log in

So you want to build a carbon capture company

Hacker News - Tue May 4 07:53

Would you like to win one hundred million bucks from Elon Musk? Carbon capture (CC) is all the rage these days, with dozens of companies springing up to remove CO2 from the atmosphere and help stabilize the climate.

I am not an expert on carbon capture but I do get asked about it from time to time. As a public service, therefore, I am offering the following rubric as a means to organize our thoughts, refine our strategy, and champion quantitative rigor when it comes to developing and evaluating a wide variety of carbon capture systems. 

Is our carbon capture scheme any good at all?

Let’s examine our hypothetical CC machine from two angles: physics and finance.

Physics

Is our machine secretly a perpetual motion machine?

In a previous life I spent a few years designing maglev systems and quite often would see concepts from other designers whose performance was too good to exist in the real world. If the system has negative drag, it is a perpetual motion machine. 

If the system concentrates CO2 for less energy than releasing it back into the air, it is a perpetual motion machine. If our machine compresses a gas stream with no expenditure of energy or generation of waste heat, it violates the laws of thermodynamics. Perpetual motion machines obviously do not exist. Check the math! 

What do we know that no-one else does?

What’s a non-obvious controversial true fact? How does our system exploit this?

How much energy does our system actually use?

CC systems sometimes use thermal cycling of sorbent beds or electrochemical separation to increase the concentration of CO2 from ambient 420 ppm to close to 100% CO2. Does our system require lots of electricity or thermal energy to operate? How is it being provided? 

If the system is electrochemical, does it use more or less power per mole of captured CO2 than aluminium smelting? This is about 1500 kJ/mol. Is the power provided high current, low voltage? Do we have a homopolar generator handy? How much copper does the power system need? If we need X electrons per molecule of CO2 at a cell voltage of Y, this works out to be about X*Y*95 kJ/mol. How close is our current system to this limit?

Does our system net reduce CO2?

If our CO2 capturing system works by weathering Calcium Oxide (quicklime) that is produced by thermal calcination burning natural gas, it will emit more CO2 than it captures over a lifetime. Whoops!

More generally, how many years of operation are needed to offset CO2 emitted during production?

How are we thinking about theoretical limits?

The Gibbs entropy of CO2 dissolution in the atmosphere is about 19.4 kJ/mol. This is not much energy, which is why no-one generates power by leveraging the osmotic gradient of concentrated CO2 in the atmosphere. Does our system get anywhere near this? Does it have to? Can it? If we’re doing electrochemical separation, how are we capturing ohmic heating and viscosity as limits to our ultimate efficiency?

Is electrical efficiency even a major constraint to our system? Does it need to be efficient, and what’s the opportunity cost for increasing efficiency by 1%? If electricity gets 1% cheaper every year, is that equivalent to a free virtual 1% increase in efficiency?

Is our machine actually concentrating atmospheric CO2?

Our machine has flashing lights and a pipe that emits CO2 at a million parts per million. We’re good, right? Well not quite. Does the machine contain carbon? Are we sure we’re not accidentally combusting part of our machine? How sure? 

I think the gold standard here is that CO2 produced by concentrating atmospheric CO2 should have a radiocarbon age of zero (relatively radioactive) while CO2 derived from, say, accidental electrolysis of a mined carbonate salt will be very, very old with no radioactivity. Testing samples for Carbon-14 requires a mass spectrometer. There are numerous labs in the US which will perform tests for a few hundred dollars, though they generally have to convert samples to graphite first. 

Carbon dating isn’t foolproof, however, as organic sources of carbon, such as vegetable oils, wood, or charcoal, are also radiocarbon young. So if our machine uses crisco as a lubricant, we should double check the math, and also our life choices.

Can we defend our results?

Do we understand our test system? Have we quantified every aspect of its operation? If we’ve produced a video showing how it works, will it confuse potential investors? Are the key points obvious? Can someone watching the video easily imagine themselves building the same system and running the same test? Are test information data and results documented well enough to enable independent verification? Do we have a good understanding of what a well documented experiment looks and feels like, or do we need to go and read a biology paper or two?

