Upgrading photosynthesis
Land is a finite resource – better photosynthesis can make it abundant.
As a kid in the early 2000s, I spent several cumulative weeks of my life playing Sim City 4. Starting from an empty, isometric landscape of tiles, you construct a city, square by square, until you realize you’ll never balance the budget, satisfy your angry residents, or deal with your insurmountable trash problems. You give up and spawn a natural disaster that wipes out the failing metropolis.
I’d like to re-design the game in order to radicalize the next generation of climate activists. If I were the game developer, children would start with a giant map populated with mostly rainforests, grasslands, and shrubs. Then they’d have to click on tiles one by one to destroy the natural habitat and replace it with a cattle ranch, or a corn field, or an occasional palm oil plantation. Only once half of the map was converted to farmland could the player proceed to install a few tiles of city near a nice body of water.
The land budget of Earth
You and I exist in this absurd version of Sim City, where about half of the game’s map is dedicated to capturing photons and turning them into food. The city itself is barely visible, with our houses and offices occupying only one tile out of a hundred. The game is really about figuring out how to provide the Sims with food, fuels, and materials before running out of tiles on the grid – a game of photosynthesis1.
Earth has a fixed budget of 141 million square kilometers of land. For path-dependent reasons, not physical ones, we’ve spent most of our land budget on inefficient feed or fodder for livestock. The result is a food system that turns only 0.0125% of received photons into calories, and a carbon opportunity cost measured in the hundreds of gigatons.
An incredible amount of solar energy reaches Earth, enough to power human civilization thousands of times over. Sunlight is not our bottleneck, and won’t be for a very long time. The real constraints are the complements to sunlight: land, water, and technology.
In my view, rethinking how we do photosynthesis at a planetary scale is priority #1 for solving climate change and most of our environmental issues. This isn’t an essay about resource scarcity – on the contrary, it’s about using technology to make land abundant. More efficient photosynthesis could free up several continents-worth of land, obviating the false dilemma between material prosperity and ecological health. That’s why it’s worth scrutinizing how we do photosynthesis today, and what a better technology stack might look like.
Lessons in photosynthesis
In the rest of this essay I’ll argue the following:
In theory, manmade technologies for artificial photosynthesis are way more efficient than biological photosynthesis in plants.
In reality, plants demonstrate quite competitive photosynthetic efficiency when compared to machines optimized for low cost.
The main advantage of artificial photosynthesis is not efficiency – it’s the ability to transform otherwise unproductive land into an abundant supply of food, chemicals, and materials.
The problem with plants is not photosynthesis; it’s their other biological idiosyncrasies. For many industrial use-cases, we’re farming the wrong crops.
The photosynthetic spectrum
Earlier this year, I explored the emerging pathways for artificial photosynthesis, a somewhat loose term for converting sunlight into chemical energy without plants. Some approaches are entirely theoretical, with others already producing products that you can go try today. Some systems have no living parts, like Savor’s thermochemical fat process, or a hypothetical solar-powered formose reaction, or a cell-free enzymatic starch pathway. Some hybrid pathways, like C1-fermentation, use electricity to produce intermediate chemical feedstocks (e.g., methanol) for microorganisms.
Looking at systems above, the line between artificial and biological systems is blurry. In reality, there is a spectrum ranging from 100% biological systems, like corn fields, to 100% engineered systems, like the German coal butter factories of WWII. Microbial fermentation and cell-free bioprocesses lie somewhere in the middle.
The limits of artificial photosynthesis
If our goal is to make photosynthesis more efficient, then it’s worth knowing what the upper bound is. What are the limits of physics and chemistry?
In this section, I’ll examine a theoretical system with no living parts that uses solar photovoltaics to power chemical or enzymatic synthesis of glucose2.
Sunlight to electricity conversion efficiency
Currently, commercially-deployed photovoltaic (PV) cells have a theoretical efficiency of 34%, the Shockley-Queisser limit for single-junction cells3. If tandem cells become industry-standard, then the limit increases to 53%4.
Electricity to glucose conversion efficiency
For making sugar, chemoenzymatic pathways are the most efficient and practical to my knowledge. One widely cited approach that was demonstrated at lab scale is the ASAP cycle, which converts methanol into starch (glucose polymers). This pathway converts electricity to starch with ~35% efficiency. Critically, this value does not account for energy losses due to enzyme production, co-factor regeneration, or balance of plant.
