CM Workshop · Beliefs Analysis

Cultivated Meat Cost (CM_01) · Post-workshop beliefs analysis · May 8, 2026 workshop · Summary · Beliefs form · The Unjournal · Public working synthesis

Publication and consent check — 20 July 2026. A name appears beside a belief only when the respondent supplied a name under a form notice saying named responses could be published; blank affiliations are not filled from our own records. Anonymous responses have no exact timestamp, and one contributor's timing and background are deliberately suppressed. This page contains no discussion from the private S2 session. These are stated beliefs, not verified expert findings or a Bayesian posterior.

CM_01 Distribution

Subquestions

Individual Responses

Methods & Notes

Next Steps

Next steps (updated 20 July 2026)

We're sharing this analysis now rather than waiting any longer. Here's what happens next — both for workshop participants and respondents, and for the additional forecasters and modelers we're now inviting.

If you took part in the workshop or already submitted beliefs

This page is a public working synthesis of the beliefs gathered so far. Please check whether your own views are represented fairly — corrections are welcome by reply, in the collaborative workshop document, or via the Hypothes.is annotation banner on this page.
We didn't want to delay sharing any further. In parallel, we're inviting a slightly expanded group of forecasters and modelers — including people without ties to cultured meat, and some who have previously given more skeptical or negative forecasts — to share their beliefs too, for balance and dialogue. This helps balance the picture (admittedly somewhat ad hoc): people without direct CM links may have been less likely to attend the conference in the first place, so their views are under-represented here. We think this is important for getting at the cruxes and making real progress in public understanding.
We now have three independent responses in this expanded round: one named response and two anonymous responses. They were submitted after the workshop, but they should be treated as pre-synthesis / pre-discussion-exposure independent views: the standalone form deliberately did not show the workshop estimates, model output, or meeting synthesis before asking for the respondent's own forecast. One anonymous response argues against the bottom-up method itself rather than merely adjusting its inputs; it is summarized under "The main dissenting case" on the CM_01 tab. Three is still thin. Once we either have a few more independent responses or decide to close this round, we'll circulate an update to everyone: the revised synthesis plus the key apparent cruxes and discussion points.
At that point — when there's a richer, more balanced picture to react to — we'll invite you to revisit and update your own beliefs. We expect that round to be more fruitful than asking for updates now.
We're also considering additional live events (see below).

If you're a forecaster, modeler, or skeptic we've invited

Invited forecasters, modelers, and skeptics now have a dedicated standalone page: cm-cost-beliefs.html. Please give your beliefs there first — before looking at the workshop participants' estimates on this page — so your forecast isn't anchored by theirs.

What we're asking: share your beliefs on the focal cost question (CM_01) and complete at least a reasonable share of the subquestions — not just one or two — with brief reasoning, via the independent beliefs form. The subquestions are where much of the value lies, so please don't treat them as optional. Roughly 30–60 minutes; partial responses are still accepted, but a fuller set is much more useful.
Why you: we're seeking independent, quantitatively-minded perspectives — including critical and skeptical ones — from people with a track record in forecasting or techno-economic modeling, to balance and stress-test the picture above.
Honorarium: $100 (Tremendous voucher or a donation to a charity of your choice) for each of the first four reasonably complete and detailed responses from invited forecasters and experts, plus a further $50 reserved for those who also return for the short follow-up/update round. (fairness note)The compensation difference is needed to elicit participation from people with less inherent interest in cultured meat. Prior workshop contributors who already gave a detailed response are equally welcome to the same honorarium for a substantive update — and we hope to be able to offer this kind of modest compensation more broadly in future.
Anonymity: you can respond named or anonymously — just let us know your preference.

P.S. — what's coming. We're planning future cultured-meat online sessions (less structured) on specific topics, and a Fermi-style modeling workshop — a rough sketch is here: modeling-hack (draft). We're applying for funding for more ambitious follow-ups; in the meantime we'll pursue the lines above as best we can.

CM_01: What will be the average production cost ($/kg, wet weight at harvest) of undifferentiated cultured chicken cell biomass in 2036, assuming large-scale commercial production is achieved?

This was the focal question, but it sits within a broader set of other interesting questions — see the Subquestions tab.

