturer of natural gas-based solid oxide fuel cells. The reason I said they are considered as the number one company is because they are a fully vertically integrated company in which not only they have their own design and engineering IP, IP means intellectual property, but also have the manufacturing capability to build natural gas-based solid oxide fuel cells. They are also looking into pioneering the hydrogen-based as well as biogas-based fuel cells moving forward, but in the hyperscaler data centre vertical, they are considered the largest manufacturer of solid oxide fuel cells that is becoming a critical requirement from a behind-the-meter power capacity point of view. Meta considered as companies who have approved Bloom Energy as one of the vendor from who they are looking to procure solid oxide fuel cells.

That is in the context of Meta, but my interaction with Bloom Energy and my understanding of Bloom Energy is wherein I understand how Bloom Energy is considered one of the leading provider for Oracle, NextEra, Crusoe and some of the other data centre hyperscaler as well as independent power producer and neoclouds that are looking for providing the behind-the-meter power capacity. SOFC is considered one of the critical category, and I can give you the reasons or I can explain more, but let me give a pause here first.

00:09:39] At Meta, what was the decision-making process for power generation? Do you look at all the upcoming and newly constructed data centre that's coming online and how much of that's accessible by grid? Grid will be your first choice. After that, how do you decide which behind-the-meter option to choose?

Expert [00:10:15]: Great question. Let me explain in terms of Meta. Meta generally will have today, as of 2026, around, I would say, around five years of roadmap that they have in the pipeline for developing the greenfield data centre and also upgrading their existing brownfield data centre. Now when it comes to sourcing the power, while you use the word and I'm not going to say that it's a wrong choice, you would say, "Hey, the first choice is always the grid." Yes, I would have agreed with that 100% like 3-5 years ago, but today, the reality is that for the hyperscaler company like Meta, there are certain types of workload wherein we are not even considering the grid as an option or as a preferred direction because for some example, for some of the locations, which could be, let's say, in Dakota or Nevada or in a very remote area of Canada, it is possible that grid company may not be able to provide the required capacity within the timeframe we are looking for.

Rather than relying on the grid, we might be looking to build a data centre, which can be called as completely off-grid. That means it is being built as an island and the entire data centre can potentially be powered by non-grid-based power and there are different type of workload which can be managed through that direction, but just wanted to give you the context that while grid certainly has been traditionally the way, but doesn't have to be that way going forward.

00:14:02] How much is going to be addressable by generators, turbines? At that time, are you evaluating reciprocal engines?

Expert [00:14:21]: Yes. I forgot to mention reciprocating engine. Yes, reciprocating engine, aeroderivative gas turbine. I would say of that 60%, my understanding is approximately 65% or more, close to two-third is going to come from primarily the turbine engines, including reciprocating gas turbine engine, simple cycle gas engine, aeroderivatives.

00:17:36] Nothing is available for nuclear at this point in time. Given the five-year window, can you allocate 5% to nuclear? Do you allocate at the back end or is that how you give your estimation?

Expert [00:17:54]: First of all, I agree with you, nuclear is certainly on a longer-term roadmap. When I included nuclear in the last 10-15%, I am attributing some small percentage, could be 3-4%, wherein we can get some kind of SMR or MMR nuclear reactor-based capacity. It is going to be a small percentage, but I do anticipate some capacity coming from nuclear.

00:18:37] At that time, did Meta not consider reciprocal engines, such as diesel engines, as power generation? It sounds like, based on the breakdown reciprocal engine, the dirty engines are not included.

Expert [00:18:54]: That's a good question you asked. I would say reciprocating engine now essentially are being preferred from companies such as Caterpillar, Cummins, Baker Hughes, which are using natural gas as a fuel rather than using the diesel-based generator because diesel generators are becoming very, very unfavourable for reasons that you very well know. They are still purchased, but primarily as a backup power, like a backup generator to provide the backup power, but for the primary power or the bridge power, nowadays, reciprocating engines are being built as a flexible skid-mounted but using natural gas as a primary fuel rather than using the diesel. Your point is totally understandable.

Client [00:19:48]: Right now, are they considered as a backup power? Is it because its burning, is it environmental reasons or any other particular reason that is not considered as a primary power source?

