INTRODUCTION

A new review finds that artificial intelligence works best on defined alternative-protein tasks, not autonomous product development. The distinction matters because cultivated meat still lacks the evidence needed to support wider deployment.

The assessment, published in Trends in Food Science & Technology and reported by vegconomist, covers cultivated, plant, insect and fermentation-based protein production. It considers research, manufacturing, formulation, nutrition, safety and consumer analysis. Its central argument treats food development as an interconnected system rather than a search for one suitable protein. AI can help researchers select candidates, monitor processes and interpret complex data. It cannot remove the need for experiments. That limit carries particular weight for cultivated food because the review finds much less work in this field than in plant or fermentation systems. Separate reporting adds a difficult commercial setting. Green Queen says European investment has shifted towards fewer alternative-protein companies, with fermentation taking most of the available capital. Food Ingredients First also reports political disputes over protein policy and evidence that terminology affects acceptance of cellular agriculture. Together, the sources present AI as a useful research instrument within a sector still constrained by finance, policy, manufacturing and public trust.

CORE FINDING

The review’s conclusion is narrower than the broad prospect of automating food research. According to vegconomist, the strongest evidence supports software that informs a particular choice inside an existing scientific workflow. Examples include ranking possible ingredients, adding functional context to protein structures and estimating process conditions that cannot easily be measured in real time. Other uses include optimisation within fixed boundaries, analysis of sensory results and interpretation of consumer data. These tasks reduce the field that scientists must examine. They do not establish whether the resulting food is safe, pleasant or practical to manufacture. The review therefore assigns experiments the final say. Its authors argue that model predictions must connect with food properties such as taste, texture, nutritional value, safety and consumer response, according to vegconomist. That standard is more demanding than showing that an algorithm performs well on a familiar dataset. For cultivated meat developers, the relevant test is whether a model improves a decision under actual production conditions.

Evidence differs sharply by production route. Vegconomist reports that published work clusters around plant proteins and products made through fermentation. Research involving cultivated meat remains comparatively sparse. The available work mainly addresses initial screening, descriptions of production behaviour and the creation of datasets. That leaves little support for claims that AI can yet manage cultivated-food development from cell selection to a finished meal. The counter-argument says sparse evidence may reflect the field’s youth rather than a weakness in the technology. That explanation is plausible, but it does not enlarge the evidence base. A model trained on limited records cannot prove that its output transfers to other cells, media, equipment or factories. The review consequently treats AI as a way to guide laboratory effort, not as a substitute for it. For a capital-intensive field, that remains useful. Reducing failed experiments has value. Removing verification would merely move the risk further down the production line.

EVIDENCE

Alternative proteins pose a difficult modelling problem because many components influence each other. Vegconomist says proteins can interact with fibre, starch, fats, pigments, salts, phenolic compounds, chitin and polysaccharides. Those interactions can alter colour, flavour, solubility, digestion and allergic response. Manufacturing adds another layer. Heating, extraction, fermentation, cell culture, extrusion and mechanical force can change a protein’s shape and the structure of the surrounding food matrix. Those changes then affect firmness, stability, aroma release and storage life, according to vegconomist. A result obtained with one raw material may therefore fail after a supplier, process or production scale changes. Conventional trial and error can investigate these links, but the review characterises that approach as expensive, slow and difficult to transfer. AI offers a way to inspect many variables together. The review does not claim that statistical pattern-finding explains the underlying food chemistry in every case.

Commercial investment provides a measure of where companies currently expect near-term progress. European alternative-protein businesses secured €236 million in private finance during the opening six months of 2026, according to Green Queen’s account of Good Food Institute Europe analysis using Net Zero Insights data. Green Queen says that sum was 56% above the equivalent period in 2025 and was Europe’s highest six-month result since early 2024. Yet the deal count fell by half, according to the same report. More money therefore did not mean broader access to capital. Fermentation businesses received 84% of the European total, Green Queen reports. The concentration broadly matches the review’s finding that fermentation has a deeper AI evidence base than cultivated meat. The sources do not show that AI caused investors to favour fermentation. They do show that technical evidence and finance are accumulating in the same part of the alternative-protein market.