Dataset filter

Aggregation method Linear mixture (recommended descriptive pool): F_agg(x) = (1/n)·ΣFᵢ(x). Each respondent contributes equal weight. For asymmetric stated intervals, the implementation uses separate lower and upper log-scale spreads so the respondent's median, p10, and p90 are all preserved. The aggregate can remain multimodal.

Geometric pool (symmetric approximation): log f_agg = (1/n)·Σ log fᵢ + C. The closed-form calculation first compresses each response to a symmetric log-normal using its interval width. Respondents with narrower intervals pull the location more. Use only as a sensitivity analysis; confidence is not the same thing as expertise or independent information.

Product-of-densities illustration: Multiplies the symmetric response densities and produces a much narrower curve. It is not a Bayesian posterior unless each submitted distribution can defensibly be treated as an independent likelihood with a compatible common prior — conditions that almost certainly do not hold here.

Default σ (no CI)For respondents who gave a median but no confidence interval, we assign a default log-scale standard deviation σ. Higher σ = more uncertainty assumed. The 80% CI would be roughly [median/exp(1.282σ), median·exp(1.282σ)].

1.0

Exclude

All-50% default submissions "FN" ($1/kg, no CI)"FN" is a respondent-entered pseudonym or initials whose meaning has not been confirmed. The response includes substantive text and does not look like an untouched default, but its identity and uncertainty are unknown. Exclude it here for a sensitivity check. Early anonymous (May 4)

Researcher / Analyst

Industry practitioner

Unknown / other

CM_01 aggregate — Linear mixture

One additional written estimate is visible but not pooled. An anonymous contributor gave a 2036 path estimate of about $20/kg per edible kg of meat, without an interval. It appears in the individual chart and response card, so the elicitation is not discarded. It is excluded from the standardized CM_01 pool because the workshop question asks about wet cell biomass at harvest; silently treating those units as identical would add false precision.

Individual estimates with uncertainty (80% CI) · log scale

Each bar spans the 10th–90th percentile of the respondent's stated or inferred distribution. Dot = median. Color = respondent category. Log scale: equal visual distance = equal proportional difference.

Stratified summary by respondent category

Only for respondents with CM_01 estimates and known categories.

The main dissenting case: a top-down benchmark against yeast

One anonymous independent response is the most pessimistic in the set — $500/kg for 2036 (80% CI $100–$5,000), roughly 5× the next-highest estimate and about 20× the workshop cohort's geometric mean. It matters less for its number than for its method. Most other substantial responses here reason bottom-up: project media cost, cell density, capex and scale, then multiply through. This respondent argues that approach is the problem, because it combines many uncertain assumptions with unknown interactions and is therefore highly sensitive to optimistic inputs. They instead anchor on a comparison with a technology that already exists at industrial scale. The respondent asked to remain anonymous but explicitly allowed the argument to be circulated and challenged.

The benchmark is yeast biomass fermentation. Yeast doubles in roughly 1.5 hours, grows on very cheap media, tolerates harsh industrial conditions, and scales on mature low-cost equipment; animal cells do none of these things. Their estimate of the total production-cost disadvantage is about three orders of magnitude (two if one is very optimistic): roughly two orders from volumetric productivity alone — lower density combined with roughly 15× slower doubling, which sets reactor volume and hence capex and opex per kg — plus about one order from growth factors and recombinant proteins that microbial fermentation never has to buy.

The empirical anchor is the part worth taking seriously even if the exact orders of magnitude are contested. Industrial yeast fermentation is one of the most optimized bioprocesses we have, and commercial yeast-derived single-cell protein still prices in roughly the same order of magnitude as animal meat. Their inference: if the most efficient biological production system available has not become dramatically cheaper than meat, it is hard to believe cultured animal cells — starting from a much less favourable position — reach commodity food cost. They accept the reframing that a cultured-vs-fermentation gap would not settle the funders' question if fermentation itself sat above animal-meat cost, and answer it directly: they read fermentation SCP as already near meat parity, so cultured sits well above animal meat rather than near it.

