How AI is Affecting Farmed Aquatic Animals. Part 2: Deployment

A new report from Rethink Priorities analyzes the deployment of AI in aquaculture, finding that AI tools are primarily used for high-value species like salmon and shrimp in the US and Norway, with adoption driven by stock and growth management needs. The report identifies 91 companies with AI-enabled technologies and examines barriers to adoption, including cost and technical challenges, with deployment expected to grow in the next five years.

| Artificial intelligence AI introduces new capabilities to animal agriculture that could alter production methods, economic structures, and animal welfare outcomes. Responding strategically requires an understanding of how quickly such changes will unfold, whether they will benefit or harm animal welfare, and what interventions will remain relevant. In this three-part series, we take a close look at how AI will be used over the next five years in aquaculture, which collectively farms hundreds of billions of animals each year for food. This report analyzes the current state of AI deployment in aquaculture, examining where AI-aquaculture tools are being used in practice, the species and regions with the greatest AI presence, what is driving and constraining adoption, and how deployment is expected to evolve. The final report will address Welfare Effects . Key findings from part two of this series: This report analyzes the current state of AI deployment in aquaculture. In the previous report, we identified which AI aquaculture products currently exist. In this report, we examine where those products are being used and to what extent their deployment may change in the short term roughly the next five years . To do this, we assessed deployment evidence for 66 aquaculture-specialist companies through targeted web searches, supplemented by three expert interviews and a review of 16 academic papers on barriers to adoption. This report is the second in a three-part series examining how AI is affecting farmed aquatic animals. In Part 1 of this series, we analysed the current state of AI innovation in aquaculture and found 91 companies with AI-enabled technologies that have direct implications for farmed animal welfare. You can see the database here https://docs.google.com/spreadsheets/d/1Jp-wd3rXnNpo ZheMrCKaDeNFNzsWRRb5DjtxCHCPhg/edit?gid=0 gid=0 . We found that innovation is concentrated primarily in the US and Norway, predominantly targets high-value species such as salmon and shrimp, and most commonly addresses stock and growth management. Part 2 builds on this foundation by examining where and to what extent those tools are being deployed. This report draws on three sources of evidence: a desk-research investigation into the deployment of AI tools from the list of companies we identified in Part 1 https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-1/ , three expert interviews, and a Claude-assisted review of academic literature on barriers to AI adoption. For the deployment analysis, we took the 91 companies identified in Part 1 and assessed where each had evidence of deploying their products see our database https://docs.google.com/spreadsheets/d/1Jp-wd3rXnNpo ZheMrCKaDeNFNzsWRRb5DjtxCHCPhg/edit?gid=0 gid=0 To supplement the desk research, we conducted interviews with three experts with direct experience of AI deployment in aquaculture. Expert backgrounds are described in the Appendix https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640- jmcclm6ci713 . The interviews covered current deployment patterns by species and region, barriers to adoption, and expectations for near-term growth. Given that two of the three experts specialize in salmonids and one in shrimp, findings from the expert interviews are most reliable for these species and should be interpreted with caution when generalized to others. To examine barriers to AI adoption, we conducted a timeboxed three-hour search for academic literature and retained 16 papers for which the full text was available—13 focused primarily on aquaculture and three drew on broader agricultural contexts. We used Claude to extract barrier themes from each paper and compared these against expert views, selecting themes with the strongest consensus across both sources. Further details on this process are provided in the Appendix https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640- zdx9fecjpg53 . This report does not constitute a comprehensive review of AI deployment in aquaculture. Our desk research identifies deployment evidence that is publicly available in English-language sources, which likely understates deployment activity in regions with limited online presence or where companies publish primarily in other languages. Expert interview coverage is heavily weighted towards salmon and shrimp, and towards Europe and Latin America. See other ways we could be wrong https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640- k5gatnnctdem below. Figure 1 : Distribution of 91 companies making AI-enabled tools for aquaculture, broken down by degree of company specialization. Of the 91 companies we found in Part 1 that produce AI-enabled tools for aquaculture, we found that over three-quarters of companies specialized in products solely for aquaculture ~70% , or aquaculture with minor applications in a closely related field ~6% see the Appendix https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640- 5dr6jmt07l73 for definitions . For three companies in the aquaculture-only category, we could not find any deployment information, leaving 66 companies for the deployment analysis that follows. To map where these 66 companies deploy their products, we recorded every country in which each company had evidence of deployment. Our inclusion criteria for deployment locations are defined in the Appendix https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640- x25gffa1f5pu . Among the 66 companies specializing in AI-aquaculture technology, we found 276 company-country pairs in total. A company-country pair is a unique combination of one company and one country. A company deploying to three countries contributes