The Good Food Institute’s APAC office published a report in July 2026 arguing that China’s industrial fermentation sector is pivoting into food protein, and that the resulting capacity is dramatically cheaper to build than its Western equivalent. The report is careful, useful, and largely correct about the direction of travel.
Its headline cost number is not a finding. It is a rule of thumb, self-reported by the companies selling the capacity, corroborated by one speech that is not publicly available.
And GFI’s own technical analysis, published three years earlier, concluded the opposite of the premise this report is built on.
The number everyone is quoting
Here is what China’s Biomanufacturing Boom actually says, on page 17, in full:
“Construction of biomanufacturing facilities is much faster and more affordable in China. A general rule of thumb expressed by Chinese companies interviewed for this report is that the USD value of facility construction in the U.S. will cost roughly the same number in RMB in China (e.g. a US$100 million facility in the U.S. will cost RMB100 million, equivalent to ~US$15 million, in China). A presentation from Cathay Biotech, a company that makes a wide range of biomanufactured products, suggests capital expenditure costs in the U.S. and Western Europe are substantially higher than China.”
Three observations.
First, it is a currency-substitution heuristic, not a cost estimate. The claim is that the numeral stays the same and the symbol changes — implying a capex differential of roughly six to seven times. It reads oddly because it is a mnemonic, not a measurement.
Second, its provenance is the party with the strongest commercial interest in it. The report attributes it to “a general rule of thumb expressed by Chinese companies interviewed for this report” — vendors describing the price of their own product.
Third, the corroborating source is unverifiable, and the report says so. Footnote 9 to that sentence reads:
“The speech is not publicly available, but elements of it have been reproduced in the link in the paragraph. The graph in the article does not give specific numbers, but the size of the capital expenditure bands for the U.S. and EU are several times the size of those for China.”
A non-public speech containing an unlabelled bar chart. That is the entire evidentiary base for the most-repeated number in this story.
How it degraded in transmission
Green Queen’s summary of the report renders it as: “Experts suggest that construction costs in the US and China are numerically similar.”
The figures are transcribed accurately — “¥100M ($15M)” is exactly what GFI wrote. The distortion is attributional. “Companies interviewed for this report” became “experts,” which converts a vendor self-report into an independent expert assessment. vegconomist’s version of the same story is notably more careful.
This is worth naming precisely because it is not fabrication. It is one adjective of drift, in one clause, and it changes the epistemic status of the claim completely.
Practical consequence: if a China capex advantage is load-bearing in your scale-up model, it needs a quotation from an engineering firm against a defined scope — not this number. Treat six-to-seven-times as an unpriced hypothesis.
To its credit, the report refuses the other big number
The single most-quoted statistic on this beat is that China holds 70% of the world’s fermentation capacity. GFI APAC declines to use it, in footnote 1 on page 7:
“While it is widely referenced (e.g. SCSP, SCIE) that China possesses 70 percent of the world’s fermentation capacity, the underlying references all link to each other and the original source of this claim is hard to locate and verify.”
That is an unusually honest piece of citation hygiene and it should be quoted more often than the capex line. It also sits awkwardly beside the report’s willingness to print the capex heuristic without equivalent hedging.
The methodology gap
The report has no methodology section. There is no stated interview count, no sampling frame, and no data-collection window. The only methodological statement is in the acknowledgements, thanking unnamed “interviewees.”
From footnotes and body text, roughly ten to twelve identifiable interview sources can be reconstructed, clustered in April 2026: the CEO of NXW, representatives of JSPC, More Meat (Guangzhou), Fushine Biotech, two companies cited as confidential, a Hunan biomanufacturing company interviewed in 2025, and an unnamed Chinese venture capitalist.
This matters because the report makes population-level claims on that base: “Every company interviewed for this report had high confidence that the government will act quickly,” and “The companies interviewed for this report are confident they can reduce their costs significantly with scale.” Both are statements about an undisclosed number of self-selected, commercially interested parties.
The quantitative work is better sourced. Figures 5 and 6 — R&D spend as a share of revenue and R&D staff ratios for eight listed bioindustrial firms, 2021 to 2024 — are credited to a team at the Australian National University with all data taken from company annual reports. Price series cite UNdata, EFFAMALL and China’s National Bureau of Statistics.