Expert [00:20:03]: Reciprocating engine. Is your question about reciprocating engine for the diesel based or the natural gas-based?

00:20:12] You can also do natural gas for reciprocal engine. At that time, did Meta simply not evaluate reciprocal engine using gas?

Expert [00:20:32]: No, I included three type of engine. One is the combined cycle gas turbine engine. The second one is the aeroderivative gas engine and the third one is a reciprocating engine.

Expert [00:22:30]: Have you heard about the Nvidia Vera Rubin, Rubin Ultra?

Client [00:22:37]: Yes, of course.

Expert [00:22:38]: Starting from later this year and going forward, majority of the greenfield data centre are turning towards 800V power delivery architecture. 800V is primarily the higher voltage compared to the traditional 48V or 52V. SOFCs are considered one of the primary mechanism that 800V can be delivered directly to the server rack. In this case, you can bypass the transformers, etc. This is becoming if I can use the word gold standard, this is very quickly becoming a gold standard in the data centre world, wherein almost every hyperscaler is adopting the 800V architecture driven by Nvidia, Google and AMD GPUs.

Client [00:23:36]: How is that relevant to the fuel cell utilisation?

Expert [00:23:44]: Because SOFCs can be directly connected to the server rack. 800V is what you are going to deliver to the server rack. You can bypass almost 40-45% copper cable, and you don't need to go through AC to DC conversion, which you otherwise require from the gas turbine engine or the grid power. You can deliver the 800V directly from the fuel cells into the servers. This is one of the biggest demand driver for the SOFCs, including natural gas-based, hydrogen-based and biogas-based as we move forward.

00:24:34] If you use fuel cell, you can deliver 800V directly to data centre. What do you have to do to make that conversion if you use natural gas?

Expert [00:24:42]: What happens is SOFCs, you can connect a direct 800V to the server rack like a GPU rack, etc. The gas turbine engine or the reciprocating engine or the aeroderivative engine, their output voltage is the AC voltage. You have to convert it down to the low-voltage AC and then from low-voltage AC, you have to bring it down to either 48V DC or 800V DC. It has to go through multiple stages of AC to AC and then AC to DC conversion, which is how you generally do with the grid power as well. The SOFC can bypass entirely those stages. That's why I said 40-45% of the copper cable can be saved by directly connecting the SOFCs to the server rack. Just keep in mind, this 800V is primarily targeted for server rack, not for your cooling or your power distribution unit or your other equipment like a server networking or storage. It is primarily catering for the servers, GPU-based servers.

00:25:57] Is the conversion process for turbines in a reciprocal engine to 800V a time needed process, an expensive process or both?

Expert [00:26:08]: First of all, it's a traditional process. Even today, how do you use the turbine today? The way you use is the output voltage of the turbine generator, which is the AC voltage, it has to be converted down to a lower voltage AC using transformer. Then from the transformer, it goes to the power supply unit and power supply unit converts it to the DC voltage. This is the same mechanism you use for the grid power, wherein you convert the grid power through a high-voltage transformer then to medium-voltage transformer, then to low-voltage transformer, then take it to the power supply unit, which converts it to DC voltage. You can bypass all these stages by deploying the SOFCs in a close proximity to the server rack outside the building and run the DC power cable to deliver the 800V.

00:27:05] I'm not familiar with the networking and racks within the data centre. You said you were able to keep more of the copper wire. What do you mean by that?

Expert [00:27:18]: What I meant was that when you are required to convert your power from a high-voltage to medium-voltage AC, from medium-voltage AC to the low-voltage AC, then from there, you have to run the cable to the power racks and you convert them to DC voltage, you require copper cable. Like in your house, when you're running a refrigerator, the power comes from outside, then it runs your cable inside the house, then you bring it to your refrigerator. All those cables, all those power cable can be significantly eliminated or reduced by approximately 45% if you are bypassing all those conversion stages and delivering the 800V directly to the server rack.

Client [00:28:29]: If you use fuel cell, you can deliver 800V directly into the data centre. Can the amount of copper cable you use be reduced by 45%?

