The implausible Bloom of an energy equipment company in Silicon Valley

How I came to invest in Bloom Energy, what's going on and why Peter Lynch was right.

“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 $2,000. Not only that, they put in another $2,000 each year for the next five years. By 1972 the fireman was a millionaire”

— Peter Lynch | One Up on Wall Street

I’ve always wondered if that was an apocryphal story since Lynch doesn’t quite give a reference to the origins of that story. Some of you who know me from (the platform formerly known as) Twitter, know that I subscribe to his philosophy and invest in Semiconductor stocks with some modest success. But that’s a field I know at least something about. Can you do this in a completely unrelated field by just being curious, observant and following your blood hound nose as he calls it?

Re-discovery

I was carpooling down the 237 one day with my friend driving. As I was idly gazing at the scenery flashing by, I did a double take uttering - “I wonder how they have the money for that?” My friend asked me what I was talking about. I pointed at the shining new looking glass building adorned with a large Bloom Energy logo. I sort of knew of the company from the past, had seen their non-descript low rise former location in a different part of the South Bay and had expected them to be struggling. This was sometime in 2019, but it turns out they had an IPO (Initial Public Offering) earlier in 2018 which explained the new digs. I made a mental note to investigate further.

Some of you from the Bay Area might remember that there was a fair bit of hype about Bloom Energy back in 2010. They had operated in stealth mode till that time when there was a grand coming out replete with a CBS “60 Minutes” feature, retired General Colin Powell as a celebrity board member and grand visions of changing the world with clean cheap energy including home units. Bloom made a Solid Oxide Fuel Cell (SOFC) based “energy server” running on Natural Gas that the company claimed could produce energy cheaper (given some conditions) and cleaner. More about the technology later. VC (Venture Capital) firm Kleiner Perkins and John Doerr were major backers, and they had first deployed their technology at Google in 2008. At the time I had done some cursory reading up on them due to all the hype. But there were plenty of skeptics. One example from a reporter in Forbes -

“Are we really falling for this again? Every clean tech company on the planet says it can produce clean energy cheaply, yet not a single one can. Government subsidies or mandates keep the entire worldwide industry afloat. Hand it to Bloom, the company has managed to tap into the hype machine like no other clean tech company in memory.”

Note that the firm was named Bloom “Energy”. The plan at the time seemed to be to sell power in a long-term power purchase type agreement to amortize the higher cost per kW (Power|CapEx) of the equipment with lower fixed cost per kWh (Energy|OpEx). This worked better in places with higher power prices (as always in California), cheaper gas costs, some of kind subsidy/clean power mandate or a combination of these. Why was it happening at that time? Well, we had a big energy price spike in 2008, then the Financial Crisis and about 10 years (typical VC fund life) since the founding of Bloom. I’ll bet that John Doerr was getting a lot of pressure to get some kind of exit. Long story short, the hype and momentum fizzled out later. The company was muddling along in niche applications where it made sense and the world mostly forgot about them. I happened to drive past their old location a few times and noted “oh, they are still alive” but mostly forgot about them as well. Till that fateful day in 2019.

As an aside, I had a few months before, opened a “fun” portfolio account to try investing in non-semiconductor stocks outside of my “serious” money main account. I decided to buy a monitoring position in BE 0.00%↑ to follow the story. Incidentally, along with that GE 0.00%↑ was one of first three stocks I bought and while GE has done well (perhaps a story for another day), its spinoff GEV 0.00%↑ is a 16 bagger with some similar themes. It’s funny how things are related sometimes. Edit: I finally did write about GE:

Sometimes bet on the jockey, not the horse

Following the story

I started investigating the company and it looked not as bad as folks made it out to be in 2010. The fundamental technology had a unique set of benefits not always available with other power sources. I talk about this later, but I’d encourage you to skim that section using the navigation menu on the left and come back. But the cost (esp. initial CapEx) was higher, and the company wasn’t profitable although getting closer. Revenue was increasing although not a high growth story (yet). So, this was a watch and follow the story kind of stock.

And I’ve followed the company ever since, listening to earnings calls and releases like an amateur analyst. As far as I could tell, the IPO wasn’t a great success and mostly seemed to be a way for early investors to get some money back on their investment. There was a brief pop after the IPO but clearly folks didn’t seem to be as excited about any kind of Energy business in the age of SaaS and Cloud computing. There were only about 3-4 actual sell side analysts who seemed to follow the company with any regularity.

