If you are trying to work out what precision-fermented protein costs to make, the honest answer is that you cannot find out from published sources. Not because the number is complicated — because the models barely exist.
The Good Food Institute, working with Hawkwood Biotech, screened 190 publications for its June 2025 meta-analysis of fermentation economics. It found 55 techno-economic models meeting quality and relevance thresholds. Of those, exactly four covered precision fermentation for food proteins: a collagen peptide, thaumatin, lactoferrin, and one unspecified recombinant nutritional protein.
Four models. And their estimates span from under $20/kg to roughly $15,000/kg — a range so wide it carries almost no information.
The more revealing detail is the shape of the gap. There are no published models in the $20–200/kg band. That is precisely where industry practitioners, working from private benchmarks, believe the current state of the technology sits.
The public literature covers everything except the part that matters.
What a techno-economic model is, and why the gap bites
A techno-economic model estimates the full cost of producing an ingredient — raw materials, energy, labour, capital recovery, downstream processing — through to finished product. It is the tool investors use to sanity-check a business plan and the tool a formulator uses to judge whether an ingredient can ever hit a target price.
When the public models cluster far away from commercial reality, three things follow. Founders cannot benchmark against anything credible. Investors underwrite against numbers that are systematically wrong. And journalists — including trade publications — repeat figures whose provenance nobody has checked.
That last one is worth naming, because this article is subject to it too. Every figure below comes from a single meta-analysis, and its authors are explicit about the limits of their own evidence.
Where the data is actually good: biomass fermentation
The contrast with biomass fermentation is stark, and instructive.
Biomass fermentation — mycoprotein and single-cell protein from fungi, yeast, algae and bacteria — has 25 published models. Costs cluster at $4–6/kg of biomass, median around $4.30/kg, with a full range of $1.27 to $18.10/kg driven mostly by scale and feedstock.
| Product | Published median cost |
|---|---|
| Fungal biomass | $21.10/kg protein |
| Structured poultry | ~$13.50/kg protein |
| Plant protein concentrate (soy, pea) | ~$6.00/kg protein |
Read that table carefully, because the headline “fermentation approaches parity with beef and pork” rests on the biomass comparison — $3.90–6.00/kg of biomass against wholesale beef and pork prices. On a per-protein basis the picture is less flattering: fungal biomass still carries a meaningful premium over poultry and a substantial one over plant concentrates.
Both framings are defensible. They answer different questions, and the more optimistic one travels further.
The 180% divergence between public and private models
Here is the finding that should change how you read any published cost figure.
Hawkwood supplied proprietary FEL-1 models for commercial-scale aerobic fed-batch fermentation of yeast and microalgae on corn glucose. At 32,000 tonnes annually, those models project $2.50/kg and $4.20/kg of protein.
Published models covering similar organisms assume roughly 10,000 tonnes annually and produce a median of $7.00/kg — a 180% premium over the private benchmark.
The divergence is not a disagreement about biology. It is a disagreement about scale. Published models consistently assume smaller facilities than commercial operators actually run, and many also assume lower titer — the concentration of product in the fermentation vessel — than industry achieves.
For precision fermentation the same gap appears in sharper form:
| Parameter | Published models | Private benchmarks |
|---|---|---|
| Annual production volume | 50–2,500 tonnes | 2,500 to >25,000 tonnes |
| Average titer | ~24 g/L | ~42 g/L |
A model built on 500 tonnes a year at 24 g/L is not modelling the same industry as a company running 25,000 tonnes at 42 g/L. It is modelling a pilot plant and calling the result a cost of production.
What actually moves the number
Across the 54 models with sufficient detail, three cost categories dominate: feedstock and raw materials (the primary driver in over half), facility and capital costs second, and fermentation process metrics — yield, titer, productivity — third.
Hawkwood’s sensitivity analysis on a hypothetical 32,000-tonne single-cell protein facility turns that ranking into something you can plan against:
| Change | Effect on cost of production |
|---|---|
| Titer +30% | −12% |
| Titer −30% | +22% |
| Corn glucose price +40% | +14.3% |
| Downstream recovery −10% | +11% |
| Total equipment cost +25% | ~+6% |
Three things stand out.