three pairs, regardless of how many species or farms it targets in each. | |---| The distribution of company-country pairs is shown in Figure 2 https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640-fig Regional Deployment . Figure 2 : Geographic distribution of AI-aquaculture technology deployed by 66 companies, counting each company-country pair once regardless of species targeted. Represents deployment to 71 unique countries from companies with headquarters in 31 unique countries. The most frequently represented regions were Europe, Latin America, and Southeast Asia , with 98, 60, and 37 company-country pairs across 26, 13, and 9 deployment countries, respectively. Within Europe, Norway and Scotland accounted for the largest shares, with 24 and 14 company-country pairs, respectively. Within Latin America, Chile dominated with 20 company-country pairs present. Outside these three regions, the US and Australia recorded notable presences with 15 and 11 company-country pairs, respectively. Figure 3 https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640-fig sanky maps the headquarters locations of 66 aquaculture-focused AI companies against their deployment countries. This illustrates that AI aquaculture tools have a larger global reach than only the countries in which they are produced, and that Figure 3 : Sankey diagram, created on SankeyMATIC , showing the headquarters location left of 66 aquaculture-focused AI companies and where their products are deployed right , independent of species. The width of each flow is proportional to the number of company-country pairs. For example, a company deploying AI products to three countries counts as three pairs. There are a total of 265 company-country pairs across the 66 companies, where five of the included companies hold dual headquarters, resulting in 276 headquarters-to-deployment pairs mapped on the diagram. Numbers on the left reflect the total company-country pairs for companies headquartered in that region; numbers on the right reflect the total company-country pairs for deployments received by that region. denotes location of the company headquarters. Norway and Scotland are the largest net exporters by absolute net flow. Norway is the headquarters of 14 companies in our dataset. These companies generate 47 company-country pairs in total, of which 13 are within Norway itself, and 33 are abroad. Scotland is the headquarters of three companies in our dataset. These companies generate 32 company-country pairs in total, of which three are within Scotland itself, and 29 are abroad. Europe excluding Scotland and Norway and Latin America excluding Chile are the largest net importers by absolute net flow. Europe excluding Scotland and Norway has 60 company-country pairs, of which 12 involve companies headquartered within the region, and 48 involve companies headquartered elsewhere. Latin America excluding Chile has 40 company-country pairs, of which eight involve companies headquartered within the region, and 32 involve companies headquartered elsewhere. This picture is supported by experts, who reported AI adoption to be most advanced in Europe and Latin America. Chile has the highest number of foreign-headquartered companies deploying there in our dataset, and no domestically-headquartered companies. Experts corroborated that Chile currently imports most of its AI systems , adding these imports are primarily from Norwegian companies. In our own dataset, eight of the 20 companies present in Chile are headquartered in Norway—more than from any other region or single country 40% deploying to Chile. Southeast Asia is the most self-sufficient region , with 24 of the 38 company-country pairs generated by Southeast Asian-headquartered companies deployed within the region itself. To assess how deployment varies by species, we count the number of companies deploying AI-aquaculture tools in each country for each species separately. A deployment instance is a unique combination of one company, one country, and one species. A single company operating in 10 countries, and 10 companies each operating in one country, would both contribute 10 deployment instances for a given species. A high deployment instance count for a given species may therefore reflect many companies targeting that species, a wide geographical spread of products targeting that species, or both. | |---| Deployment instances are comparable across species and countries, but their sum across both represents only total deployment activity across the dataset and cannot be decomposed into a count of companies, countries, or species. Shrimp and salmon have the broadest deployment footprint in the dataset in absolute terms 136 and 131 instances, respectively , followed by trout 88 , sea bass 69 , sea bream 70 , tilapia 56 , and yellowtail 56 . 2 https://forum.effectivealtruism.org/feed.xml fngdeulkf0r1v Table 1 https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640-tab country species shows the top targeted species, as defined by deployment instances, for the five countries with the highest number of company-country pairs as shown in Table 1 : Top-ranked targeted species by deployment instances for the five countries with the highest number of company-country pairs. Where species are tied, all are shown. Top targeted species by number of deployment instances | ||| |---|---|---|---| Country | 1st | 2nd | 3rd | Norway | Salmon 20 | Trout 9 | Sea bass/sea bream 7 | Chile | Salmon 17 | Trout 8 | Sea bass/sea bream 7 | US | Shrimp/salmon 8 | Trout/yellowtail 6 | Sea bass/sea bream 4 | Scotland | Salmon 12 | Trout 8 | Shrimp/charr/sea bass/sea bream/tilapia 3 | Australia | Salmon 5 | Shrimp 4 | Sea bass/sea bream/tilapia 3 | In Part 1 https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-1/ of this series, we found that salmon and shrimp co-ranked as the top-most targeted species by AI-aquaculture product innovation. We find that the picture for salmon is reflected in deployment: salmon has the highest overall AI-aquaculture presence and dominates the top spot for most targeted species in every country. For shrimp, the numbers suggest differently: despite an