I recently exchanged some emails with Casey Handmer, the founder and CEO of Terraform Industries, which captures carbon dioxide from the atmosphere and converts it to natural gas — thus producing hydrocarbons with zero net carbon emissions. His blog is one of my favourites; I always look forward to reading what he has to say.
Here is the transcript of our email exchange.
MM: In Cryptonomicon, Neal Stephenson writes - “As it happens, Alan has become fascinated by the incarnations of pure ideas in the physical world ... Turing is neither a mortal nor a god. He is Antaeus. That he bridges the mathematical and physical worlds is his strength and his weakness”. What, to your mind, are the most beautiful examples of the physical manifestations of a pure math idea?
CH: Computation in general is pretty cool, and widely underappreciated. Any computer can simulate any other computer is kind of mind boggling.
MM: From a first-principles standpoint, multi-junction solar cells seem like the obvious successor to monoPERC and TOPCon for breaking past the Shockley-Queisser limit. Yet they remain a niche, space-only technology. In your view, what is the single biggest physical or manufacturing bottleneck preventing their terrestrial dominance, and what non-obvious breakthroughs are needed to unlock a cost-competitive learning curve for them?
CH: I don’t know much about emerging PV technology. The biggest obstacle for PV in general is permitting reform. For perovskites and other more exotic PV technology, the roadmap is clear - get to positive unit economics and then ride a cost curve that is at least as steep as the incumbents.
MM: The 2023 Terraform white paper (version 2) says that synthetic fuel will use 80+% of all solar generation long term — this implies that in the long term, converting photons to molecules will outcompete storing electrons in batteries (and discharging when needed). Could you walk me through the reasoning for why you believe >80% of solar power will be used for DAC fuels and not to charge batteries for battery energy storage?
CH: The leading order term is that chemical synthesis is intrinsically inefficient, so that's a 3x boost right there.
MM: What is the first principles reason for why 1 MW solar module produces 6,500 cubic feet of CH4? In that chain from solar panels to electrolysis to the Sabatier reaction, what are the key bottlenecks that prevent an even greater throughput of CH4 per MW of solar generation?
CH: 1 MW solar array produces about 6 MWh per day (sun is underground half the time, and pointing in the wrong direction another half of the time). 6 MWh is equivalent, energetically, to about 20,000 cubic feet of CH4, but the synthesis process is not very efficient. First, most of the power goes to making H2, at perhaps 60% efficiency. Then half that hydrogen just gets converted into water instead of methane. But you get a little bit of efficiency back because the carbon atom that’s released to become CH4 from the O2 that becomes water, also has positive enthalpy of combustion. So all up, about 30-35% efficiency. There are known ways to push this into the high 40s, and physically 60% efficiency is probably possible - but it would not be close to cost optimal.
MM: There are several emerging trends in AI data centers - e.g., bitcoin mines being converted to AI data centers, direct liquid cooling to the chip, orbital/in-space data centers, modular/containerized data centers etc. Which of these (or any others) do you find most promising, and why?
CH: I think the most promising is direct DC HV systems, including racks powered in series. All the other stuff is either too exotic, or I’m not qualified to speculate, or both.
MM: In your podcast with Dwarkesh, you spoke about the potential for integrated solar with chips (power with compute), since both are fundamentally made of the same material — silicon — albeit at different purity levels. Could you paint a picture for what such an integrated solar-GPU might look like and how TSMC might build it?
CH: That’s a hypothetical about the distant future state of humanity. With our current manufacturing tech, the chips cannot be larger than the lithography field of view, which is too small. So let your imagination run wild.
MM: Like yourself, I am majoring in physics and math. What are the most profound and elegant concepts you learnt in college that fundamentally rewired your thinking? Could you convey that elegance and beauty to motivate why one must study them? (E.g., in freshman year I found Fourier transform and Tychonoff's theorem the most beautiful).
CH: Probably the earliest “aha!” moment I can remember was showing that central inverse square laws give rise to “orbits” broadly construed that map perfectly to the conics, which is to say projections of cones and planes. Something quite profound about that duality, and it provides a strong hint of similarly deep principles, such as holography, that I encountered later.