A measurement point that applies to everyone's answer, not just theirs. CM_01 asks about undifferentiated biomass. The respondent notes that a real meat product additionally needs a differentiation phase, in which cells stop dividing but keep consuming nutrients for maintenance, plus scaffolding and organoleptic work. None of that is inside the CM_01 boundary, and the CHO-style proliferative numbers that several optimistic estimates lean on ignore it. If they are right, every figure on this page — optimistic and pessimistic alike — understates the cost of finished cultured meat by an unknown but non-trivial factor. They also flag overflow metabolism (glucose shunted to lactate) as a reason a cheap $/L medium still yields an expensive $/kg output, and argue feed conversion should be modelled explicitly rather than left implicit in an assumed L/kg.

Their specific bottlenecks, each with a stated mechanism, are: slow doubling and low volumetric productivity driving reactor volume and therefore capex; fragility, since animal cells lack a cell wall and are shear- and stress-sensitive, capping reactor scale and forcing expensive sterility; senescence in primary cells, which prevents the indefinite continuous operation yeast allows, while the immortalized lines that avoid it are the least meat-like and carry food-safety and consumer concerns; GMP at scale, where contamination-proneness means cleanliness costs do not amortize away; and per-cell-line characterization, which is largely non-transferable and repeated for every new product. On the workshop's own optimistic claims they are unpersuaded but not dismissive: hydrolysates are plausibly cheaper but their batch-to-batch variability and endotoxin risk sit awkwardly with a reproducible-quality food; cell-line engineering is real but speculative and not, in their view, enough to cross an order-of-magnitude gap; and large-scale perfusion raises density at the cost of shear, fouling and heavy media consumption, unproven at scale.

They add a structural argument that is separable from the biology: a large sustained gap is partly self-perpetuating. While cultured meat sits far from parity there is little rational incentive to fund the deep, coordinated, largely non-appropriable R&D needed even to reach its biological floor — so that floor may never be realized. They see the field compounding this by fragmenting, with much funding spread across companies that each develop and characterize their own cell line, duplicating the hardest and least-transferable step rather than solving it once. And because cost is a product of many drivers, isolated wins such as cheaper growth factors do not compose into a low headline cost unless media, productivity, senescence and capital are solved together.

Three questions they put to the optimists. These look like good candidates for the next elicitation round, since they are answerable and would separate the positions rather than restate them.

Estimate cultured-meat production cost top-down, benchmarked against yeast biomass production. What volumetric productivity (g/L/day) and doubling time do you assume for 2036, and how do they compare to yeast?
Why should animal cells undercut the yeast SCP cost floor, when yeast — far more efficient and mature — only reaches the same order-of-magnitude parity with meat?
In your bottom-up TEA, which single assumption, if wrong by one order of magnitude, breaks the result?

They state what would move them: feed-grade growth factors at scale, several productivity breakthroughs, or economical large-scale perfusion. Their counterfactual claim is that plant-based foods and microbial fermentation offer higher expected returns per research dollar — not because cultured meat is uniquely hard, but because those platforms are already biologically efficient and industrially validated while remaining, in their view, scientifically immature and open to innovation.

Effect on the aggregate, for transparency (recomputed 20 July 2026): adding this one response moves the equal-weight linear-mixture median from $27.3 to $31.0/kg, the arithmetic mean from $34.9 to $70.7/kg, and the p90 from $88.7 to $125.6/kg. That asymmetry is a property of the pooling rule — an extreme response fattens the right tail much more than it moves the centre — and is a reason to read the median and the mean together rather than quoting either alone. The three independent-cohort responses have a geometric mean of about $100/kg against about $18/kg for the ten standardized workshop-cohort estimates, but with n=3 in the independent set that gap is suggestive rather than established.

Further details & notes

Pre / post workshop comparison

CM_01 Supplementary — 2027 near-term target

The original workshop form included a supplementary question: "Same question as CM_01, but as of December 31, 2027, across all large-scale plants in the world?" Workshop respondents skipped it, but the independent form now has three 2027 estimates: Paul Wood at $93/kg (80% CI $75–$119), one anonymous response at $150/kg (80% CI $50–$250), and another anonymous response at $5,000/kg (80% CI $500–$50,000). These near-term estimates are all pessimistic relative to the 2036 medians and may be useful for calibration, even though the sample is still very small. The $5,000 figure comes with an argument worth separating from the number: the respondent holds that a 2027 "average" is dominated by pilot economics at tiny volumes with near-zero learning, and that reported costs understate full economic cost because much current output is subsidized R&D. That claim is checkable against company disclosures independently of whether the $5,000 estimate is right.