There are plenty of very confused people out there in the area of CC, and we need to normalize a high level of rigor in our approach to documentation. We’re not planning on posting our trade secrets online (or are we?) but it’s unreasonable to expect investors to part with their money on a hope and a prayer.

Can we scale it?

Are there any fundamental physical limits on deployment? If we’re going to capture 10 GT of CO2 a year by planting trees, how much water will we need to irrigate them? What are the fundamental constraints on scaling? Capital availability? Indefinite returns on investment (ROIs)? Rare reagents? Flaky co-founders? Utility energy supply? Legal status of carbon taxes?

Is our CC machine ready to escape the lab?

The AC propulsion prototype used at Tesla in the early days was notoriously unreliable, using dozens of analog op amps to drive an AC induction motor. History is littered with projects whose costs were unsustainable because they were insufficiently mature to be put into production.

Do we have a desktop demo we can show people? Does it actually work? Is it quite clear which parts are hacks and which parts actually matter? Is it safe enough to put in a room with members of the public?

Is the tech ready to put into production? Can we hand the prototype to an average mechanical engineering graduate and say “make 10,000 of these” and have reasonable confidence that they’ll come off the line functional, reliable, and with decent yield? Have we worked out the bugs before the major capital investments, or is it still a science experiment?

Finance

What is our CO2 price?

Can we produce CO2 for $1000/tonne, $100/tonne, or $10/tonne? Where are we? Where do we want to be? Where do we need to be? How do we stack up against the competition? How credible is our path to improvement?

How expensive is our CC machine?

What is the CAPEX structure? How many tonnes of CO2 does the machine have to capture to pay only for the machine, less opex, financing costs, depreciation? How long does that take?

If our machine captures 1kg a day at a $100/T price point, it will earn $36.50 a year. If our machine costs $500 to build, it will take 15 years of operation just to cover construction costs. $500 for parts and labor falls somewhere between a nice cake and a very basic dishwasher in terms of overall scale and complexity. A half decent technician should be able to assemble half a dozen per day, which means our production rate must be at least 1500/year. Even then, total additional revenue will be about $50,000 which is barely enough to put one person through grad school.

If CAPEX is amortized over a decade or three of operation, how are we estimating our capital costs? Do we expect/rely on congress to underwrite big loans to ensure low interest rates, like with home mortgages? Are we going to become the underwriter for our customers’ loans to buy devices from us? How are we going to diversify risk in this sector given that many risks (technology, regulatory) are extremely correlated?

Or, can we make back the cost of construction in a few months or a couple of years, and thus access short term financing or even self-finance?

How quickly does our machine wear out? Do we have to depreciate it more quickly than we can pay it off? Are we going to self-cannibalize with version 2 and strand our early customers? Are they okay with that?

How expensive is our CC machine to run?

What are the operating expenses (OPEX)? Do we require labor for maintenance? What are the machine’s expendables, such as reagents, valves, fittings, pumps, electrodes, software?

How do operating expenses compare with the amortization schedule for CAPEX? Are we spending more on operations than CAPEX payments, and thus could justify adding complexity to the system to reduce ongoing expenses? Or is the machine so reliable, so set and forget, that NASA will use it for atmospheric regulation on a Moon base? 

Are we deploying in our backyard or in the middle of the desert somewhere? How do we access and support customers with hardware in remote or difficult to access places?

Energy, again?

Are energy costs important to the financial picture? Ten years ago, electricity costs made green hydrogen (produced by electrolysing water) prohibitively expensive compared to blue hydrogen (derived from natural gas via steam reforming). Today, solar PV electricity during peak hours is >10x cheaper. How does our business model and system optimization shift if electricity becomes more expensive, or cheaper, over the lifetime of our machine?

Is our process energy intensive? Is it comparable to refrigeration or electro-refining of magnesium? Could we be investigated for running an illicit growing operation or a data haven? 