The global comparison tempers the European rise. Companies in Europe collected more than three-quarters of worldwide alternative-protein investment during the first six months of 2026, according to Green Queen. Global funding nevertheless declined by 10%, from €341 million in the corresponding 2025 period to €306 million in 2026, the publication reports. Europe’s gain therefore took place inside a shrinking global pool. Precision-fermentation companies in the region obtained €100 million during those six months, exceeding their €97 million total for all of 2025, according to Green Queen. Biomass-fermentation businesses raised €99 million in the same 2026 period. Green Queen compares that with €61 million across the whole of 2025. Those figures help explain why fermentation offers more opportunities to generate process data, test models and validate results. Cultivated producers, by contrast, have fewer funded programmes from which to build shared evidence.

MECHANISM

One useful method identified by the review is soft sensing. This approach uses measurements that are available to estimate a production state that is difficult, costly or slow to observe directly. Vegconomist includes soft sensing among the AI applications with current evidential support. In a cultivated process, its value would rest on the accuracy of the available measurements and the relevance of the training data. The supplied sources do not name a particular cultivated-meat deployment, so claims about factory performance would go beyond the record. The broader mechanism is clear. A model looks for relationships between recorded signals and a target condition. Operators can then use that estimate to decide whether to adjust a process or collect a physical sample. Laboratory tests remain necessary to check the estimate. This arrangement keeps human judgement and measurement inside the workflow. It also reflects the review’s preference for bounded decisions over autonomous control.

Constrained optimisation follows the same logic. Researchers define the acceptable limits, the model compares possible choices, and experiments test the candidates it ranks. According to vegconomist, AI can also help map sensory information and draw patterns from consumer research. These applications matter because protein chemistry alone does not determine whether a product succeeds. Processing changes can affect how flavour is released, while other ingredients can alter colour or texture. A technically efficient cell or fermentation process may still produce an unsuitable food. The review therefore links prediction with endpoints that consumers and regulators can recognise. Food Ingredients First supplies a commercial example of a narrower formulation tool. Food ingredient supplier Ingredion introduced Ask Ingredion as a conversational AI service for research teams, the publication reports. The tool helps users choose ingredients and connects them with samples and specialist advice. That design supports the review’s case: software guides a task, while people and physical materials remain involved.

The review also identifies candidate ranking and protein annotation as supported uses, according to vegconomist. Ranking can direct scientists towards options that merit scarce laboratory time. Annotation can place a predicted structure beside information about how that structure may behave. Neither function proves that a protein will perform as expected in food. A useful prediction must survive changes in composition and processing. Cultivated systems add biological variability to that burden. Vegconomist says the limited cultivated-meat literature has concentrated on screening and process description rather than complete product development. Data generation is itself one of the reported activities. That point matters because models depend on records that represent the conditions in which they will operate. Generating consistent data may be less conspicuous than launching an automated platform. It is also the prerequisite for judging whether such a platform works.

COMMERCIAL CONTEXT

Cultivated-meat finance remains modest beside fermentation. European businesses in the category raised €18 million during 2026 before July, according to Green Queen. The publication compares that with €20 million across all of 2025 and says the latest result remains below the sector’s 2023 high. British cultivated pet-food startup Meatly accounted for €12 million of the 2026 amount, while German cultivated-meat startup Innocent Meat secured €6 million, Green Queen reports. Those two rounds equal the reported regional total. The concentration leaves little basis for treating the increase as a broad recovery. Helene Grosshans of Good Food Institute Europe argues that funding sources must expand if producers are to reach commercial operation, according to Green Queen. The AI review points to a related technical need. Tools must work beyond the dataset, instrument or facility in which developers created them. Both problems meet at scale-up, where a company must finance new equipment while proving that its process remains predictable.

Public support is already carrying part of the burden elsewhere in alternative proteins. Fermentation companies received €67 million in grants during the first six months of 2026, according to Green Queen. The entire European category, including cultivated and plant-based businesses, had obtained €45 million in the equivalent 2025 period, the publication reports. Gas-protein producer Solar Foods also secured a €78 million package from Business Finland for a commercial-scale Solein plant, according to Green Queen. The package illustrates the use of public finance alongside other capital rather than reliance on venture investment alone. The European Innovation Council supplied fermented-dairy protein maker Vivici with €12.5 million through equity and grant support, Green Queen reports. Mycelium-food developer Adamo Foods led a consortium that received a €10 million grant from the EU-backed Circular Bio-Based Europe Joint Undertaking, according to the publication. These are fermentation and mycelium examples, not evidence of equivalent support for cultivated meat.