Submission timeline

When did responses arrive, relative to the workshop on May 8, 2026?

The workshop ran 11am–3pm ET on May 8. Workshop-form submissions are split into before, during, and after that window. Earlier versions treated all May 8 submissions as "pre/during" and sometimes shortened that to "pre-workshop"; this revision corrects that. Independent submissions made later are not evidence of workshop-driven updating: the standalone form asked for the respondent's own view before showing this synthesis or the workshop distribution. Exact timing is suppressed for specially protected anonymous records.

Technical sub-questions and expert distribution questions. Only respondents who answered each question are shown.

CM_02 AW investment CM_10 AW benefit CM_12 Hydrolysates CM_14 Media cost CM_17 Food-grade CM_20 Bioreactors E6 GF pathways CM_13 GF cost

CM_12 — Hydrolysate adoption by 2036

Probability that most commercial CM uses hydrolysate-based basal media by 2036 (%)

CM_14 — Basal media cost ($/kg biomass)

Median estimate of basal media cost per kg cell biomass output, excluding growth factors

CM_17 — Food-grade media adoption (%)

Share of commercial CM using food-grade rather than pharma-grade media by 2036

CM_20 — Companies building own bioreactors (%)

Share of CM companies (capex > $10M) designing and building their own bioreactors by 2036

E6 — Growth factor innovation: probability each pathway achieves meaningful cost reduction by 2036

Five pathways to reduce GF costs. Sliders from 0–100%. Each respondent's five values shown as a grouped row. Color = category.

CM_02 — Relative AW investment value

Qualitative assessment of CM funding vs. next-best AW intervention, plus numeric benefit-share estimate (% of next-best, where 100% = equal; >100% = CM is better). Not all respondents gave a numeric estimate.

CM_10 — AW benefit vs. proven interventions (%)

Probability a $100K CM investment exceeds the AW benefit of proven alternatives

CM_13 — GF cost per kg biomass ($/kg)

Expected growth factor cost contribution per kg biomass by 2036

Process mode (E5) — % of production by mode in 2036

Two respondents provided E5 values, and they disagree sharply on where production lands. Elliot Swartz: fed-batch 20%, perfusion 50%, continuous 30%. An anonymous independent respondent: fed-batch 80%, perfusion 15%, continuous 5% — reasoning that fed-batch stays the low-capex default, perfusion grows only where higher density justifies the complexity and media use, and continuous flow stays a frontier share given limited demonstrated maturity at scale. Since perfusion is one of the levers the optimistic case leans on, this split is a crux rather than a detail.

Discussion responses (open text)

Individual responses

All human submissions (including one anonymized written response; test and machine submissions are dropped — see Methods). Sorted by CM_01 estimate (low to high; no-estimate at end). Click to expand. One anonymous independent card carries a fuller written argument shared with the respondent's permission.

Methods, notes and annotations

Statistical approach

Respondent distributions: Cost is represented on the log scale because it is positive and often right-skewed. For the recommended linear mixture, a respondent with median m and 80% interval [lo, hi] gets a two-piece log distribution: the lower spread is ln(m/lo)/1.282 and the upper spread is ln(hi/m)/1.282. This preserves all three stated quantiles even when the interval is asymmetric. Respondents without an interval are assigned a user-selectable symmetric log-scale σ (0.3 = narrow to 1.8 = very wide).

Three aggregation methods are available:

Linear mixture (recommended descriptive pool): F_agg(x) = (1/n)·ΣFᵢ(x). Each respondent contributes equal weight and asymmetric intervals are preserved. The result can be multimodal. Conceptually: "give equal credence to each respondent's full stated distribution." This is descriptive, not a claim that the group forms a calibrated posterior.
Geometric pool (symmetric approximation): log f_agg = (1/n)·Σlog fᵢ + C. The closed form requires compressing each response to a symmetric log-normal. Location is the precision-weighted mean of log-medians and width is √(n/Σ(1/σᵢ²)). Narrow intervals therefore receive more influence, which is not automatically desirable: confidence can reflect overconfidence or shared evidence rather than accuracy.
Product-of-densities illustration: Uses the same symmetric approximation but width 1/√Σ(1/σᵢ²). It becomes dramatically tighter. Calling this a Bayesian posterior would require each submitted distribution to be an independent likelihood with a compatible common prior; the workshop respondents share literature and discussion, so those assumptions are implausible. It is included only to show sensitivity to a strong pooling rule.