In theory, glucose can also be synthesized through a purely chemical approach (no enzymes), with an efficiency of 31% (paper). However, this intentionally pie-in-the-sky pathway assumes a magic catalyst that overcomes the selectivity issues of the formose reaction.
In the absolute best-case scenario, combining a tandem PV cell with enzymatic sugar synthesis results in an 18.6% photosynthetic efficiency. For comparison, this is about 100x higher than an average field of corn.
The limits of biological photosynthesis
There are two notable variations of photosynthesis done by the plant kingdom: C3 and C4, which refer to the number of carbons in the intermediate molecule that CO2 is first attached to.
C4 is more efficient under current atmospheric levels of CO2 (~400 ppm) because it sidesteps a pesky issue called photorespiration, but this will actually change if we ever exceed 700 ppm of CO2 in the atmosphere5. Let’s hope we never see that experiment play out.
You might come across a few different photosynthetic efficiency numbers for plants, depending on what denominator is used and what losses are accounted for. I include the figure above from this paper, which helped clarify my mental model.
If you’re skimming this section, the two most relevant numbers are 4.6% efficiency for C3 crops (e.g., rice) and 6% for C4 (e.g., maize).
At the risk of confusing the reader, my subsequent chart also includes efficiency numbers for “pathway only” photosynthesis, which excludes losses due to respiration: 6.5% for C3 and 8.5% for C4. These are the best values you could achieve for a cell-free system, sometimes called an “artificial leaf”. Higher efficiency is possible in a nonliving system that has the same biochemistry as a plant but doesn’t require energy to stay alive.
In theory, human engineering surpasses biology
How do plants stack up against manmade systems?
In theory, tandem photovoltaics and enzymatic synthesis are more efficient than C3 plants by a factor of four. Both of these subsystems have been demonstrated at lab scale, though not at the theoretical efficiencies and not as part of an integrated system.
Nevertheless, biology is surprisingly competitive: 4.6% and 6% for C3 and C4 crops. If it weren’t for respiration, the energy tax paid to keep a plant alive, the efficiency of biosynthesis in C3 and C4 crops (“pathway only”) is within a factor of two of the best artificial system. Unfortunately, spending ATP to fight entropy is a fundamental requirement when manufacturing using living things.
Winning the photosynthesis Olympics
David Hula is a farmer in Virginia who has broken the world record for corn yield five times. His current record stands at 623 bushels per acre, over 3x the national average that sits below 200 bushels.
By my calculations, Hula’s world record corn yield achieved only a 1.13% efficiency in converting sunlight to edible grain. If we include inedible biomass as well, this number is likely above ~2% – enough to win medals, but a large margin below the 6% theoretical efficiency of a C4 crop like corn.
By compiling average and world record yields for other crops like sugar beets, potatoes, wheat, and rice, I found a consistently large yield gap between theory and practice. Our highest-yielding crop varieties, even when managed by world-class farmers, convert at most ~1% of sunlight into food and ~2% into total biomass.
The global average farm (shown in blue) typically sees yields that are 3-5x lower than the world record. This wide gap demonstrates how much crop yields vary based on factors like geography, climate, farming practices, and access to inputs like synthetic fertilizer. We should try to close these gaps, but recognize the fact that biological systems have high variance across environments.
The cost of cheapness
In practice, both plants and engineered systems struggle to live up to their photosynthetic potential. For plants, the volatile environment is largely to blame for this yield gap. For manmade systems, there is a fundamental tradeoff between efficiency and cost, and cost is what ultimately drives engineering decisions.
My go-to example of an artificial photosynthesis company is Terraform Industries, which produces synthetic fuels from sunlight. They’ve aggressively optimized their system for cost, at the expense of efficiency. At 6.5 kcf/day of natural gas output and a land footprint of roughly 5 acres (whitepaper), Terraform’s photosynthetic efficiency is about 2.11%, excluding the energy that went into manufacturing and installation.
But methane, a single carbon molecule, is arguably easier to synthesize than the six-carbon sugars in harvested corn. Optimistically, going from methane to glucose through further chemical reactions would result in a system-level efficiency of only 1.16% (calculations). Incredibly, this is about nearly the same as David Hula’s world record corn (calculations), and below the efficiency of aquatic plants like algae or duckweed (more on those later). For producing carbohydrates, the low-cost, low-efficiency Terraformer machine is no better than a highly productive corn or sugar beet farm.
