Workshop Summary & Notes

David Manheim (Technion/ALTER) and Mirjam Capuder (University of Maribor) participated. It covered introductions, a walkthrough of the interactive cost model dashboard, and early framing questions about key modeling uncertainties — including Manheim's perspective on expert elicitation methodology and how to structure belief-updating around contested TEA assumptions. Substantive points from this session are incorporated where relevant in the S1 and S3 notes below.

Introduced the Pivotal Questions initiative and the purpose of the workshop: to gather calibrated expert beliefs on CM_01 and its subquestions, and to use today's discussion to update the cost model. Framed the TEA disagreement: Pasitka et al. (Nature Food, 2024) projects ~$6/lb under continuous production, drawing on assumptions about hydrolysate substitution, high cell densities, and custom bioreactor costs; Humbird (2021) projects $17–23/lb and concludes cost parity is "highly unlikely" given the biology and engineering constraints.

Two new inputs since the major TEAs were published: GFI's December 2025 amino acid supply chain analysis (suggesting Humbird overestimated AA costs by 2–10x based on real supplier quotes); and the growing commercial relevance of gene-edited cell lines, which barely appear in published TEAs but may substantially change the cost picture. These are the two most significant updates to consider.

Gave an overview of how cell line choice and gene editing cascade through the entire cost picture. Immortalisation strategy determines proliferation rates, achievable density, growth factor dependence, and media requirements — making it arguably the most upstream cost-relevant decision.

On regulatory constraints: the US FSMA framework has cleared gene-edited cell lines for food use in most applications; EU regulatory posture is more conservative, with gene editing in food production facing EFSA scrutiny analogous to GM food applications. Italy's CM production ban (2023) and lingering EU regulatory uncertainty create a geographic split that matters for where first commercial production occurs.

Kubinyecz flagged three key uncertainties for the group: (1) on what timeline will GF-independent cell lines reach commercial-scale validation? (2) Will EU regulators treat gene-edited CM differently from gene-edited crops, potentially creating a more permissive pathway? (3) Does cell line origin (biopsy from live animal vs. continuous line) affect the regulatory dossier in ways that affect cost?

Presented the FEASTS project (circular cell culture): using hydrolysates — plant-derived or by-product protein digests — as a full or near-full substitution for purified amino acids in cell culture media. The key economic argument: hydrolysates cost orders of magnitude less than purified amino acids at commodity scale. If a suitable hydrolysate can be validated for cell culture quality and consistency, the base media cost component drops dramatically.

Key challenge flagged: batch-to-batch variability in hydrolysates is a serious QC issue at commercial scale. Cells are sensitive to the precise nutrient profile; a variable hydrolysate that works in the lab may require expensive supplementation or in-process monitoring to perform consistently in a bioreactor. The FEASTS circular approach — using spent media and cell by-products to reduce fresh inputs — also introduces bioprocess complexity.

Fuchs expressed high confidence (100%) that precision fermentation and plant molecular farming will successfully produce recombinant growth factors at cost-competitive prices by 2036, and moderate confidence (70%) in widespread custom bioreactor adoption. She was more cautious on autocrine cell lines (20%), noting that the engineering required to make cells self-sufficient for GFs is harder than external supply substitution.

Media costs will be quite low. Questions remain around productivity and capital costs. — CM_01 beliefs submission, pre-workshop

Swartz agreed that Humbird's amino acid cost assumptions are significantly overstated relative to current supplier data — GFI's December 2025 analysis showed purchasable prices 2–10x below Humbird's projections. However, he was more cautious about full hydrolysate substitution:

Hydrolysates may become incorporated in a decent number of manufacturers' media formulations over the next decade. It is unlikely you'd fully replace purified AAs. You'd still likely need to supplement them in. It's also unclear if hydrolysates are truly cost beneficial. — Elliot Swartz, CM_12 written comment

His CM_01 estimate reflects this position: media costs will fall substantially, but not as dramatically as Fuchs projects. The binding uncertainties are cell productivity (density × growth rate × bioreactor occupancy) and capital costs — not media chemistry. (Specific estimates withheld pending the post-workshop update round.)