The methods need not agree on location. In particular, the equal-weight linear mixture's median is generally different from the precision-weighted location used by the other two methods. Report the method whenever quoting an aggregate.

Geometric mean of point estimates: exp(mean(ln(mᵢ))). Appropriate for log-symmetric distributions; less sensitive to the highest estimates than the arithmetic mean. Shown alongside the precision-weighted geometric mean (where respondents with tight CIs get more weight).

Exclusions: One test submission is excluded entirely. "FN" is a respondent-entered pseudonym or initials whose meaning has not been confirmed. That submission includes substantive text and does not look like an untouched default, but it has no interval; the control above exposes its influence as a sensitivity check.

Machine submissions (revised 18 July 2026). Six submissions carried every slider at exactly 50% with no free text, no CM_01 estimate, and no identifying fields. Earlier versions described these as human "no-interaction" responses and merely excluded them from the default view. The shared known mail-crawler address ranges and clustered arrival pattern make automated link-scanning the more consistent explanation. They are now dropped during ingest. The rule is deliberately narrow — an all-default payload, with no text of any kind, from a known crawler range — so a real respondent who genuinely leaves the sliders untouched would still be kept and flagged. The form's touchedSliders mechanism now prevents untouched sliders from being submitted at all; these submissions predate that fix.

Respondent categories

Categories are assigned from submitted affiliations and, for named workshop participants, the public workshop roster. "Researcher" includes academic researchers and nonprofit analysts; "Industry" includes company practitioners. Anonymous respondents are left as "unknown / other" unless they supplied a public category themselves. We deliberately do not infer a protected anonymous respondent's background for the public stratified analysis.

The May 4 anonymous submission has a consumer-perspective reasoning ("current chicken price in my country") and a very wide CI — it is flagged as non-expert and can be excluded via the control on the CM_01 tab.

2027 near-term question

The original workshop form did include a supplementary CM_01 question for December 31, 2027, but workshop respondents skipped it. The independent form has produced three estimates: Paul Wood at $93/kg (80% CI $75–$119), one anonymous response at $150/kg (80% CI $50–$250), and another anonymous response at $5,000/kg (80% CI $500–$50,000). Treat these as a very small independent sample, not as a workshop update.

Update analysis (pre/post)

Two workshop-cohort post-workshop CM_01 responses have arrived: PersonABC and Andrew Stout. The ingest step uses exact timestamps to distinguish before-workshop, during-workshop, and post-workshop submissions instead of treating all May 8 entries as pre/during; the public data retain only the date, and protected anonymous dates are withheld entirely. The independent round adds Paul Wood and two anonymous responses, but these are post-workshop only by calendar date and pre-synthesis / pre-discussion-exposure in the relevant information sense. For workshop-effect comparisons, the dashboard excludes the independent cohort and still has no paired pre/post responses from the same person. When paired responses exist, the dashboard will compute and display the shift in μ (log-scale location) and σ (uncertainty) for each respondent.

Hypothes.is annotations

Public annotations: Install the Hypothes.is browser extension and annotate freely. Public annotations on this URL are monitored by the workshop team.

Data completeness

Most useful next reads: Next Steps → Subquestions →

Note on one anonymized written response and unit alignment (click to expand)

One anonymized contributor submitted detailed answers using an earlier Pivotal-Questions wording, which differs from the workshop form on several questions. Their CM_14 and CM_17 map directly. The original CM_01 estimate is shown but not pooled; CM_12, CM_13 and CM_16 remain qualitative rather than being converted using assumptions the contributor did not state:

CM_01: ~$20/kg in 2036 per edible kg of meat, with no interval. Visible as stated; excluded from the standardized wet-biomass pool.
CM_12: answered the different "replacing growth factors" wording; no inferred probability is plotted on the base-media question.
CM_13: ~$500 per gram of growth factor; no $/kg biomass conversion without a stated loading.
CM_16: ~1.5×10⁸ cells/mL as a frontier maximum; no g/L conversion or typical-value inference.

Full reasoning is on the anonymized card under "Individual Responses." This keeps the actual elicitation visible while avoiding fabricated precision.