So, while engineered photosynthesis has a high theoretical efficiency of nearly 20%, a practical system designed for simplicity and cost is closer to 1%. Against this practical benchmark, our commodity crops don’t look so bad.
Solar farm efficiency is overstated
A common assumption underlying analysis of artificial photosynthesis is that modern solar panels are about ~21% efficient. This number is misleading; it’s the module-level efficiency under ideal conditions, not an achievable system-level efficiency.
Solar panel efficiency drops off when temperatures get too high, or when dust coats the panels, or if the sun angle isn’t perpendicular to the panels. To maximize efficiency, panels need a tracking system that rotates them to follow the sun. But this adds significant cost, so the solar farms that give you the cheapest energy are not necessarily the most land-efficient ones. Due to the steep tradeoff between efficiency and cost, it may never make economic sense to build a solar farm that actually approaches the ~21% module efficiency.
If we examine the system-level efficiency of actual solar farms as compiled by Leger et. al., the average efficiency is about 5%. The best solar farm in the dataset, a five hectare project in Cambodia, reaches 12.86%. For our purposes, commercial solar farms are only 5-10% efficient, nowhere near the 21% that is often taken out of context.
This is a large reason why Terraform’s end-to-end efficiency is so low! After the first step, when photons are converted into electricity, 90-95% of the energy has already been lost. This is the brutal reality of turning sunlight into anything using solar farms optimized for LCOE.
Powering fermentation with photovoltaics
A number of companies, like Solar Foods and Farmless, have demonstrated microbial fermentation with electricity. This is a hybrid approach that involves turning solar energy into an intermediate feedstock like hydrogen, formate, or methanol.
While this approach can produce significantly more protein per unit area than soybeans, the energy efficiency is actually quite unremarkable. The best pathway identified in this paper, which uses methanol and the RuMP metabolic pathway, results in a practical photosynthetic efficiency of 0.8% – in the same range as oil palms, sugar beets, and maize. Other feedstocks and pathways are worse.
Why is PV fermentation efficiency so low? As before, the biggest hit to efficiency is starting from ~5% efficient solar energy. Then there’s energy lost in the feedstock production, and in microbial metabolism. Significant process energy is used in CO2 capture, heating, cooling, and downstream processing.
It was never about efficiency anyways
Would you rather replace 1 hectare of rainforest with a 6% efficient farm, or replace 6 hectares of desert with a 1% efficient factory?
This simple thought experiment reveals that efficiency isn’t everything; geography matters just as much. Even if artificial photosynthesis is no more efficient than a high-performing corn field, we can do it in places with fewer tradeoffs.
Solar panels don’t need water, fertilizer, good soil, or fair weather to operate. And as a side benefit, they’re more resilient. Not only can they survive harsh weather, but there are no pathogens that could lead to the sudden extinction of commercial solar farms or bioreactors.
About 14% of Earth’s land is classified as “barren” – the deserts, salt flats, beaches, and sand dunes that you might see on a cross-country flight. Conveniently, these marginal lands tend to have clear skies, intense solar radiation, and few competing uses. If artificial photosynthesis becomes cost-competitive, places like Nevada, Mongolia, or Australia – not known for their lush, rainfed farmland – could become the world’s artificial breadbaskets.
Factories without a roof
The vast majority of farmland is used to grow a handful of commodity crops, like corn, soy, wheat, rice, oil palms, and sugarcane.
If the purpose of a system is what it does, then farms are factories for raw materials, not food. For example, most corn in the U.S is used for animal feed, ethanol production, and separation into raw sugar and starch. For every ear of corn you see in the grocery store, a thousand others went to a feedlot or factory.
Zooming out to the Sankey level, we see that agriculture is mostly a process for producing refined raw materials, like starches, sugars, oils, proteins, alcohols, and so on. We grow complex plants just to separate them into their molecular building blocks and then assemble those back into products. If this process of assembly, disassembly, and reassembly seems absurd, it’s because it is! Growing corn to make ethanol or growing soybeans to make biodiesel is like farming dinosaurs in order to produce plastic.
