There is a graph inside the vertical-farming industry that almost nobody reads correctly. In Japan, where the government has surveyed plant factories for years, the share of operators that are profitable or breaking even splits cleanly by how the light gets in. Sunlight-based greenhouses: about 73%. Fully artificial-light farms — the stacked, LED-lit, climate-sealed towers that get all the magazine covers — about 43%. The whole field averages around 59%.1 The more futuristic the facility, the worse the books. Technological intensity and solvency run in opposite directions.

I spent a while thinking the explanation was a food story: indoor farming is a bet on feeding cities from inside cities, and the bet is not paying off. That framing is not wrong so much as too polite. When I put it against the primary literature, the honest version turned out to be an energy story, and underneath that, a thermodynamic one. And it embarrassed a hypothesis I had walked in with.

The number that ends the argument

Start with a single figure from a 2025 back-of-the-envelope paper in Plant Physiology by Lovat, Noor, and Milo. In a fully artificial-light farm, the overall conversion of electricity into edible dry matter runs at roughly 1% to 2%.2 Not the LED. Not the wiring. The whole chain, electricity in, food out.

Run that forward. At 2% efficiency, producing a kilogram of dry plant matter takes on the order of 250 kWh. Even at a generous four cents per kilowatt-hour, that is about $10 per kilogram of dry matter — before anyone is paid, before the building is amortized. Meanwhile the farm-gate price of the world’s staple dry crops — wheat, rice, soy — is under $1/kg, and can be as low as $0.10/kg.2 To grow a staple grain under lights costs, in electricity alone, ten to a hundred times its market price.

That is the whole indictment, and it is why “just wait for cheaper LEDs” is a category error. I had assumed the LED was the frontier where the gains would come from. The paper corrected me twice. First, the binding constraint is not the lamp; it is photosynthesis itself, which converts light into biomass at a low single-digit percentage no matter who supplies the photons. Second — and this is where I have to be honest against my own instinct to overclaim — the biological ceiling is not frozen. The paper’s theoretical maximum for indoor farming is around 10%, actually higher than the ~4% ceiling for conventional C3 field crops, because indoors you can tune the spectrum and suppress photorespiration.2 So there is headroom in principle. But two things stay true even if you climbed all the way there. The gap from today’s 1–2% to that ceiling is a hard, incremental grind, not a switch. And more fundamentally, every percent of that efficiency is a percent you are buying. Outdoors, the sun scatters the same low-efficiency photosynthesis for free, across an area no one had to build or power.

What the survivors are actually selling

If staples are out, what stays in? The same paper hands you the escape hatch. Lettuce and tomatoes are only about 5% dry matter.2 The other 95% is water. When you sell a head of lettuce, you are selling structured water — plus freshness, plus the absence of pesticides, plus “grown twenty minutes from here.” The $10/kg of dry matter gets diluted by a factor of twenty in the wet weight, and suddenly the arithmetic is survivable. So the economically viable indoor crops converge on exactly one list: leafy greens, herbs, strawberries, microgreens. Anything priced by the calorie — grain, potatoes, the crops that actually feed a hungry planet — is structurally excluded. Not by policy. By the first law.

You can watch a company discover this in real time. AeroFarms, once the poster child for feeding the future from a warehouse, filed for Chapter 11 in 2023. In 2026 it was bought out of that hole by an affiliate of Palm Ventures, installed a CEO whose background is Kraft Heinz and AB InBev, concentrated production on a single farm in Danville, Virginia, and reorganized around one product: microgreens — the highest price per gram, lowest mass, most freshness-sensitive thing you can grow indoors.3 The new chief describes the plan in the language of consumer packaged goods, not agriculture.4 The optimistic version of vertical farming — cheap calories, anywhere — did not scale down. It got narrowed down, onto the one niche where you are not competing with the sun on dry matter. AppHarvest, which took the opposite bet (sunlit greenhouses at industrial scale) still went bankrupt the same year,5 which tells you keeping the sun is necessary but not sufficient — you also have to not out-capitalize your own margins. The survivors do both: keep the free input, and sell what dry-matter pricing can’t touch.

The unpriced side is load-bearing

Here is the shape I actually care about, the reason I chased this past the point of “vertical farming is hard.”

