ICYMI: Bloom Energy odds and ends
In case you missed it, the Bloom Energy updates from my Substack notes
“I made this one longer only because I have not had the leisure to make it shorter.”
— Blaise Pascal
I tried to make my first post on Bloom as interesting and readable as possible, but alas, I might have failed when it came to conciseness as it clocked in at around 5000 words. When you’ve been following something for years, it’s not easy to put in words the web of knowledge you’ve acquired. That is despite spending days of effort on it. I now have a newfound appreciation for what it probably takes to write a book.
Since I didn’t want to keep adding to the post, I started writing short Substack notes and instead updated the post to link to them as footnotes. Unfortunately, it doesn’t look like Substack notes are a heavily used feature at least based on the stats I see for mine. Very few folks seem to have clicked on them despite the post itself being popular. As an experiment I’ve decided to slightly edit and collect them all into this post, so everyone gets it in an email. If you’ve read them, feel free to skip or go back to the original post (link below) and read it in context. Feedback welcome if you like this or not.
You can find my previous post on Bloom energy here:
The implausible Bloom of an energy equipment company in Silicon Valley
“There’s a famous story about a fireman from New England. Apparently back in the 1950s he couldn’t help noticing that a local Tambrands plant (then the company was called Tampax) was expanding at a furious pace. It occurred to him that they wouldn’t be expanding so fast unless they were prospering, and on that assumption he and his family invested $2,00…
Valuation
Since it’s bound to come up, I might as well address it first. My Bloom post was a personal investment story and dive into the technology without any focus on price or valuation. Moreover, with a very low basis and a long horizon, the risk calculus is naturally very different from someone wanting to buy after a short run.
Said differently, I might not buy more, but neither would I sell as long as the story checks out other than perhaps for extreme position size risk management. This is one of the hardest decisions to make for an individual investor and I’ve talked about it before -
The valuation today totally depends on the future prospects and the probability of that happening, but a low basis gives you a margin on safety on not needing to be super accurate but directionally right. A vivid example of this is what happened to Cisco during the dot com boom. Everyone likes to bring it up as an example of the poster child top stock of the dot com bust that went nowhere (till very recently). However, let me show you something else that folks miss. If you bought Cisco early enough and held on, you would have done absolutely fine. Let’s say you bought it a month after the “Netscape moment” in Sep 1995 (note that this is much later than Cisco’s IPO) convinced the internet will need lots of routers and held on, you would still beat the S&P 500 and never dip below it, despite a greater than 86% drawdown -
I’m not saying you completely forget about valuation. If a bubble is obvious and you can identify it, you probably want to sell.
Another data point: from around the point in time when I bought nVidia in mid 2013, even if it now drops another 99%, it will still beat the S&P 500! Can it drop that much? Anything is possible but it’s less likely IMO as the story stands currently. Moreover, if you sell in a taxable account, you immediately face the bear known as taxes. Even your psychologist cannot tell you what to do. You have to sell down the point that makes you sleep well at night and that point changes with experience, wisdom and wealth level. Ultimately, you have to do you own work and make your own decisions.
A reader had more questions about valuation and when one might buy. Even for me, it’s not a very big position size so I would not mind buying more at some kind of significant drawdown. But the question is what price would that be? There are just too many variables including the whole trajectory of how this AI buildout will play. Are we going to have some kind of “trough of disillusionment” phase at all or is it different this cycle due to all the Hyperscaler/Sovereign Wealth/Private Credit/Whatever kind of money involved? Honestly my best answer at this time is that I don’t know right now what that price/valuation will be, but I think we will know it if and when we see it.
If someone were to put a gun to my head, here’s how I might approach it. Assume that Bloom is only able to ship the 1GW in 2026 that they have capacity for right now without any expansion and then apply a price/sales multiple similar to what GE Vernova has (which is a bigger potentially more stable company but that’s balanced by lower growth prospects compared to Bloom). That gets you in the $40-60 range depending on what values you use. That’s also conveniently in the same range as the previous meme high in 2021. Who knows if it will ever get there? It might need some kind of macro/economic shock, but I would seriously consider buying more at that point if the fundamental story hasn’t changed.