Expert [00:28:39]: Up to 45% and that's not the only reason. The other reason is the power conversion efficiency, because when you go through multiple stages of power conversion, you lose a lot of power through the heat, transmission wastage, conversion wastage. You can eliminate a lot of that wastage by using lesser number of conversion stages. Hyperscalers are looking at around 7-8% efficiency gain in the power consumption by using these mechanisms.

00:29:24] When you're evaluating for power generation, especially for behind-the-meter options, is there a pecking order of where you choose or what location? Do you say you're going to do gas turbine first and then fuel cell, followed by renewable energy?

Expert [00:29:52]: Yes. I will say, generally, when we are selecting a particular data centre site, or even somewhere like a remote region. Before the site selection is done, one of the calculation is how much capacity of compute we need to plan in that location? What is the size of the land that we are able to secure? How much power is required? Do we require 200MW, 300MW or 500MW in that location? This calculation about the wattage or the power is generally determined by what are the different sources of power that we can potentially secure the power from? Can we secure 20% from the grid and 80% is going to come from behind-the-meter or can I get 80% from the grid or 60% from the grid and 40% I can get from behind-the-meter? Even from the behind-the-meter, we have to look at does this location have a proximity closer to a large amount of natural gas resources and there are logistics and supply chain of gas turbine engine available or does this location have hydroelectric power as a potential capability. We do perform like a site-related survey and site-related analytics.

We have to do a very clear calculation about the total power capacity that will be available by the time the data centre becomes fully ready, and that might take a three-year, four-year timeframe. Speccing the power supply options, the priority is based on the location, the size of the land, the capacity we are looking to deploy, availability of the infrastructure like a gas turbine engine or other resources. We do look at all those factors before we make a decision to build a certain capacity at a certain location.

00:35:54] What was your evaluation after you did some testing with Bloom? Was it as good as expected, better than expected? Was it performing? Is there anything concerning after the pilot programme?

Expert [00:36:14]: From a technical performance point of view, certainly, Bloom Energy's SOFCs are able to provide a power conversion efficiency of around 40%. The industry standard right now is around 35%. It is projected that with some better control system of their fuel cells, they are looking to achieve a power conversion efficiency of around 52% in the next 2-3 years. What I will say is, Bloom Energy's power conversion efficiency is the highest among the industry for the SOFCs. The reason for that is they have a very strong software and firmware, which is able to provide a better energy conversion efficiency, and they use some of the semiconductor controls to do even a better management of power conversion of their fuel cells. They certainly are, I would say, ahead of the game. One of the closest competitor of Bloom Energy is, have you heard about the name called Doosan?

Client [00:37:21]: No, not really.

Expert [00:37:23]: Doosan, this is a company which is considered a close competitor. The challenge with Doosan is they don't have the in-house solid oxide fuel cell technology. They have purchased this technology from a company called Ceres Power. Ceres Power is a UK-based company, which sells its IP to the other vendor. Doosan is considered as a very strong competitor to Bloom. There's a Chinese company called Weichai. They are also coming up very rapidly for our fuel cells.

Client [00:38:03]: Is it performing ahead of its fuel cell peers regarding power conversion?

Expert [00:38:11]: Power conversion and the cost of the fuel cell, the cost also.

00:38:18] People have mentioned that price per CAPEX per gigawatt or per megawatt is higher than gas turbines. Is cost a consideration when you evaluate fuel cell technology in general?

Expert [00:38:42]: Yes. Fuel cells are definitely very expensive, significantly more expensive. It is certainly much more expensive. The reason for that is there is not yet enough of the manufacturing capacity or the scale. There are certain areas of the manufacturing, which require a lot of upgrading. I would say, in the due course of time, the cost of the fuel cell will come down, but compared to gas turbine engine, fuel cells are definitely, more expensive. Now, there is a direction that the power conversion efficiency of fuel cells is going to be better than gas turbine engine. Secondly, the operational cost or the operational and maintenance cost and the ability to recycle the heat from the fuel cell, this will allow the total cost of ownership of fuel cells to become lower in the 5-6-year timeframe for which they will be installed compared to gas turbine engine. Have you heard something called LCOE, levelised cost of energy?

Client [00:39:58]: No, please explain.