I'd say hydrolysates will contribute to a substantial portion of the basal media used at this time, but almost certainly supplemented by some amount of purified amino acids. ~80%+ chance that both hydrolysates and purified AAs will play substantial roles. — Claire Bomkamp, CM_12 written comment

The switch to food-grade ingredients represents a very substantial "low hanging fruit" cost reduction opportunity — I would therefore expect that few products beyond the very earliest launched would use substantial amounts of pharma-grade inputs. — Claire Bomkamp, CM_17 written comment

Bomkamp's CM_01 estimate reflects greater uncertainty about capital costs: "I'm less certain about capex costs than ingredient costs in general." The shift to food-grade ingredients is an early, achievable cost reduction that may happen quickly once regulatory approvals normalize. (Specific estimate withheld pending the post-workshop update round.)

I estimate the industry will achieve or surpass price parity in 10 years; based on the rate of progress to date with relatively limited amounts of funding. Numerous companies have near-term targets in the range of $0.10–0.20/L media and ~20L media/kg conversion rates. Large obstacles (FBS, growth factors, albumin, suspension culture) have been or are in the process of being overcome through process innovation. — Anonymous submission, beliefs form

Discussion ranged across the five GF cost-reduction pathways: precision fermentation, plant molecular farming, autocrine cell lines, small molecule substitutes, and thermostable variants. Fuchs's position — 100% probability that PF and plant molecular farming will deliver at cost — contrasts with Swartz's more qualified view that these pathways are plausible but not certain by 2036.

Small molecule substitutes (mimicking GF signaling without the protein itself) were seen as scientifically promising but with significant remaining uncertainty about whether cells at commercial density respond appropriately. Thermostable variants reduce cold-chain costs and degradation losses but don't eliminate the synthesis cost.

Abraham is an industry practitioner working at commercial-scale bioreactor densities. The key framing question for this topic: what cell densities are realistically achievable in a 20kL+ bioreactor by 2036, and what is the binding constraint — oxygen transfer rate, metabolite accumulation, or shear stress?

Challenged cell density as a standalone metric — a critique with direct implications for how TEA models compare scenarios and what reported "records" actually mean:

Cell density itself, as a number, it's meaningless. Because you never know what you really added. You can go to super high cell densities, but use a thousand times more whatever input you are using, and make everything expensive. Or you can go into lower cell densities and have a sweet point in between input and output that makes a bit more sense. — Jordi Morales-Dalmau, S1 (transcript)

He framed the practitioner's follow-up question whenever a news item reports a new density record: what are the inputs, and at what scale can this actually be done?

Agreed and proposed a better unit:

Cell count itself is misleading — protein content rather than cell number, because cell diameter has a huge influence on the mass, and ultimately the mass is what we're trying to make. — Elliot Swartz, S1 (transcript)

Swartz offered the biopharma CHO cell as a historical benchmark for what focused long-term optimization can achieve: 10–20 million cells/mL was considered good for decades; fed-batch processes now reach ~40 million/mL; perfusion has exceeded 100 million/mL. The implication: CM cell lines haven't had 40 years of CHO-like optimization, and current density benchmarks will likely improve — but on a multi-decade horizon, not by 2036.

In an ideal world, I think we would see a substantial proportion of companies using fit-for-purpose bioreactors built by B2B companies. I would expect this number to go up as companies move away from pharma-grade bioreactors, but then come back down as standardisation emerges. — Claire Bomkamp, CM_20 written comment

Assuming that a supplier steps in to design/build/standardize at low-cost. Doesn't seem to make sense for CM companies to own custom designs moving forward. — Anonymous participant, CM_20 written comment

Swartz added that companies are "unlikely to build them themselves" — the expectation is specialist B2B suppliers developing standardized food-grade bioreactor designs. The key open question is timing: pharma-grade equipment is expensive but available now; custom food-grade reactors could dramatically reduce capex but require a supply ecosystem to develop.