Edit: The above is old price analysis and I did buy more early in the Hormuz War doldrums as I mentioned in later posts
Can Bloom scale production?
William Martin Keating (great semiconductors Substack) asked an astute question about Bloom’s ability to meet demand -
“I guess my question is now whether Bloom can meet all of the anticipated demand that cannot be met by the national grid in a timely fashion? Do they have the capacity to ramp up production?”
This was my stab at it - I guess the answer depends on what the demand is ultimately. Meeting the incremental demands of the entire grid for the country is for sure hard. The marginal demand for high priority DCs that are really constrained by power and willing to spend, can be met. Many DCs would want the diversification of being grid connected to begin with or later if available so it’s not like all of them will go this route.
So, what could Bloom deliver? Bloom has so far delivered about 1.5 GW of installed capacity in its life. But they currently have capacity to produce about 1GW per year and have already announced plans to increase that to 2GW in 2026. They have two production facilities in Newark, Delaware and Fremont, CA that can be expanded further without a very high level of CapEx. It’s in the order I think of $100M per GW of capacity.
The other thing to keep in mind is that the same modular nature of their power generation that helps their reliability also I believe helps in production. It is similar to semiconductors in the sense that once you perfect the process/recipe/flow then you are replicating it at a mass scale. Incidentally, Intel might be giving up “Copy Exact”, but Bloom mentions that their production facilities use such a “Copy Exact” methodology where they can easily stand up another facility by copying existing ones. I believe there are other aspects of getting the “yield” of their ceramic cells and stacks that are similar. You still need to setup the factory/fab but then the incremental cost/effort isn’t that big. Of course, it’s not quite as efficient as shipping small silicon dies. We are still talking about a bunch of medium sized cabinets full of ceramic stacks, metal, piping etc. But compare that to forging/casting and assembling huge CCGT (combined cycle gas turbine) assemblies. Hot section single crystal super alloy blades can take 60+ weeks to produce. It seems you literally need to pull a mold through a furnace at millimeters per minute to grow one grain from bottom to tip. Even shipping these turbines is non-trivial. They are also backlogged till 2030. One of my uncles happens to be a metallurgical engineer with experience in forging (nothing fancy like turbines) so I have some directionally correct knowledge about what it probably takes.
(Edit 07/03/2026: I finally spent some time finding the right video to explain single crystal turbine blades. While the Jon Y one is pretty good, I think this one is even better. No affiliation, I just like the video. Start at 12:48 for a faster intro) -
Note that a CCGT gas turbine IMO is the right comparison for a baseload type generation that Bloom excels at. I’m aware that aero derivate gas turbines can be deployed faster, but a real utility would not buy them for baseload. But Utilities do buy Bloom’s SOFCs.
If there really is a need, some of the big names who have even speculatively talked to the extent of funding their own semi fabs, could I suppose spend the money to increase the production of SOFCs either for Bloom or any other viable competition that manages to show up. You don’t need super expensive Lithography machines from ASML to make this happen. Certainly, if needed it should not take till 2030 to increase capacity.
Now some might object that SOFCs are filled with rare earth, but I believe the Bloom stack is a Scandium-stabilized Zirconia and they have claimed that their main modules are not dependent on materials from China or any war-torn regions. This can only mean Scandium which is also produced outside of China in Philippines (and Zirconium is not a problem). More supply is coming up online in other places.
Finally, none of this is proof that just because they can do it, they will be able to do it for sure. Execution is still key. But we do have some track record to go on.
Who are the real potential competitors of Bloom Energy?
There’s a surprising amount of uninformed commentary out there about who the real competitors of Bloom are. Many fuel cell companies don’t even do SOFCs. And of those who do, most are years behind. Let’s take a look in a bit more detail.
There was chatter a while ago on X about some potential new competition which led to a reply by me and interaction with some informed folks. More about that and links later.
Apart from some local players in the US (see backup power section) and Japan, currently the leading companies with the technology which comes closest to Bloom, if not the operational and production/shipping experience are Ceres Power and Elcogen. Ceres Power is a UK based company that primarily licenses its technology for manufacture of stacks and complete end systems by others. Elcogen is an Estonian company that manufactures cells, stacks and modules but complete end systems are up to others. Incidentally both companies were founded at the same time as Bloom yet haven’t shipped even a small fraction of what Bloom has managed.