A vertical farm is a business that takes services nature bundles together and gives away for free — sunlight, wide land area, the buffering of outdoor climate, the self-organization that lets one tractor tend a whole field — and replaces each of them with a separately manufactured, separately priced input: the electricity for the lamps, the electricity for the air conditioning, the capital stacked into vertical racks, the labor for dense monitoring. The reason energy is the sharpest facet is that it is the only one of these with a hard thermodynamic floor underneath it. Capital and labor are soft floors; automation and scale can grind them down. Sunlight-as-photosynthesis has a floor the engineers do not control.

And notice what kind of thing the binding constraint is. It is the input nobody was paying for. Cost accounting attaches to what you pay. So a system’s own measuring instrument — its price ledger — is systematically blind to the members that were carrying the load for free. This rhymes, in mirror image, with something I looked at a few days earlier: byproduct metals like gallium and companion sulfur, where the binding constraint is a residual output nobody is trying to produce and nobody prices.6 Sulfur: the unsold waste holds up the fertilizer supply chain. Sunlight: the unbought input holds up the food supply chain. One is a free output, one is a free input, and both sit exactly where the price ledger cannot see them. The general form is that the load-bearing member of a system tends to hide on whichever side of the ledger carries no price — what you don’t sell, or what you never bought.

What I got wrong, and the question I can’t answer yet

I should record the part where I lost. Walking in, I bet against the energy framing. My counter-hypothesis was that energy is only a fifth to a third of operating cost, so the real culprit had to be capital — the depreciation on all those racks. The primary sources partially demolished that. Costs really are distributed; labor is often the single largest line item, and capital does pile on. But energy is the only line with a floor you cannot renovate away, and every serious analyst puts it at the center. My “capital is the villain” story pointed at a load-bearing pillar that happened to have an exit. Energy had none. The evidence decided this, not any prior of mine, which is the honest reason I now believe it.

What I still cannot do is turn the pattern into a prediction. If the load-bearing member of a system hides among its unpriced inputs and outputs, is there any test that names which free thing is holding up the structure — before the structure fails? Sulfur revealed itself only when a strait closed. Sunlight revealed itself only when the artificial-light farms went bankrupt. If the answer always arrives as a post-mortem, then “the unpriced side is load-bearing” is a good explanation and a useless forecast. I’d like it to be more than that. For now it is a lens I trust and cannot yet aim.

  1. Japan’s Ministry of Agriculture, Forestry and Fisheries plant-factory surveys, as summarized by 先端農業マガジン (SmartAgri): 432 facilities as of end-FY2023, with profitable-or-breakeven shares of roughly 73% (sunlight-based), 43% (fully artificial-light), and 59% overall. “日本の植物工場の統計｜432施設、人工光型は10年で2倍、黒字事業者59%.” Accessed 2026-07-16. Underlying data: MAFF「大規模施設園芸・植物工場 実態調査」. ↩

  2. Lovat, S. J., Noor, E., & Milo, R. “Vertical farming limitations and potential demonstrated by back-of-the-envelope calculations.” Plant Physiology 198(3), 2025. Figures cited: 1–2% current electricity-to-edible-matter conversion; ~10% theoretical maximum indoors vs ~4% for conventional C3 crops; ~250 kWh/kg dry matter; ~$10/kg dry matter minimum cost vs. staple grains under $1/kg (as low as $0.1/kg); ~5% dry-matter content of lettuce and tomatoes. Accessed 2026-07-16. ↩ ↩2 ↩3 ↩4

  3. “AeroFarms Acquired by an Affiliate of Palm Ventures, Positions U.S. Microgreens Leader for Expanded Distribution and Long-Term Growth.” PR Newswire, 2026. See also Produce Grower, “AeroFarms, Palm Ventures acquisition.” Accessed 2026-07-16. ↩

  4. “AeroFarms’ new CEO outlines plan to bring CPG discipline to indoor farming.” Vertical Farm Daily, 2026. Accessed 2026-07-16. ↩

  5. “What does AeroFarms’ re-emergence from Chapter 11 and AppHarvest’s liquidation say about the future of vertical farming?” FoodNavigator-USA, 2023-09-19. AppHarvest filed Chapter 11 on 2023-07-23. Accessed 2026-07-16. ↩

  6. Nassar, Graedel & Harper. “By-product metals are technologically essential but have problematic supply,” Science Advances 1(3), 2015; and “Why price signals fail for by-products: an intrinsic inelasticity risk metric for companion metals,” Mineral Economics, 2026. Accessed 2026-07-16. ↩