I slowly came to the realization that there wasn’t a single other SOFC provider in the world in production for onsite power. There are a couple of other small US manufacturers, but they do mostly backup type applications.¹ They had a huge first mover advantage with the cumulative experience of having learnt from all the systems they shipped. Energy and power production is a very difficult business to break into and productize new innovations.² Almost all other SOFC companies fell by the wayside and there’s just one other company that’s trying to commercialize it for onsite power but is far behind. It’s practically a miracle that Bloom even survived. The only reason is that VC money propped it for quite a while since John Doerr/Kleiner Perkins at one point in time were into “green tech” (and criticized by others for wasting money). Had it been anywhere else in the world or even anywhere else in the US but the Bay Area it would be dead as a dodo now. Any guesses as to where the others developing SOFC tech were? Even the one distant competitor is being propped by a Korean company since that is one market where Bloom found early success and I guess they would like an alternative.³

Near death

As fate would have it, sometime in Sep 2019 Bloom came under a short seller attack by Hindenburg Research. Yes, the same Hindenburg Research that later took down Nikola Motors. The main allegations were that Bloom wasn’t profitable, had a lot of debt, weird questionable accounting, its energy wasn’t really clean, and its fuel cell stack life was lower than promised leading to higher costs.

The debt, profitability and accounting were real issues but surmountable if they kept improving the product at the rate at which they had been. Not being very clean was a silly argument since they compared it to the grid which had other clean sources mixed in. Fact was that Bloom was still the cleanest way of producing energy from a fossil fuel where CO₂ is unavoidable. The point is that due to the lack of combustion a whole bunch of SO_x (sulfur oxides), NO_x (nitrogen oxides) and other pollutants are avoided. In any case the grid itself needs natural gas plants to operate which is the real comparison.

The stack life issue was the most worrisome where they claimed that it was only around 3 years. However, they seemed to have missed that all existing competitors could barely manage 1 year. Bloom started out with 12 months stack life way back in 2008, which quickly improved to 18 months and somewhere near 3 years for the generations that Hindenburg likely looked at. Currently Bloom claims about 6 years for the stack life so they have made steady progress there. Stacks can and are replaced easily in the field as different modules can be shut down as needed. In comparison ICE engines have significant maintenance needs if run continuously for long, as any Generator Set owner knows. Gas turbines also need periodic maintenance. SOFCs have very few moving parts so it’s just basic physics and engineering that the maintenance needs are less. In fact, there are niche SOFC backup power manufacturers who do a decent business just based on this lower maintenance (for remote unmanned locations).

If the company had not died in the midst of all the Cloud/Mobile/SaaS boom in Silicon Valley and actually managed to IPO, this kind of attack could wound it badly but not quite kill it, was my assessment. They had some genuinely unique technology despite all the financial warts, which would be worth something to someone. The stock did take a beating and dipped below 3 dollars having IPO-ed at 15 and a high of 38. The risk vs. reward looked good to me. I decided to buy a medium sized position in my “fun” portfolio -

Green Shoots?

Right on the heels of the short seller attack came the COVID-19 pandemic and it looked bleak, but what was interesting is that the market was more worried about the short seller attack than the pandemic and it never touched the absolute lows seen then. Probably few sellers left. I bought some more at the worst of the pandemic crash in the market. This is a huge underappreciated advantage that individual investors have. It would be very difficult for professional money managers to pull the trigger in a situation like this, what with the career risk and others questioning the move.

Bloom was even helping refurbish ventilators at this time. I’m not quite sure if they had anything else to do but this was a nice charitable gesture.

Soon after they pivoted to adding an Electrolyzer solution. SOFCs can easily be reversed to make a Solid Oxide Electrolyzer Cell (SOEC). In fact, this was part of the background of the Bloom founders as scientists at NASA with the technology. SOECs are more efficient as an electrolyzer compared to PEM (Proton Exchange Membrane) electrolyzers. In addition, they can take advantage of heat, say waste heat in a nuclear reactor, to produce so called “Pink Hydrogen” (with a test at Idaho National Labs to prove the concept). The same applies even better to a Solar Concentrator based application (partnership with Heliogen, went nowhere and since acquired by Zeo Energy) for “Green Hydrogen”. Batteries work for some problems as they get cheaper, but they don’t work for all applications like heavy industry and heavy transport where liquid fuel’s high energy density or specific energy (energy per unit of weight) based compactness or combustion heat is important. Hydrogen does have a future if society decides that being green is important. But as abundant cheap renewable energy is still not widespread, I consider this more of a futuristic scenario.