Titer is asymmetric. A 30% gain buys you 12%; a 30% loss costs you 22%. Losing titer hurts nearly twice as much as gaining it helps, because fixed-cost recovery and downstream efficiency compound in the same direction. Process reliability is therefore worth more than process ambition — a strain that holds 40 g/L consistently beats one that reaches 55 g/L sometimes.
Downstream recovery is nearly as powerful as feedstock price. A 10% recovery loss adds 11% to cost. Purification is routinely treated as the unglamorous back end of the process; these numbers say it belongs in the same conversation as strain engineering.
Capital is the least sensitive lever at scale. A 25% equipment cost overrun adds only about 6%, because fixed costs dilute across high volume. This is the mathematical case for building big — and simultaneously the reason capital is so hard to raise, since the economics only work at a scale that requires the capital in the first place.
The practical consequence: if you are optimising in order, titer stability and downstream recovery come before capex negotiation. And any cost projection that does not state its assumed annual volume and titer is not a projection — it is a number.
The limitation the authors name themselves
GFI is direct about this, and it deserves repeating rather than burying.
Most of the report’s conclusions about industry competitiveness rest on private data from a single modelling firm. Hawkwood is a consultancy with commercial interests in the sector. That does not make its models wrong — they are almost certainly closer to commercial reality than the published literature — but it means the most decision-relevant figures in the most comprehensive public analysis available cannot be independently verified.
So the state of knowledge is: published models are demonstrably unrepresentative, and the representative models are proprietary. Both halves of that sentence are problems.
Why this connects to everything else
Two threads from our earlier coverage meet here.
Cultivated meat’s failures were cost failures, not regulatory ones — Believer Meats ceased operations after raising over $390m and after securing USDA clearance. We covered that pattern in our analysis of the alternative protein funding contraction. If the sector cannot independently verify what production costs, it cannot distinguish a company with a real cost pathway from one with a good deck.
And the regulatory picture compounds it. A European company can hold US clearance while its home market cannot buy the product, meaning scale — the single biggest determinant of cost — is gated behind a market-access problem. Volume drives cost down; regulation determines whether volume is reachable.
GFI counts 165 companies focused on fermentation-derived proteins as of 2024, with over 200 more running related business lines. That is a substantial industry operating without a shared, verifiable cost baseline.
What we could not establish
The $10–50/kg and $8–12/kg figures in circulation. Several secondary sources cite current precision fermentation COGS of $10–50/kg at scale, and projections of $8–12/kg by 2028. We could not trace these to a primary model, and they sit inside the exact band GFI identifies as having no published models at all. Treat them as industry expectation rather than established fact.
Hawkwood’s methodology. FEL-1 is a front-end loading estimate class, typically carrying wide error bars by design. The report does not publish the underlying assumptions in a form that permits reconstruction.
Whether the four precision fermentation models are representative of anything. Collagen peptide, thaumatin and lactoferrin are specialty ingredients with very different economics from a commodity dairy or egg protein. Four models covering four unlike products is arguably worse than none, because it invites averaging across things that should not be averaged.
What to watch
Does anyone publish a precision fermentation model in the $20–200/kg range? This is the single most useful thing that could happen to the sector’s economics. GFI has an open submission process for new techno-economic models.
Do producers begin sharing anonymised process parameters? GFI recommends it; the incentive runs the other way, since titer and recovery figures are competitively sensitive. Watch whether any consortium forms to pool them.
Does a standardised reporting framework emerge? The proposal is something analogous to ISO standards for life cycle assessment. Until models declare volume, titer and recovery in a comparable format, meta-analysis will keep comparing pilot plants with factories.
Until then, the correct response to any precision fermentation cost figure — including the ones in this article — is to ask what volume and what titer it assumes. If the answer is not immediately available, the number is not one.