Expert [00:40:01]: Levelised cost of energy means, let's say, you're buying the power from the grid. Generally, you pay around USD 35 per megawatt hour. With a solar power, you pay around USD 40. With a gas turbine engine, you are paying around USD 55-60 per megawatt hour. Fuel cell, today, you are paying around USD 80-85. That is the CAPEX. Now levelised cost of energy means on a per megawatt basis, how much am I paying for the equipment, for the fuel, etc. The LCOE is different from TCO. TCO means it includes your operational expense, your fuel expense, maintenance costs, etc. It is almost sure that SOFCs can never reach the TCO of the grid because grid obviously is an established category and you can never achieve the same level, but because of the other factor like a lead time, capacity issue, etc, customers are willing to pay higher LCOE for some of these premium sources of energy such as SOFC because this allows us to build our data centre rather than waiting for the grid power to be available.

00:41:22] It looks like on the testing programme, the pilot, you're satisfied with the performance. Is that a small sample? Did you recommend at the time that Meta should go forward with Bloom or was there any other consideration causing you to evaluate a little more before doing so?

Expert [00:41:49]: Generally, we go through the vendor evaluation process. There could be some like a company like a Schneider or GE Vernova or even ABB, even these guys are looking to build SOFC in future. Meta certainly is going to look at multiple options. Now in the due course of time, the design and the technology of the SOFC also will become very commoditised. Rather than being dependent on a single player, we might have an opportunity to buy from common marketplace from whoever we can get the required capacity or the price point. Bloom Energy, as I said in the beginning of the call, they are at the leading edge, but that doesn't mean that other companies will not be able to catch up.

00:43:04] Are the turbine folks thinking of getting to fuel cell energy as well? Given what you know and evaluate, what is your biggest concern after your pilot testing with fuel cell? You talked about some positive aspects.

Expert [00:43:35]: Yes. Number one concern is capacity, whether Bloom will be able to deliver on the megawatt or gigawatt of capacity that as a customer we require, how do we hedge our risk of supply chain on to one or more vendor. From that point of view, because Bloom Energy is still a relatively smaller company compared to other big player, in order for them to deliver on the capacity, they have to have capital raise. They have to have the manufacturing capacity built up. They have to have the logistics of natural gas as a resource pipeline connected. There are many factors which are going to determine their manufacturing capacity, and that is one of the biggest concerns that hyperscalers have in with Bloom as a reference.

Client [00:44:28]: The ability to scale manufacturing. How did Meta come to you to alleviate that concern?

Expert [00:44:41]: It is not about they able to alleviate. They also realised that while the market demand for SOFC is going to become, remember, I told you it is going to become almost like a gold standard. Bloom realises that the demand is going to be high, but they have to really show their manufacturing capacity, their power conversion efficiency, their ability to bring the price lower. There are some of these technical and logistical factors, which Bloom has to overcome or to take account of in order to provide us a guaranteed capacity. In the public domain, they have announced they are going to deliver around, I think, they have signed a 2.8GW deal with Oracle. They have announced they have a deal. I have a reason to believe that they do know what their capacity is going to be, but each one is going to be signed as a one-on-one contract with the customer.

00:47:45] Oracle announced that its data centre may be delayed because of regulatory reasons, since it didn't get the permit. In the political heat map is there bipartisan effort to slow down new data centre build-out? If data centre deployments delay, it's going to impact the behind-the-meter power generation demand. How should we think about that?

Expert [00:48:28]: I would say delays, cancellation, adjustments in the capacity, hyperscalers generally are talking about 25-30% of their capacity that they're announcing today, not entire 100% or not 70%. They're saying 25-30% of the capacity announcement might go through some type of adjustment correction, delays, etc. That part I will totally agree with you. I'm also not going to be surprised because as you can imagine, when companies like Meta, Google, Microsoft, they are announcing capacity, these guys are doing a lot of smart analytics internally when they announce these capacity in the public domain. Sometimes they also know that not entire capacity is going to be met with. That is one part I will totally align to. The part which I would like to say is not necessarily a concern is whether that slowdown or delay or cancellation, will that have a big impact on the behind-the-meter. The reason is the power supply situation is so tight that even if the grid company and the behind-the-meter companies are continuously looking to add capacity, they will still not be able to catch up the timeline at which hyperscalers are looking to deploy.