The economic driver is real and well evidenced
Nothing above disputes the report’s core mechanism, which is documented properly.
Chinese producers hold roughly 38% of global vitamin exports and 50% of global amino acid exports as of 2024. European amino acid prices collapsed between January 2022 and November 2025: L-threonine from about €4.6 to €1.1 a kilogram, L-valine from about €8.5 to €2.3, 70% L-lysine sulfate from about €3.6 to €1. China recorded 41 consecutive months of producer price index decline, with the share of Chinese companies reporting losses rising from 17% in 2021 to 25% in 2025.
On top of that, biopharma capacity expanded while biosimilar competition intensified. The report’s formulation: “the attractiveness of some pharmaceutical manufacturing markets has lessened.”
Policy reinforces it. An August 2025 list of 36 target biomanufactured products includes ten foods or food additives, three of them directly alt-protein. Heilongjiang has offered 30% reimbursement of total investment for pilot facilities since February 2026.
So: margins compressed in the incumbent business, the state has designated the destination, and capital is moving. That is a genuine and important story.
It is simply not the same story as “existing pharma tanks are being converted to food.”
What the report actually documents being built
Figure 1 is titled “Examples of recent alt protein facilities and partnerships in China (publicly disclosed).”
| Company | Partner | Product | Current | Planned |
|---|---|---|---|---|
| Fushine | None stated | Mycoprotein | 1,200 t | 20,000 t |
| More Meat | Guangzhou Shuangqiao | Mycoprotein | not stated | 20,000 t |
| Angel Yeast | None stated | Yeast protein | 11,000 t | 20,000 t |
| Calysta | Adisseo | Microbial protein | 20,000 t | not stated |
| Jupeng Bio | Inner Mongolia Yitai | Microbial protein | 20,000 t | not stated |
| Zhonggu Junchuang | None stated | Mycelium protein | not stated | 18,500 t |
| NXW (Nutrition From Water) | JSPC | Microalgae protein | “commercial-scale production run completed” | not stated |
| Nourish Ingredients | CABIO | Precision-fermented fats | capacity to formulate 170,000 t of finished products | not stated |
Read down that column and a pattern emerges: not one entry is a converted pharmaceutical fermenter.
Angel Yeast’s 11,000-tonne AngeoPro line — the best-documented item, confirmed operational on the company’s own site in November 2025 at above 80% protein — is biomass fermentation on Angel’s existing yeast platform. Fushine’s 1,200 tonnes is described on its own website as a pilot production line, completed September 2023; the 20,000-tonne figure appears there as something the company expects “in the near future.” Calysta/Adisseo is a pre-existing animal-feed joint venture whose investment dates to the 2010s. Zhonggu Junchuang uses solid-state fermentation, which shares essentially no equipment with submerged liquid fermentation, pharma or otherwise.
The remaining entries are contract-manufacturing arrangements in which the asset is never described.
Two figures that need unpacking before reuse
Nourish Ingredients and CABIO: “170,000 tonnes” is not a capacity. CABIO’s own release of 9 April 2025 says the achievement enables “the supply of Tastilux®, a high-performance meaty fat that meets demand for 170,000 tonnes of end-product at minimal inclusion rates.” That is finished-product formulation demand, not fat produced. Because inclusion rates are minimal, the actual fermented-fat tonnage is undisclosed and is a small fraction of it. GFI’s figure caption words this correctly; secondary coverage tends to compress it into a headline capacity number. The accompanying “1,700% increase in production capacity” is a percentage off an undisclosed base.
Fushine: a 200,000-tonne figure is circulating and is almost certainly a unit error. The regulatory consultancy CIRS states the company is “constructing a Phase I project with an annual output of 200,000 tons.” Fushine’s own English-language site says 20,000. The likely explanation is a dropped 万 (ten-thousand) character. The wrong figure has already propagated into search summaries. Use 20,000, and mark it aspirational.