But what about my comment in the original post about a Korean competitor? That was about Doosan Group which has a fuel cell division (HyAxiom) with an interesting story and path to SOFCs. They bought a fuel cell company in 2014 called ClearEdge Power, which itself had bought the fuel cell division of United Technologies UTC (now RTX Corp). That division probably had the longest history with fuel cells starting with the Apollo space program in the 60s. However, that still did not lead to much commercial success. Their PAFC solutions have lower efficiency so they had to license from Ceres to get their SOFC line rolling. This is currently shipping only in South Korea and the efficiency IMO isn’t on par with Bloom due what seems like lack of complete internal reforming.
So, who is the new potential competitor? It’s none other than GE Vernova. SOFCs can be a very good complement to turbines so it makes sense. They have some history with fuel cells, but as I had to point out this is just an announcement and by their own CEO’s admission it will take 3-5 years before they are even in production. See the following X threads that led to this note -
Unlike GE Vernova’s posturing, if gas turbine manufacturers are going to do SOFCs then Mitsubishi Power has a better chance since they have a longer history of actually doing SOFCs in Japan.
Bloom’s cost reduction trajectory
I was looking at some of the notes I took over the years listening to Bloom’s earnings and decided to look again at a presentation made in Oct 2020 where a slide of their cost reduction trajectory was shared. The initial costs were even higher than I remembered.
The trend has slowed, but I believe they have line of sight to get to $1.5K/kW. The question really is a matter of the volumes they ship if you understand experience curves. If the AI boom/bubble lets them ship multiples more or folks invest in them to scale production further, then they could get there a lot faster than most are expecting.
What’s special about $1.5K/kW? CCGT gas turbines were about $1K/kW before COVID. Add about 30% inflation and some more for the supply chain effects and it’s going to be hard for them to drop below that never mind the scalping going on now. At that point they might have real competition that’s better in many other ways.
More insights into Bloom’s experience curve and energy costs
I had thought I was done talking about Bloom with my various notes, but stuff kept happening like SemiAnalysis’ piece on power for AI datacenters, where they mention Bloom and I answered a bunch of questions in the comments about Bloom’s learning rate experience curve and energy costs. Incidentally SemiAnalysis later acknowledged in a comment that my deep dive was part of their research. Some kind of attribution in their article would have been decent, but I suppose acknowledging in a comment is better than nothing.
Grant asked if Bloom just benefits from economies of scale or there are fundamental improvements (emphasis mine) -
“One thing everyone seems to miss on Bloom is how quickly their costs have come down. From 2020 Q3 to 2024 Q3 (the last time they reported exact kWs sold), their inflation-adjusted COGS / kW dropped 49%! Maintenance costs have dropped similarly. We don’t know how much of this is due to economics of scale, which should ramp quickly, or fundamental design and manufacturing improvements, which will be slower. However if the trend continues, it seems like they may eventually be able to compete against combustion technologies on cost.”
My answer was that it is both due to economies of scale and fundamental improvements. Apart from my Bloom post and the associated notes, if you want further proof, see slide number 18 below (Energy Server Evolution) from this presentation -
If it was pure economies of scale, they would just be able to buy materials in bulk at cheaper costs. But note that the amount of material used per “energy server” is coming down and the power density is increasing as well. SemiAnalysis says that Bloom itself doesn’t reveal what materials they use but it’s not hard to find if you know where to look and infer whether they are beholden to China for rare earths again as mentioned in my notes.
Grant further asked - “That advertised difference in specific power is incredible, but we're only at 325 kW today. Have you heard anything on why the 500 and 750 kW servers don't exist?”