Regardless, the market got quite excited about all this during the Covid lockdown SPAC and meme stock driven mini bubble of 2021 and the stock breached its post IPO highs. While on paper I had decent gains, I decided not to sell as I believed that the company had a better future with the on-site power market if they could keep reducing costs. The one good that did come out of this is that they managed to refinance their debt as lower cost longer duration “green bonds”.

The AI inflection point

The AI boom has pushed power demand of Data Centers into uncharted territory. I won’t rehash all that as you can find plenty of coverage about it. Utilities are typically slower-moving organizations that are unique in American capitalism for being “regulated monopolies”. For the privilege of being allowed to be a local/regional monopoly (for the scale benefits needed for a highly capital-intensive business) they agree to be regulated by the government. On top of this constraint, there are backlogs for gas turbines, transformers etc. due to all the demand. The last time the utility industry saw a growth phase like this was probably more than half century ago during the sunbelt migration and air conditioning growth. I’m not sure there remains much institutional knowledge on how to handle a rapid growth in power demand.

Bloom is well positioned to supply demand when others just cannot on time or at the same costs as before. First, they announced a deal with the utility AEP, then came the deal to supply power to an Oracle DC in 90 days (actually delivered in 55 days as per latest earnings call) and finally the deal with Brookfield as the infrastructure power provider for their AI projects. Why is all this happening? In addition to the unique aspects of Bloom’s solution that I talk about later, there are a few more specific ones like the ability to provide Direct Current (DC) power. A big shout out to Irrational Analysis who covered Bloom’s presentation at OCP (Open Compute Project) Global Summit about this. He’s the one person who has done more than anyone else to make my thoughts better known to other folks on the platform formerly known as twitter.

BTW this is the most frustrating part of following the company. They seem to make these kinds of presentations and don’t make them public even though they should know better that they are obligated to. Without Irrational Analyst even someone like me who has followed them for a while, wouldn’t know (even if I could guess a lot of what they would say). Edit: It seems the material is available elsewhere, but not on Bloom’s site. In addition, they have hosted seminars for which they don’t share the slides.

While I don’t know exactly what the Bloom speakers said and Irrational Analyst did not quite elaborate, there was this cryptic image -

I believe what it represents can be explained with the following Ragone plot which shows SOFCs combined with capacitors and/or batteries can cover the entire gamut that gas engine/turbines can.

Edit: Irrational Analyst shared that what they said was a highly tortured metaphor that the AI rabbit is constrained by the tortoise utility! I still think my analogy is better 🙂

Unique benefits of Bloom’s SOFC solution

I mentioned earlier that there are some unique benefits of Bloom’s solution. Let’s unpack that a bit. Note that I’m mentioning this before a discussion of how the technology works, but you can read that section first to better understand how these benefits might come about.

Parameters where Bloom is better than most other fossil-fuel based solutions -

• Emissions. Since it uses an electrochemical reaction instead of combustion, it produces just trace amounts of gases other than CO₂. No combustion-based solutions like turbines or ICE (Internal Combustion Engines) can match that. This also helps with any kind of Carbon capture solution as it’s hard and expensive to separate Nitrogen from typical exhaust outputs. In comparison Bloom’s SOFC produces 95% pure CO₂ which makes it a lot easier to do real economic Carbon Capture.

• Availability. Being a modular Lego like approach, Bloom has built-in redundancy and distributed production which naturally leads to high availability.

• Land Use (and curb appeal). While bloom’s footprint is similar or even lesser than gas turbines and ICE solutions (depending on what exactly you are looking at), it can be double, triple stacked vertically which reduces the land use significantly. This can be quite desirable for dense urban locations along with its lower noise and emissions levels. It also helps with NIMBY-ism against on site power for a Data Center. Bloom has demonstrated this in one of its most important markets, densely populated South Korea (5th most dense in the world, barring small countries) -

  

• Time to power. Even without the well-known backlogs for gas turbines, Bloom has an advantage since its modules can be easily assembled on skids ready to install. Their stack production is not that capital intensive and it’s easy to scale up in comparison to the significant investment needed for gas turbine manufacturing.⁴

• Operating cost. With the improvements in their latest gen servers and their efficiency, the pure operating cost per kWh (not accounting for CapEx) is lower. While the efficiency is on par with the best gas turbines, there are cost savings in maintenance and avoiding electrical distribution infrastructure for an onsite installation. This was the original competitive vector for Bloom with the Power Purchase Agreement lease type arrangements.