Practical consequence: for every Chinese capacity number, establish three things before using it — installed or planned, product tonnage or formulated-product tonnage, and whether the source is the company or a third party paraphrasing it. This is the same basis problem we have documented across alternative protein figures, and it is worse here.
The six-tonne tank
The report’s most-cited proof of concept is the partnership between NXW (Nutrition From Water) and JSPC, the Jiangshan Pharmaceutical Company — a global leader in vitamin C, formerly DSM-owned, with DSM exiting in 2024. The partnership was announced in March 2026 after a December 2025 technology transfer.
The report states the partners “completed a commercial-scale trial production run in a six-tonne fermentation tank in May 2026.”
Six tonnes is roughly six thousand litres. For scale, the same report cites Marlow Ingredients in the UK as operating the world’s largest mycoprotein facility at over 67,000 tonnes a year. A six-thousand-litre vessel is a pilot-to-demonstration asset. Describing a run in it as “commercial-scale,” without qualification, is exactly the kind of elastic vocabulary that makes capacity claims in this sector impossible to compare.
What did transfer in that partnership is described precisely, and it is instructive. The JSPC technical contribution is listed as “improvements to processes such as spray drying, steam use efficiency, and increased density.” Those are commodity-fermentation unit operations learned from making vitamin C and amino acids. The cost anecdotes elsewhere in the report are about steam prices falling from 320 to 220 yuan a tonne, land, wages, and a 4.6 MW photovoltaic project.
That is a real competitive advantage. It is process engineering and utility economics — not bioreactor reuse.
GFI’s own earlier analysis says the tanks do not transfer
This is the part that has gone unremarked. In June 2023, GFI published a capacity study with Integration Consulting. Its conclusions run directly against the 2026 report’s framing, and the 2026 report does not cite it.
On retrofitting:
“Success cases show that retrofitting used equipment for fermentation is feasible from a technical perspective, particularly for solid and liquid biomass fermentation… However, due to the complex technical requirements of precision fermentation equipment, experts are skeptical of the economic viability of retrofitting for precision fermentation at all scales.”
On downstream processing:
“Compared to the broad potential held by core fermentation equipment, DSP equipment has rarely proven to be suitable for retrofitting.”
And on pharma-adjacent capacity specifically — the direct answer to the 2026 thesis. The 2023 study put globally available capacity at approximately 16 million litres across 89 in-house producers and food-exclusive contract manufacturers. A further 53 contract manufacturers can make food-grade product but also serve pharmaceutical and industrial customers. Of those:
“These categories typically involve manufacturing higher-margin products (up to 70% profit margins for pharmaceuticals). This poses a challenge for fermentation-derived product manufacturers in terms of accessing these facilities, suggesting that the estimated functional capacity is much closer to the lower bound of 16 million liters.”
That is GFI concluding that pharma-capable fermentation capacity is functionally inaccessible to food because of margin competition.
The 2026 APAC report’s implicit answer is that the margin gap has now narrowed enough to unlock exactly that capacity — which is a coherent argument, and the PPI and biosimilar data support it. But the report never makes the argument explicitly, never cites the earlier finding, and never addresses the equipment objections. Two GFI publications, three years apart, reaching opposite conclusions about the same asset base, with no reconciliation between them.
Note also that the parallel industries the 2023 study identifies as most suitable for retrofit are beer breweries, ethanol plants and wineries. Not pharmaceutical plants.
Why over-specified equipment is a liability, not an asset
The engineering case against pharma reuse is not that the equipment is inadequate. It is that it is too good, and the food product cannot pay for it.
CRB’s Sebastian Bohn puts it directly:
“Biopharma equipment and facilities are expensive and designed for a heavily regulated industry where manufacturers are required to adhere to much stricter standards… That works when kilograms of a drug can be sold for billions of dollars. But it doesn’t make sense when producers are hoping to reach a target price for a pound of cultivated protein that is the same as, or lower, than that of traditional meat or dairy.”
His quantified version: “For manufacturers to be commercially viable in the alternative protein space, the cost of a bioreactor has to be one-third of a conventional, pharma-grade bioreactor.” And on the compliance overhead: “Food facility inspectors are not looking for the same types and amounts of certifications required by pharma inspectors.”