My answer - I took a look at my notes and re-checked presentations from the past. The 500 kW and 750 kW servers were planned at some point but would have used more power modules (PM). These are the power cabinets in the energy server cluster. Today in one energy server you have 5 PMs and 3 other modules for power distribution, telemetry and water etc. The 500/750 ones were planned with 10-12 PMs. For whatever reasons Bloom has decided to keep the older module arrangement but the individual PMs have benefited from the power density improvements. Bloom 2.5 had a 42 kW PM for a 250 kW server. The 750 was planned with 75 kW PM. Instead, we have 65 kW PMs today. So, in effect they have reached the equivalent of a 650 kW energy server. Interestingly, Bloom 1.0 from 2008 to 2011 was a 25 kW PM with a different sized cabinet. The server evolution slide seems to allude to this and likely shows that the 500/750 were planned with different cabinet sizes. They could have decided not to bother with changing that.
Bloom’s delivered energy costs
Wing's Investment World asked what Bloom’s power cost figures will look like if you include tax credits -
“I have a question regarding Bloom Energy’s total cost of ownership (TCO). If the new tax credit benefits are taken into account—specifically the 30% capital expenditure credit under TBBB—how would the revised TCO figures look?”
My answer - You don't have to look at SemiAnalysis' numbers. Bloom had publicly posted its PPA figures for purchasing power from them which include such credits and it was 9.9 cents/kWh. And that assumes some kind of profit margin for them. No idea if that has been revised up due to all the AI demand but you can be reasonably sure those are in the ballpark.
There were doubts expressed about whether that included stack replacement costs. I pointed out that these are typically long-term agreements where the maintenance is an important part. For example, the Series 10 PPA Press Release includes the following phrase -
“...The offer includes maintenance and 24/7 monitoring.”
That price almost certainly includes tax credits since it’s from 2023. For those not aware of the minutiae, the section 48 ITC was going to go away (for “non-green” natural gas SOFCs) starting in 2025 as per the previous admin’s Inflation Reduction Act. But the current admin’s One Big Beautiful Bill Act re-enabled that with slightly different terms starting in 2026. That effectively means that 2025 was no man’s land. You had to start construction before Jan 1, 2025 or in 2026. You can imagine the revenue effects in 2026. Maintenance being part of the PPA is consistent with what Bloom has filed with the SEC since the S1.
Backup SOFC solutions and stack life
In my original post I mentioned that there are a few smaller manufacturers that seem to do a good business with back up and other applications. There are some sell side analysts who believe that SOFCs make for a bad backup power solution. However, that is misunderstanding of some of the other unique benefits that can be applied to special needs. Let’s take a look at a few of them.
Let’s start with RedHawk energy systems that can install an SOFC based backup power solution which can go without maintenance for years(!) and can operate between -40°F to 122°F just running off standard propane tanks (which are available just about anywhere). Note that the specs say 3000 hours of life which is fine for a backup application as it works in conjunction with a battery and/or online power. But the actual fuel cell manufacturer is now owned by Edge Autonomy, and they have a longer life version. This one has a life of about 15K hours which is a bit less than 2 years. Compare that to Bloom which has a stack like of about 6 years which IMO alludes to a first mover learning rate advantage.
There are some other manufacturers like WATT fuel cell that even have home units, but these seem to have lower efficiencies as they don’t yet use reforming reactions (see my original post for details about these reactions). These currently seem to be geared towards parallel grid backup for utilities in rural remote areas (gas distribution can be more resilient than electricity because worst case you can just use propane tanks). Unfortunately, I can’t seem to find cell life details on their site. My point is that there are other SOFC solutions which are finding applications in specific niche areas where its unique advantages are being put to good use.
Conclusion and future updates
I hope this was useful. I’m certainly not a professional analyst but these kinds of insights are possible from the rare person who has followed and analyzed something for years. Certainly, it will be hard to find it for free from someone else who has.
I still have things to say for at least a couple of more posts apart from following the earnings calls and significant future updates.
Disclaimer: None of this is meant as advice. I would encourage you to do your own research and invest according to your circumstances.
Excellent analysis! Your Bloom insights are so valueble.
Curious to know your thoughts on whether you are actually of the opinion that they are pretty likely to scale up production meaningfully from here on, given that the recent air-permit filings especially the one in Texas, seems to showcase tremendous demand for their technology.
2GW by H2 2026 and if demand remains as strong as it seems to be, perhaps even further capex to the 4GW mark, which is what they can achieve by some rather cheap brownfield expansion.
Applying a $4B topline per GW, the revenues can far exceed what the market currently prices in.
Thoughts on the above?