• Local CHP (combined heat and power) Applications. While Combined Cycle Gas Turbines (CCGT) can do CHP as well, these are typically very large power plants and are less likely to be near places which can use the heat economically without some kind of transport infrastructure. With Bloom, if you are doing on site power it’s easier to just tap the heat output if you have a need and achieve close to 90% efficiency. Note that this can be used for absorption cooling in a Data Center. While Bloom has not publicized any such DC installation, they have done CHP installations and utilize absorption cooling at their Fremont, CA manufacturing plant -

Parameters where Bloom is better than gas turbines -

• Water consumption. Bloom’s SOFCs only need water for startup to initiate the reforming reactions that convert Natural Gas to Hydrogen (see the technology section). Once startup is done, it generates enough water from the electrochemical reaction of Hydrogen with Oxygen to not need any inputs. This can be a huge advantage in water constrained areas. Of course, ICE solutions are similar (only need water for engine coolant) but more polluting.

• Ramp rate. Bloom’s SOFCs can load follow and ramp up output very well. This can be done because modules can be kept in hot standby mode using the heat from running modules to quickly turn them on as needed. You can add capacitors for even more responsiveness.

Parameters where Bloom lags behind the best gas turbines -

• Levelized cost including CapEx. This was the original disadvantage and while costs have improved quite a lot, it certainly was more expensive than gas turbines before the AI related backlogs and cost increases. Bloom is on the order of $3K per kW of capacity. But due to all the demand and backlog, large turbines are now reaching close to that cost without all the other advantages, Bloom can be very competitive depending on the application.

• Electrical Efficiency. While raw figures are on par or even better than the best turbines, there has to be an allowance made for stack life and needing to replace these as the efficiency drops with age. As stack life improves in newer generations this should improve, but right now an average value during the serviceable life will be a bit lower than the best Combined Cycle Gas Turbines.

• Startup time. SOFCs take up to an hour for startup due to the need to build up the heat and start self-sustaining reactions. This can be mitigated with the hot standby mentioned in the ramp rate but the very first startup will still be slow. Batteries can be used to mitigate this if really needed and are used in remote backup type applications.

Technology

Let’s take a look at the technology and how it works. This section can be skipped for a first reading and looked at later to better understand why SOFCs and Bloom’s implementation have unique advantages.

A fuel cell can be thought of as similar to a battery in that it produces electricity, but the power is produced via a continuous stream of fuel instead of stored charge. What’s more the fuel is used in an efficient electrochemical reaction like in a battery and not a combustion process (like in an engine or turbine). There are various other kinds of fuel cells apart from Bloom’s SOFC (Solid Oxide Fuel Cell) like -

• PEM(FC) (Proton Exchange Membrane)

• AFC (Alkaline Fuel Cell)

• PAFC (Phosphoric Acid Fuell Cell)

• MCFC (Molten Carbonate Fuel Cell).

Of these SOFC and PEM seem to be the most mature/deployed solutions with PEM being preferred in transportation. However, only SOFCs, PAFCs and MCFCs can work with natural gas instead of Hydrogen/Ammonia in the others (producing Hydrogen is still not cheap and will need abundant cheap renewable sources before it becomes reality, so for the purposes of this discussion, we are going to ignore it). MCFCs are large and complex with corrosive materials and haven’t become mainstream. PAFCs use expensive catalysts and have lower electrical efficiency. That leaves SOFCs as the one unique fuel cell type that can work with Natural Gas (and Hydrogen) in an efficient manner.

The basic structure of an SOFC as shown above consists of a solid electrolyte (usually a ceramic) positioned between an anode and a cathode. Fuel is delivered to the anode and oxidant, typically air, is delivered to the cathode. The electrodes are solid porous structures that allow the fuel and air to diffuse to the electrolyte and the products of the electrochemical reaction on the anode side to diffuse away from the electrolyte. The electrolyte conducts the oxygen ions formed by the electrochemical reduction of molecular oxygen from the cathode side to the anode side of the SOFC. Fuel diffuses through the anode to the anode/electrolyte interface. Here it reacts catalytically with the oxygen ions, releasing electrons that are transported through an external circuit, producing electricity.

