The model in plain terms

A short guide: what the model works out, the few numbers that set the cost, how to read the price lines, and how far to trust it

Written by Claude (AI) with David Reinstein, 17 November 2026. Every number on this page is read from five-drivers-2026-11-17.json, which scripts/five-drivers.mjs computes from the current model (1,000,000 simulated worlds). Not yet reviewed by engineers or cell-culture experts.

In brief

1. The cost is five numbers

The cost of a kilogram of dry cells is:

litres of medium per kg × (price of medium per litre + growth-factor cost per litre) + plant running costs + capital charge

“Medium” is the nutrient liquid the cells grow in. “Growth factors” are signalling proteins added to the medium in tiny amounts; today they are very expensive. “Plant running costs” are labour, maintenance, energy, consumables and a few other proteins. The “capital charge” is the yearly cost of paying for the plant, spread over each kg it makes.

How to read it.

Which of the five matters most?

One way to ask this: if we knew one of the five numbers for certain, how much of the spread in the model’s cost would disappear?

So nearly all of the model’s uncertainty is about medium: how much of it a kg of cells needs, what a litre costs, and what the growth factors in it cost. Running costs and capital hardly affect the spread across all worlds. They matter in a different way.

Running costs and capital are a floor

2. What lies behind the five numbers

The five numbers are where the cost is added up. They are not the root causes. Each one depends on which technical developments succeed, and those developments are what research and investment could actually change.

So the developments that matter are few: a dense production process (perfusion or continuous culture) that works at commercial scale, cheap medium, cheap growth factors, and a plant that is cheap to build and run. The model represents the first three thinly, as yes-or-no switches with a probability each plus a price range, and the fourth not at all.

How much the process matters on its own
Do the breakthroughs arrive together?

3. Reading the price lines

The pages quote costs in three ways, which is a common source of confusion:

  • Per kg of dry cells. This is what the model calculates. Cells are mostly water, so a kg of dry cells is several kg of meat.
  • Per kg of meat made only of cells, at the water content of chicken meat. This is the dry-cell cost multiplied by 0.275.
  • Per kg of a “hybrid” product: cells mixed with plant-based ingredients costing $3 a kg. This is how products are likely to be sold, and it is what the threshold cards on the main model page use.
The hybrid figures assume 25% cells. That is a convention, not a finding.

There is no agreed share of cells in a hybrid product. These pages use 25% as the baseline. Products on sale may contain less: in November 2026 UPSIDE Foods described the cell share of its new blended chicken only as “double-digit”, meaning at least 10% (Green Queen). The fewer cells a product contains, the less the cost of the cells matters. The table below therefore shows a 10% version alongside the 25% baseline. You can change the share on the Advanced model under Hybrid product.

Two things follow from this table.

4. How far to trust it

What has been checked. The arithmetic. The model has automated tests, and Alex Bates independently rebuilt it in November 2026; with the same settings his version reproduces ours exactly, world for world. His rebuild also found two logic errors, which were corrected on 17 November 2026 (what changed).

What has not. Most of the input ranges. Several of the inputs that matter most have no published source:

  • plant running costs (the largest cost in the cheapest worlds);
  • the probabilities that hydrolysate medium, cheap growth factors and food-grade proteins work by 2036 (75%, 50% and 70%);
  • the favourable ends of the ranges for nutrient yield and medium price, which Alex’s literature review finds more optimistic than the published evidence supports;
  • how strongly good outcomes are assumed to arrive together (the “maturity” factor, explained in the technical reference).

What that means. The share of worlds that come out cheap depends mostly on how far those favourable ends reach and on whether they happen together. Neither has been asked of experts yet. Treat the model as a way to see what would have to go right, and how much, rather than as a prediction. The next step is to ask experts about the five numbers directly, and about the handful of breakthroughs behind them.

5. What this can and cannot tell a funder

What it can tell you.