Redundancy, validation burden, clean-steam utilities and revalidation cycles are capital and operating costs that a food P&L cannot amortise. Inherited pharma configuration shows up on the wrong side of the ledger.
GFI Europe’s July 2025 modelling gives the technically precise version of the same point, and it does so on a single molecule. It treats “food-grade purity (~95% ±10%)” and “high purity (>99%)” as separate downstream configurations with separate cost structures, then models alpha-lactalbumin under both. Same protein, same 20 g/L titre at harvest, same 80 m³ culture volume:
| Purity target | Base cost | With 50% improvement in four cost drivers |
|---|---|---|
| Food-grade (~95% ±10%) | $24/kg | $12/kg |
| High purity (>99%) | $100/kg | $86/kg |
A four-fold cost difference on the same fermentation, created entirely downstream. A pharma purification train is built to reach the bottom row using chromatography steps the top row does not need and cannot pay for — which is precisely why downstream is the part that does not transfer, and why inheriting one is a cost, not a saving.
Practical consequence: when evaluating a Chinese contract manufacturer, price the downstream train separately from fermentation. Fermentation capacity is the part that is plausibly cheap and available. Purification to food-grade at food economics is the part to diligence.
The one clear case of pharma taking food work
There is a genuine example, and it deserves stating fairly because it cuts the other way.
All G’s chief executive Jan Pacas told AgFunderNews in December 2025: “We have already started the technology transfer for the CMO in China, which manufactures for some of the largest global pharmaceutical companies and has a lot of expertise in our organism.” All G was the first company globally to receive Chinese approval to sell recombinant bovine lactoferrin, in late 2024.
That is a large pharmaceutical contract manufacturer taking food work, on the record. Two caveats: the partner is unnamed, and lactoferrin is a high-value protein whose economics are far closer to pharma than to bulk dairy protein. It is the exception that shows where the boundary sits, not evidence the boundary has moved.
Notably, All G describes purifying lactoferrin “through the same chromatography process used by the dairy industry.” The downstream analogue it reaches for is dairy, not pharma.
The regulatory clock, with one data point
China’s approval route runs through the National Health Commission with risk assessment by the CFSA. The Fushine mycoprotein approval is fully dated by the consultancy that filed it:
| Date | Step |
|---|---|
| 7 February 2025 | NHC accepts application, acceptance no. 卫食新申字(2025)第0010号 |
| 10 July 2025 | CFSA releases draft for public comment |
| 27 November 2025 | NHC Announcement No. 7 of 2025 — approved |
Nine months, submission to approval, for mycoprotein from Fusarium venenatum strain A3/5 — the Quorn lineage.
That is fast. It is also the easiest possible case: an ingredient with over forty years of consumption history and existing approvals in the EU, US, Canada and Australia/New Zealand. One data point on the easiest molecule is not a statutory timeline, and GFI’s own framing is softer than the trade-press version — companies expressing “high confidence that the government will act quickly” is a sentiment, not a track record.
One loose end: the August 2025 target list names Fusarium brachygibbosum protein against Fushine, while the November approval is for Fusarium venenatum. The report does not reconcile these and neither can we.
The counter-argument
The strongest case for the China thesis does not depend on the capex number at all.
Capital allocation by fermentation incumbents is a more durable advantage than cheap steel. Angel Yeast did not convert a pharma tank; it built a purpose-built 11,000-tonne protein line using decades of yeast-fermentation operating knowledge, and it is running now. That is exactly how the vitamin, amino acid, solar and battery industries were won — not by retrofitting, but by incumbents with adjacent process mastery deciding to build new capacity at scale and accepting thin margins to hold it.
On that reading, the equipment objections are beside the point. Nobody needs to convert a pharma fermenter if the same firms can build new food fermenters faster and cheaper than anyone else, and are willing to. The NDRC’s own 2026 annual report makes the intent explicit: “In emerging industries, we will allow appropriate surplus capacity and encourage competition and innovation.”