The diagram above shows Hydrogen, but the principles are similar for Natural Gas. For those of you with a Chemistry background, the difference is that there are reforming (wet/dry) and water-gas shift (WGS) reactions that converts the Natural Gas to Hydrogen and Carbon dioxide. After conversion to Hydrogen, the process is the same as using Hydrogen as a fuel.

The high temperature operation of SOFCs makes this possible and is key to its fuel flexibility.

Individual cells are linked with a metallic interconnect in electrical series to increase voltage and power and can be stacked to achieve an optimal stack size. A number of stacks can be arrayed within an enclosure to form an SOFC module. Single or multiple modules may be aggregated and integrated with subsystems to deliver air and fuel, to recover heat, and to convert the generated electricity from direct current (DC) to alternating current (AC), forming an SOFC power system. The above illustrations from Bloom shows how cells are stacked up and combined into a basic block which they call an Energy Server.

Bloom’s Energy Server is currently rated at 325 kW, and is comprised of various modules, all of which are typically factory-packaged and mounted on a pre-wired skid that includes all necessary interconnections—electrical cabling, piping, and auxiliary equipment—for simplified installation and deployment.

The key to understanding Bloom’s future potential

“Not hearing is as nothing compared to hearing; hearing is not as good as seeing; seeing is not as good as knowing; knowing is not as good as doing.”

— Xunzi (Confucian thinker)

The key to Bloom’s future potential is to understand their steady progress on bringing the costs down via the experience curve effect. They started out with a cost greater $10K/kW. Now they are down to about $3K/kW, and its been a steady more double-digit improvement per year as shared in earnings releases over the years.⁵ They are the largest player in SOFCs who have shipped regularly in large enough quantities to ride that curve. Unfortunately, they have recently resorted to not sharing this metric arguing that we need to look at other parameters than just cost.

Even if the above slide from Bloom is an optimistic take, it certainly is in the same ballpark as batteries and solar. It all boils down to if you believe they can continue progress on the experience curve and even drop costs down to $1.5K/kW. At which point a whole world of applications open up. Data Centers are just a part of that potential world.

Interestingly, they claim the learning rate of their SOECs, which is earlier in the cycle, is close to 30%. That would be quite remarkable if they can maintain it and ship in volume.

Conclusion

What’s shocking to me is how few have known about it or had discovered it till recently. A Bay Area resident with a clue could have bought it dirt cheap during the short seller attack by Hindenburg Research like me but that takes balls of Tungsten Carbide if not diamond. Also proof that Peter Lynch was right!

The one and only big money exception so far is Paul Wick of Columbia Seligman who has boots on the ground presence in Silicon Valley. He seems to also have bought around the same time. Just to be clear, I have no connection to him or his firm. I like to keep tabs on other owners of the stock.

Yes, Bloom’s solution is more expensive right now but that depends on what kind of application and needs you have. If Bloom manages to use this AI boom/bubble to make its technology more mainstream and continue the learning curve cost reduction, then they may very well achieve the shock of breaking into a mature established sleepy industry with something new. We shall see.⁶

Disclaimer: None of this is advice. I have done my own original research over the years and invested. I would encourage you to do your own research and make your own decisions.

1. See the later note about other SOFC applications -

2. If you have an engineering background or know thermodynamics, then lookup the Allam-Fetvedt Cycle and see the trials and tribulations the VC folks first and then the public company Net Power is going through to commercialize it. And that is with big companies like Occidental Petroleum, Baker Hughes and Constellation Energy involved. This is super hard stuff that can take decades. They would be a natural competitor to Bloom, but it’s comical how tough it has been for them and how long it might take. Gas turbine-based power IMHO is almost Rube Goldberg-esq in comparison to the simplicity of modular SOFCs.

3. See the note about competitors -

4. William Martin Keating wanted to know more about Bloom’s ability to scale. My reply, including possible connection with semi fab manufacturing, which I published as a note -

5. See my later replies in a SemiAnalysis post about AI and onsite power for more insights on the experience curve -

6. If you are wondering about valuation -