That means the correction or the slowdown that you're talking about, the Oracle, etc, I honestly do not believe will have any direct or immediate impact on the behind-the-meter capacity procurement planning because that capacity development is very much way behind, like the gap between the demand and the supply is so large that behind-the-meter is not going to catch up even if the demand gets cut down by 20-30%. I don't believe it is a demand cut down. It is more about the delays than anything else.

00:50:34] Power generation is relatively new to me. The consensus among investors is that people will look at gas turbines and reciprocal engine first. Once exhausted, they will go to fuel cell or other options. You make evaluations data centre by data centre, location by location. Are there any instances where a site was better for fuel cell over gas turbines? Have you come across that?

Expert [00:51:43]: I'll say, yes, I have. We have done, and we have chosen the site wherein on a certain location, securing the SOFC is a better option than the gas turbine engine, but at the same time, in this, we are assuming, you and I are assuming, that you use either the gas turbine engine or SOFC because the purpose of the gas turbine engine primarily is much more broader. You can use the gas turbine engine to provide power for the entire data hall, which includes your server, networking, storage, your building, etc. The purpose behind SOFC is very, very straight and very unique. It is only and only for the server. There is some level of a common application, but there's a large level of non-overlapping application as well. That means when you ask me the question, whether you decide to use SOFC vs gas turbine engine, my answer is yes, but only in the context of the power that goes into the server rack. Only from that point of view, there's a common competition between them. Otherwise, you can think of SOFCs and gas turbine engine decision being made on a cumulative basis rather than on a sequential basis. That is how I will say it is rather than thinking of one or the other as an option.

Client [00:53:19]: In other words, in certain data centre, you would recommend both? You have turbines, would you have fuel cells together?

Expert [00:53:27]: Absolutely. Think about this way. The gas turbine engines certainly are certainly the most preferred, but the advantage with the SOFC is that you can bypass a lot of these dependency on the gas turbine engine, particularly for your server rack by connecting these SOFCs directly. You still require a gas turbine engine, but those can be used for your other non-IT infrastructure or for some of the less critical server, while SOFCs can provide a very, very focused power for high-end servers such as Nvidia Vera Rubin, Rubin Ultra, etc. Even though both technologies use natural gas, from an application point of view, SOFCs are very much restricted to the servers, high-end server, while gas turbines cover the entire footprint of the power that can be delivered to server as well as non-server infrastructure.

00:56:46] It sounds like a lot of decision-makers, when thinking about behind-the-meter, want to choose the most mature technology and supply chain ecosystem. This is less execution risk. Even though TCO [total cost of ownership] could be lower for fuel cell, is the natural choice, if that's available, going with that because they have the most mature technology? That's going to gas-based. To the extent you cannot, if turbines are in short supply and lead times are three years, then you have to evaluate fuel cell and other technologies. I have high-power Rubin chips, they use 800V, and is fuel cell suitable for that?

Expert [00:57:46]: Yes. I would say the decision to use SOFCs is pretty much happening in parallel. Let's say, hypothetically, today, Meta is building a data centre in which they have zero SOFCs planned. The same location in 3-5 years, they can add SOFCs-based power. This way, they can add higher compute capacity at the same location. I do not want to give you the impression that once you choose the gas turbine engine, you cannot add SOFC in future. You can add. That is a possibility. The point which I agree with you is that because of the lack of maturity, because of the cost, because of the scale, today, I don't have the luxury of buying SOFC, but that doesn't mean that in the next 3-5 years, when SOFCs become more available, I can certainly add those on to that location where I did not have SOFC to begin with. That is going to be part of the calculation that hyperscalers will be making that how much incremental capacity they can add to the existing site.

Client [00:59:02]: Is having a more mature technology ecosystem having more vendors? It can't be fuel cell being the biggest vendor and several other small vendors there. Fuel cells have a wider adoption from the major players. Like you said, some turbine folks are getting involved, is that a positive for the fuel cell?

Expert [00:59:25]: That is correct. Correct. Like GE Vernova, ABB, Schneider, etc.