Most participants expected fed-batch to dominate near-term commercial production, with perfusion and continuous approaches following as the technology matures. Fuchs's FEASTS project uses a continuous culture approach and is an outlier in this view.

Fed-batch; perfusion/continuous technologies may take longer to scale than approaches for optimizing fed-batch (e.g., reducing metabolites). — Anonymous industry participant, CM_16 written comment

The key driver here is operational complexity: perfusion requires precise cell retention systems (filters, hollow fibers) that add failure modes; continuous culture requires maintaining a steady state that is harder to achieve at scale. Fed-batch is simpler to operate and validate for regulatory purposes — important in a GMP-adjacent production context. Perfusion does allow higher densities, but the engineering challenges mean it's likely to come after fed-batch is established.

Kubinyecz returned for discussion of regulatory timelines. Key points:

Bruno Cell (Lattanzi/Schiavo) represents a concrete case: Italy's 2023 CM production ban directly affected the company's operations. This is not a hypothetical — regulatory risk is already a material business constraint. The EU regulatory environment is unlikely to be resolved quickly enough to matter for pre-2030 commercial scale.

Lattanzi's written beliefs submission attributed his cost estimate to "growth medium costs, bioprocessing efficiency, scaffolding solutions" — a production-side framing. Regulatory costs are an additional factor that practitioners navigating hostile regulatory environments face beyond what TEA models capture. (Specific estimate withheld pending the post-workshop update round.)

The session addressed four structured topics:

1. S1 ground-truth: Which technical claims from S1 hold up in production practice — and which discrepancies matter most for cost?
2. Real-world cost benchmarks: What do CDMO production costs actually look like? Where do TEA assumptions diverge most from operator experience?
3. Paths to commercial viability: What does commercially viable CM production actually require, and on what horizon?
4. Key uncertainties and research gaps: What data or research would most reduce cost uncertainty?

Industry practitioners from Bruno Cell, Ivy Farm Technologies, Cultimate Foods, and others participated alongside researchers from GFI and Tufts. An internal notes summary is being prepared for registered participants. No S2 content will be attributed externally without explicit opt-in.

Four cruxes identified from S1 and S3 discussion. See the structured crux map for fuller analysis, grounding, and key questions for skeptics and optimists.

Presented the interactive cost model dashboard as a working tool for critique and improvement — explicitly not a definitive analysis. He introduced it in S1 with a clear disclaimer:

This is one of the equations that underlie our very first pass model, which I'm not presenting that you should necessarily use or trust. It's largely been vibe-coded back and forth from earlier work. I want to map what the cruxes are — how do we disagree about modeling this? — David Reinstein, S1 opening (transcript)

In S3 he asked participants to engage, but urged them not to over-invest before a round of back-and-forth to identify what is most off:

Engage with it to some extent — don't dump a whole bunch of time into engaging with it until we have some back and forth, because there might be whole sections that are just way off, and I don't want to waste your time and effort. — David Reinstein, S3 (transcript)

Identified a fundamental modelling challenge — the difficulty of building a model that generalizes across cell types with distinct cost profiles:

When you build a model of cost of production, it's hard to make a generalized model, because production for fat could look different than production for muscle or fibroblasts — you lose some of that nuance when you create a general holistic model. — Natalie Rubio, S1 (transcript)

Proposed "performance-to-cost ratio" as a better organizing metric for TEAs — one that can be compared across cell types and process modes without conflating media and capital costs:

I have not had a chance to look at your model in detail, but what are the key performance metrics? Something that Elliot and I published in our paper last year that we're suggesting as a key metric is a performance-to-cost ratio of the process: how much material can I make, with the equipment that I have, and what does that equipment cost — independent of the media costs. Depending on how we structure the key metrics, you know, could influence how the model is built. — Bert Frohlich, S3 (transcript)

He pushed for structured sensitivity analysis as the core tool for identifying where to focus model improvement:

TEAs are useful in identifying the key levers — where are the factors most important or sensitive to future success — and then zero in on that. That may be the basis of categorisation of additional topics. — Bert Frohlich, S3 (transcript)

Frohlich noted he had just launched a consortium on modeling of large-scale bioreactors — a parallel effort he offered as a potential collaboration point.