We think this is right, and it is the argument the report should have led with. It also carries an uncomfortable implication for anyone planning to buy that capacity: the mechanism that will make it cheap is subsidised overcapacity — the same dynamic that just destroyed amino acid and vitamin margins and caused this pivot in the first place. GFI notes that “the only way to get sustained surplus capacity is with some level of subsidisation,” and that large capital subsidies are “discussed behind closed doors.” The cause of the pivot is the likely outcome of it.
The risks the report does document
To be clear about what Biomanufacturing Boom gets right, its risk section is candid:
- Investor resistance. Several foreign companies said Chinese manufacturing plans were “a red line for many U.S. investors.”
- Trade defence, with precedent. The EU applied a roughly 50% anti-dumping tariff on lysine in 2025 for five years, and 34% to 233% on erythritol. India imposed duties up to $20.87 a kilogram on vitamin-A palmitate from China.
- Export controls. Under the FY2026 National Defense Authorization Act, federal agencies are barred from procuring biotech equipment or services from designated companies of concern, including BGI, MGI, Complete Genomics and WuXi AppTec.
- Import substitution cuts against the export thesis. China introduced beef import quotas in early 2026. The report’s own conclusion: “The political direction is clear, and it will apply to alt proteins as much as conventional foods.”
- IP. The recommended structure has pre-existing IP staying with the foreign partner, but “process improvements generated during scale-up work may be owned by the Chinese company.” In a business where scale-up process knowledge is the moat, that is a material concession presented as standard practice.
- Presence costs. Successful scaling “will require monthly visits to China for several years, or a local presence.” JSPC’s Amy Shi: “Fermentation scaleup is ultimately a physical process, and many of the practical constraints and opportunities only become apparent when scientists and engineers work side by side.”
That last item is an unpriced line in every model built off the capex heuristic.
Also worth noting: all companies interviewed are selling business-to-business, and the report concedes the products are “not necessarily one-for-one replacements” for animal products, including applications in fortification and animal feed.
What we could not establish
- Whether the More Meat 20,000-tonne facility began commercial runs in June 2026. The report states it in the past tense but sources it to an April 2026 interview, meaning it was a projection at the time of writing. We found no independent confirmation.
- Whether Fushine added a 20,000-tonne facility in June 2026, as the report’s body text states. The company’s own timeline, whose last entry is September 2025, does not record it.
- The source of Angel Yeast’s “planned 20,000 t.” The company’s release states 11,000 tonnes operational and only “built-in expansion capacity,” with no number. The report gives no citation for the larger figure.
- Jupeng Bio’s 20,000 tonnes and Zhonggu Junchuang’s 18,500 tonnes. Both are single-sourced to the report’s own figure with no corroboration available.
- The exchange rate underlying “RMB100 million, equivalent to ~US$15 million.” That implies roughly 6.67 yuan to the dollar. We did not verify a contemporaneous rate, so the implied multiple should be read as approximately six to seven times, not precisely 6.67.
- A reported “60,000-litre scale” for All G. It appeared in a search summary and on no page we fetched. We have not used it.
- Any disclosed capacity for JSBio, which appears in trade coverage only as a photo credit and does not appear in the GFI report at all.
What to watch
- Whether GFI reconciles its two positions. The 2023 capacity study and the 2026 APAC report cannot both be right about the accessibility of pharma-adjacent capacity. A 2026 fermentation State of the Industry report is the natural place for that.
- Whether any named Chinese facility publishes an audited utilisation figure. Nameplate capacity is being disclosed freely; utilisation is not disclosed by anyone, anywhere.
- Whether the six-tonne NXW run is followed by a run at least an order of magnitude larger within twelve months. That is the single cleanest test of whether the technology transfer worked.
- Whether a Western company publishes a like-for-like capex comparison against a defined scope. Until one does, the six-to-seven-times figure remains a rule of thumb.
Our position, recorded so it can be checked: China’s advantage in fermentation-derived food protein is real, and it is being built out of new purpose-built capacity, incumbent process engineering and subsidised industrial policy — not out of repurposed pharmaceutical tanks. If we are wrong, the evidence will be a named pharmaceutical fermenter, with a stated volume, producing a food protein at commercial rates. As of today, no such example appears in the report that made the claim.