How sensible is our supply chain?

Does our machine depend on any unusual materials? What can’t I get from McMaster-Carr, or Ali Baba? Or Silk Road? Is our supply chain fungible or do we depend on the business and good graces of a single supplier in outer Mongolia? Do we absorb CO2 with amines, zeolites, or MOFs? How expensive are these specialty materials? Are we related by blood or marriage with a lab that can actually make them? Can they scale production as quickly as we can scale business, and at what marginal cost? MOFs cost WHAT exactly?

Does our Bill of Materials contain anything (ANYTHING) considered more than usually toxic or requiring special handling? Any plutonium? Prohibited substances? FOOF? Piranha solution? Do we need certified technicians to do the work? Can we afford their fancy insurance? Are we going to get a visit from the DEA or DHS?

Does our process depend on the availability and good graces of one or more highly trained PhDs? Do we have a talent retention plan? How exotic is our process?

Do we need miracle materials to work?

Does our system only work with 99.999999% pure anything? Contamination: no problem, reduced efficiency, or spontaneous combustion? Will our catalyst get destroyed by exposure to common air pollutants, such as water vapor, oxygen, or the smell of pad thai?

Does our system expend its catalysts? Are they actually secretly consumables? Do we have a plan to supply, service, and replace stuff that we weren’t planning on breaking? How much cobalt do we actually need per tonne of CO2?

Do we need a miracle of scale?

Everyone knows that cars are only relatively cheap because of enormous complex, expensive tooling that enables a few hundred thousand to be made, exactly the same, every year. 

Does our CC machine have the same issue, where we can’t get CAPEX down to a reasonable level until we’ve designed and built a million square foot alien dreadnought fully automated lights out factory for it? Why can’t it be assembled like LEGO? Have we personally ever built a huge automated factory before? Is this expertise actually our value-add? Have we considered an “alien dreadnought factory as a service” start up instead?

More generally, is there a critical scale below which our system makes no sense? Can we justify the economies of scale or are we waving our hands because our system costs more to build than 20 years of operation at $1000/T can justify?

Do we have a revenue stream?

Or do we need, long term, to rely on coordinated legislative action by a few dozen national governments to emplace a reliable carbon tax/dividend so we actually have an infinitely deep, zero elasticity market to sell our CO2 to? 

Where does our concentrated CO2 go? Turning into fuel? Plastic? Carbon black? Graphite? Cement? Underground? Carbonated drinks? What is the annual capacity of these markets? How much of that can we capture? How much of that can we expand?

If we’re selling our CO2 only to PepsiCo, it goes back into the atmosphere very quickly. Do we have a plan to generate a more durable store of CO2?

Who is willing to buy our CO2, in what form, how much, and for what price? What does our business look like with this market saturated? For example, if we’re selling 1000 T/year for deep well injection at $100/T, our business has a revenue of $100k/year. Is that enough to support the team?

Where is the value generated in our business?

If we’re building a CC machine that must be amortized over 20 years, we’re selling very expensive widgets to debt-happy customers, and hopefully lots of them. What is expensive in that machine? Where are we adding value?

Let’s say we’re building CC machines that use a swing cell with zeolites, similar to the life support system on the International Space Station. A major cost of these systems is new zeolites. To reduce cost and improve quality control, we have decided to vertically integrate manufacturing of zeolites, and by doing so have improved the cost by 20%. As the zeolites were about 90% of the initial CAPEX of the machine, more than 95% of our company’s value-add is now in making zeolites. So are we really a zeolite factory in disguise?

More generally, if long term industrial scale CO2 capture does turn out to depend strongly on mass production of an otherwise exotic material, just as the computer industry has depended strongly on photolithographic etching of insanely pure silicon crystals, does verticalization of the industry make sense? Where are we starting on this value chain, and where do we intend to end up? Chemical supply as a service?

Concluding thoughts

What should we be thinking about?

Have I missed any obvious questions? Any un-obvious ones? Does this help us understand what needs to be done?