Identified five specific data gaps that currently prevent meaningful model validation — a list that defines what would most sharpen future TEAs:

There's basically no published studies that really quantify feed conversion ratio — what it actually is. There's also no information about inclusion rates — what inclusion rates are actually going to yield tasty products. There's also very little information around what is the actual cost of equipment. How much capital do you need to build a facility of a certain size? Is a scale-up or scale-out approach more prudent, given capital constraints? — Elliot Swartz, S3 (transcript)

He argued the workshop itself had illustrated the scope problem — cost is too large a question to resolve in a single session:

Trying to tackle the question of cost is just too big of a nut to crack at once. Chunking things into workshops on separate topics — media, bioprocess, product design — would be a more prudent, stepwise path, because it's extremely complex to cover it all at once. — Elliot Swartz, S3 (transcript)

Flagged the challenge of finding parameters that generalize across different bioprocess scenarios — the model needs a common language before comparisons are meaningful:

Could we come up with a set of parameters — like these four numbers — that are going to translate well across multiple bioprocess scenarios? — Claire Bomkamp, S3 (transcript)

She suggested cell volume per milliliter as a potentially more cross-scenario metric than cell count — consistent with Morales-Dalmau's and Swartz's earlier critiques of density as a standalone figure.

The question for AW funders: does a $100K investment in CM development yield more animal welfare benefit than $100K for proven interventions like corporate campaigns (cage-free, BCC commitments)? Participants offered very different views:

Cultivated meat is a permanent, structural solution — saving trillions+ of animals whenever impact is realized, even if this takes decades. The most important factor is whether the $100K is donated effectively vs. redundantly — but even if it supports the field indirectly (e.g., workforce training), I'd still argue it exceeds AW corporate campaigns. — Anonymous participant, AW funding reasoning, beliefs form

Short term I believe funding CM will not exceed the benefit of AW campaigns. However long term (>10 years) it will, since research and development of this technology together with consumer adoption, will provide a longer-term benefit. I believe however that all these interventions (campaigns, development of alternatives) need to go hand in hand. — Anonymous participant, AW funding reasoning, beliefs form

If research obtains a real scale up of CM and this impacts meaningfully on intensive farming, CM would have a much higher impact, making it probably the best investment for AW. — Stefano Lattanzi (Bruno Cell), CM_02 AW reasoning

There was broad agreement on the conditional logic: CM investment has enormous expected AW value conditional on successful scale-up. The crux is the probability-weighted timeline and cost to that scale-up. Lattanzi's more pessimistic cost estimate does not translate to a pessimistic AW investment case — he assigns high value to CM if it succeeds, and sees the main risk as timeline rather than ultimate feasibility.

CM has broad and extensive potential for animal welfare benefit. The risks in/path to meaningful market adoption creates uncertainty as to extent of realized benefit on a given timeline, certainly, but we are past the point in which we question if this can be achieved. This benefit profile (impact x likelihood) is massively favorable when taken on the long term view because of just how significant the impact potential is. And despite significant investments to date, albeit largely to private companies, I would argue CM is relatively neglected as compared to investment profiles seen for other emerging technology spaces with comparable potential benefit impacts to society. — Matt McNulty (Tufts CCA), CM_02 reasoning (author-revised, June 2026)

David Reinstein outlined next steps: commissioning peer evaluations of the key TEAs (Humbird, Pasitka/Beefy-R, CE Delft, Goodwin et al. scoping review) through the PQ project; updating the cost model with beliefs gathered at the workshop; producing a post-workshop synthesis document; and circulating a follow-up beliefs update round to all participants within ~1 week to capture any shifts from S1 and S2 discussions.

Several participants expressed interest in contributing to formal evaluations. Elliot Swartz and Aleksandra Fuchs were identified as potential contributors on media cost components. Jakub Kozlowski had offered pre-workshop to present on model composition and forecasting uncertainty — this may be incorporated into a follow-up session or written synthesis.