As a source of industrial raw materials, the crops we currently farm aren’t particularly suitable. Even if we forgive their <1% efficiency, most crops are actually quite impractical to work with. They need water, fertilizer, pesticides, good soil, the right climate, and are prone to all sorts of diseases and natural disasters. Due to their slow biological lifecycle and unpredictable G x E interactions, breeding a new crop variety requires thousands of field trials over nearly a decade.
Finding new commodity crops
Aquatic plants are living proof that biological photosynthesis isn’t inherently inefficient, just the living package it comes in.
For example, duckweed and microalgae are widely regarded as some of the fastest-growing and most photosynthetically-efficient plants on Earth. By my calculations, the highest reported efficiencies are 2.34% for duckweed and 2.98% for microalgae (source). These are floating photosynthesis machines that forego the need for structural biomass, providing a key advantage over crops on land that must invest energy in stalks, leaves, roots, seed pods, and so on.
Duckweed and algae can both achieve high protein contents (above 50% dry weight). Most of that protein is Rubisco, which is nutritionally complete and widely regarded as a superior functional ingredient for meat and dairy alternatives. They could also be tuned for oil production (in the case of algae) or starch (in the case of duckweed).
For techno-economic reasons I plan to elaborate on in a future post, I see these prolific aquatic plants as a promising upgrade to our current commodity crops. They’re fast-growing, efficient, and minimal chassis for biomanufacturing.
A roadmap for upgrading photosynthesis
We can do better than carpeting the Earth in inefficient commodity crops and pastures that throw away more than 99.9% of energy they receive. The status quo arose through path dependence (“what else can we do with all this excess corn?”) and not first principles. It’s an environmentally destructive and unimaginative way to power civilization.
We have a broad toolbox of photosynthetic pathways that could allow us to make food, fuels, and raw materials using at least 10x less land, freeing up the equivalent of several continents-worth of land6. The following is my compressed theory of change.
Make the lowest trophic level (plants) match the price, taste, and convenience of the level above (animals). Drop-in replacements for meat and dairy will improve the effective photosynthetic efficiency of agriculture 10x or more. This alternative protein lever may seem intractable, but it is also the most impactful.
For industrial production of most macronutrients (e.g., protein isolates, starches, fats, and animal feed), replace commodity crops with aquatic plants grown on non-arable land. For high-value functional ingredients, electricity-driven precision fermentation might make more sense. These changes involve a significant infrastructure buildout, but unlock another 10x reduction in land use.
Move production of chemicals and materials to where the cheap electricity is: solar farms. Co-locate production of synthetic fuels, synthetic fats, fertilizers, materials, and their heat supply with solar farms. This solar-industrial revolution, combined with rapid data center buildout, will actually increase land use, since we’re moving from geologically-stored sunlight (fossil fuels) to “real-time” sunlight. But the expansion can be done on marginal land, and will be overwhelmingly offset agricultural reductions.
What do we do with all the land that becomes available? Let biological photosynthesis do its thing, providing a small restorative nudge where needed. Open more national parks. By letting natural vegetation regrow on the land currently occupied by animal agriculture, we could sequester enough carbon to pause global warming for 30 years. I think we could all use a break.
Appendix
Fossil fuels are also the product of photosynthesis, accumulated over geological timescales.
It is not strictly necessary to use electricity as an intermediate. For example, a light-powered enzymatic process could use photosynthetic pigments to produce electrons, like plants.
Multi-junction PV cells can surpass the 34% limit by using varying bandgaps to capture more wavelengths of light energy. But these cells are much more complex and expensive to manufacture.
I arrive at 53% efficiency based on analysis in this paper (section 5). Under one sun of illumination, you could theoretically add infinitely many junctions to your tandem cell to achieve an efficiency of ~65%. But due to diurnal and seasonal variations in the spectrum of sunlight, adding junctions beyond 8 or 9 actually decreases efficiency. With varying spectrum, the best achievable efficiency is 53% in Denver (8 junctions) and 58% in Singapore (9 junctions). I take the lower of the two values as the multi-junction efficiency upper bound.
Good systems thinking and putting numbers to hard-to-compare classes. Thank you!
What, in your opinion, are the most important obstacles to making the lowest trophic level (plants) match the price, taste, and convenience of the level of animals?