Good Food Institute Europe gathered more than 40 investors to discuss alternative ways to fund production, according to Green Queen. Participants concluded that agrifood ventures do not fit a model built around rapid growth and strong returns within a few years, the publication reports. Their proposed answer combines grants, debt and equity so that several funders share the risks of building facilities. Grosshans argues that public funding can encourage private investment rather than displace it, according to Green Queen. That proposal addresses capital structure, not technical validation. Even so, it bears directly on the review’s conclusions. Reliable AI depends on data from varied production conditions. Companies need access to factories and demonstration lines to obtain such records. Investors, in turn, want evidence that a process works outside the laboratory. Blended finance may help create the equipment needed for validation, but the supplied sources provide no result showing that it has improved cultivated-meat models.

POLICY AND RECEPTION

Policy does not yet give every protein route equal support. Food Ingredients First reports that the European Union’s Protein Plan set a production objective for protein used in animal feed but omitted an equivalent goal for plant protein eaten by people. Plant-based groups criticised that choice, according to the publication. The plan appeared beside the EU Livestock Strategy, Food Ingredients First reports. This dispute concerns plant food rather than cultivated meat, but it shows how policy can direct infrastructure and demand towards particular uses. The World Business Council for Sustainable Development separately said companies were increasing spending on protein diversification, according to Food Ingredients First. The council identified China, the European Union and the United States as leading markets. It also called for clearer rules, aligned regulation and financial incentives. These requests sit outside the AI review’s technical analysis. They matter because a validated tool has little commercial value if the product faces an uncertain route to market or lacks production support.

Public language presents a different constraint. A Tufts University study found that consumers in the United States and Germany preferred cellular-agriculture meat described as cultured or cultivated over the term lab-grown, according to Food Ingredients First. The supplied account gives no sample size or effect magnitude, so the strength of that preference cannot be assessed here. The finding still identifies terminology as a variable in consumer research. That makes it relevant to the review’s discussion of AI-based consumer analysis. A model may detect patterns in survey or market data, but its output will depend on the labels, questions and populations used to create those records. Food Ingredients First concludes that trust and branding may matter as much commercially as the underlying production method. The technical review offers a compatible warning from another direction: results tied to one dataset may not carry into another setting. Neither source shows that AI can resolve distrust by itself.

LIMITS

Weak data practice remains the main technical obstacle. Vegconomist says the review identifies fragmented records, limited standardisation, poor transfer between settings and inadequate validation. Integration with factory needs and regulatory demands is also deficient, according to the publication. Small changes in raw inputs or operating conditions can alter finished-food behaviour. A model that works with one instrument or site may consequently fail elsewhere. The counter-argument holds that larger datasets and more advanced models will solve this problem. The review does not support that conclusion on its own. More records help only when measurements are comparable, endpoints are relevant and experiments verify the prediction. The authors instead call for better data systems, benchmarks designed to test performance under change, and workflows checked by people, according to vegconomist. Cultivated meat starts from a weaker position because its published applications remain concentrated in early research.

The sources also leave several questions unanswered. Vegconomist does not provide counts of the studies examined, model error rates or comparisons between AI-assisted and conventional development. Food Ingredients First gives no numerical result for the Tufts University terminology study. Green Queen reports investment flows but does not connect any cultivated-meat round with a specific AI programme. The evidence can therefore establish where researchers use AI and where capital is moving. It cannot calculate the savings from AI, prove that a tool performs across factories or show that investors reward its adoption. Nor does the supplied material identify a regulatory authority that has accepted an AI-derived result for cultivated food. These gaps do not make the tools useless. They set the boundary between demonstrated decision support and an untested claim of end-to-end automation.

CONCLUSION

The review supports AI as disciplined research equipment, not an autonomous cultivated-food developer. Progress should be judged by validated food outcomes and transfer between facilities; anything less is a model performing for its own dataset.