  • How far away cheap cultured meat is, and what has to go right together. For example, cheap medium alone is not enough (section 1), and plants using fed-batch almost never get near chicken prices (section 2). Work that cannot lead to a dense process, cheap medium and a cheap plant together does not lead to price parity.
  • Which unknowns are worth resolving first. The numbers that carry most of the uncertainty, and those with the weakest sources, are where better evidence or expert judgement would change the answer most.
  • Whether a claim is consistent. A company’s stated price, cell share and scale can be checked against each other.

What it cannot tell you.

  • How much a research dollar changes the odds. The model takes the probability that, say, hydrolysate medium works as an input. It contains nothing about how that probability responds to funding. So on its own it cannot say whether to fund cultured-meat research, or which line of research.
  • Cause and effect. “Worlds where cells cost under $36” are combinations of inputs that happen to come out cheap. They show what cheap cultured meat would look like, not how to bring it about.

What would be needed to get from here to a funding decision. First, judgements from experts about how much specific research or investment would shift the probabilities and ranges (“with this programme, and without it”). The model can then turn those shifts into a change in cost. Second, a view of how a lower cost changes how much cultured meat is sold and how much conventional meat it displaces; the cost and demand page is a start on that.

A possible shorter list of questions for experts

This is a draft for discussion. It has not been adopted, and it does not replace the questions the project already asks. It is here so that the team and the experts can react to it.

The current plan asks experts about many of the model’s individual inputs. Because the cost comes down to five numbers, one possibility is to ask about those five directly, and then about the few developments behind them. That would be about eight questions.

Read the draft questions

Each question is about a large plant built with the best technology available in 2036, and about cost in today’s dollars. For each of the first five, we would ask for two values: your best single estimate, and a favourable value that you think has only about a 1-in-20 chance of being beaten.

The five numbers

  1. Medium needed. How many litres of fresh medium would it take to produce one kg of cells (dry weight)?
  2. Medium price. What would that medium cost per litre, leaving out growth factors and other added proteins?
  3. Growth factors. What would growth factors cost per kg of cells (dry weight)?
  4. Running the plant. What would it cost to run the plant per kg of cells (dry weight): labour, maintenance, energy, consumables and added proteins, leaving out medium, growth factors and the cost of building the plant?
  5. Building the plant. How much would it cost to build the plant, per kg of cells (dry weight) it can produce each year?

What would have to happen

  1. How likely is each of these by 2036: (a) food-grade medium that costs a small fraction of today’s, by any route; (b) growth factors that are cheap, or cell lines that need very little of them; (c) perfusion or continuous culture working reliably at commercial scale?
  2. If the first of these happens, how likely are the other two? (This tells us whether good outcomes tend to arrive together, which the model currently has to guess.)
  3. Which single development would most change your answers to questions 1 to 5, and by roughly how much?

What this would gain and lose. Experts would answer fewer questions, in units closer to what they see in practice, and the answers would bear directly on the numbers that decide the cost. The detailed inputs would still be in the model, but they would be set to agree with the answers rather than asked about one by one. What would be lost is the detail on why a number is high or low; question 8 is there to recover some of that. Whether experts find these five numbers natural to estimate is itself something to test before relying on them.

Terms used on these pages

Open the list
  • Simulated world (also called a “draw”): one run of the model with one value chosen for every input.
  • Dry cell mass: the weight of the cells with the water removed. Sources often quote costs per kg of wet cells, which are 70–80% water, so a wet figure is roughly a fifth to a third of the dry figure.
  • Nutrient yield: grams of dry cells that one litre of medium can feed. It sets the least medium a kg of cells can use.
  • Medium-use multiplier: how many reactor-fulls of fresh medium are used per reactor-full of cells harvested.
  • Hydrolysate: a cheap digest of plant or yeast protein that could replace the purified amino acids in medium.
  • Hybrid product: a product that is part cultured cells and part plant-based ingredients. These pages assume one quarter cells.
  • Maturity: one number per simulated world that makes several technologies slightly more or less likely to work together. Its effect is small.
  • Price-competitive zone: on Alex Bates’s page, cells costing between $12 and $36 per kg dry, that is, between matching ordinary chicken and matching organic chicken.
  • Minimal set: on Alex’s page, one of the smallest combinations of things going well that gets a simulated world into the price-competitive zone.

Where to go next