equally large total deployment footprint, shrimp reaches top rank only in the US, where it ties with salmon. The following sections list the countries with the highest salmon and shrimp deployment instances in our dataset. These five countries account for ~50% of AI-aquaculture tools deployed targeting salmon 65/131 deployment instances across 44 countries . Consistent with our data, experts noted that highest adoption rates are seen in Chile, Norway, and Scotland, though this is in part expected given that countries with the largest Atlantic salmon farming industries could reasonably be anticipated to show the highest absolute deployment regardless of adoption rates. The experts reported that Norway leads in AI testing, attributing this to the ability of salmon farming companies to absorb the cost premium of current AI systems. Conversely, they suggested that Chile leads in adoption due to a younger UNCTAD, 2006 https://unctad.org/system/files/official-document/iteiit200512 en.pdf , more hierarchical management structure than Norway Experts estimate that AI technology is used by approximately 15% of all salmon producers, rising to upwards of 75% when looking at only top salmon producers. 5 https://forum.effectivealtruism.org/feed.xml fnorvhhs3f07g Figure 4 : Geographic distribution of AI-aquaculture deployment instances targeting salmon across 44 countries, drawn from a dataset of 66 companies. Graph shows 131 deployment instances, where 1 count represents one company present in one country for that species. These 7 countries account for ~37% of AI-aquaculture tools deployed targeting shrimp 50/136 deployment instances across 52 countries . Consistent with our findings, experts ranked Ecuador first by some distance. One expert further specified Mexico, Brazil, and Vietnam as other developing markets. Notably, despite the US ranking joint first with Ecuador for shrimp deployment instances in our dataset, it was not identified by any expert as a significant shrimp AI-adoption market. This discrepancy may simply reflect the scale of US shrimp production: one expert noted that US shrimp production is negligible, suggesting that even if adoption rates of AI technology among US producers are relatively high, it may not translate into the US being considered a significant AI-adoption market in global terms. Experts estimate that AI technology is used by approximately 5–10% of all shrimp producers and up to 20–30% among top shrimp producers. 6 https://forum.effectivealtruism.org/feed.xml fnnwaizb4r6t Figure 5 : Geographic distribution of AI-aquaculture deployment instances targeting shrimp across 52 countries, drawn from a dataset of 66 companies. Graph shows 136 deployment instances, where 1 count represents one company present in one country for that species. Of the 66 companies included on the graphs, 27 target salmon and 26 target shrimp, meaning the average salmon company is deployed in ~4.9 countries, and the average shrimp company in ~5.2 countries. However, the average deployment instances per company do not capture the evenness of the spread across countries. To quantify this geographic concentration, we calculated the Gini coefficient for salmon and shrimp deployment instances. The Gini coefficient is a measure of concentration ranging from 0 to 1, here applied to the geographic distribution of deployment instances across countries. A value of 0 indicates perfectly equal distribution across all countries where companies operate; a value of 1 indicates complete concentration in a single country. Salmon has a Gini coefficient of 0.54, compared to 0.41 for shrimp, indicating that salmon AI-tool deployment is more geographically concentrated in a smaller number of markets —particularly Norway and Chile. Shrimp deployment is less geographically concentrated than salmon’s, with company activity distributed more evenly across the countries in which they operate. Despite being a larger market by volume, AI tools are less developed and deployed in the shrimp sector than in salmon. Experts attribute this to the lower value of shrimp per tonne, and more fragmented ownership: unlike salmon, which is farmed predominantly by well-capitalized companies see also The Fish Site, 2011 https://perma.cc/DH8Z-TPJA , shrimp production spans family-run smallholdings to corporate producers see also Expert 1 ranked biomass estimation highest among in-water tools for salmon. Expert 2 reported that bioscopes—tools that simultaneously capture biomass, lice counts, wounds, respiration, and swimming behavior—are widely deployed across Chile and Norway, with biomass estimation having gained the most traction among their functions. Biomass is the primary asset farmers are selling, and knowing its current value and trajectory through a production cycle allows farmers to monitor and protect profitability. Biomass data also informs slaughter scheduling: weaker groups can be harvested first to avoid mortality that would render fish unsaleable, while healthier groups remain in the sea to continue growing, maximizing yield and revenue. Biomass is furthermore a key input to calculating feed quantities. The experts we spoke with estimate that initial stocking numbers can be off by 5–20%, which can result in significant under- or overfeeding. Given that feed is the dominant production cost see Part 1 https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-1/ , biomass estimation is particularly economically valuable. Feed optimization tools appear less embedded in salmon farming. Expert 1 ranked these last among AI product categories, attributing this to the technical complexity of modeling feeding behavior and integrating with large feeding systems. Expert 2 noted that uptake is stronger in Chile than Norway, where feed represents a larger share of margins and management culture favors automating feeding decisions. In contrast, Expert 3 reported that feed optimization tools dominate shrimp AI deployment. Traditionally, feed quantities were calculated from population estimates—you needed to know how many animals you had to determine what percentage of their total body weight to feed. AI-managed feeding has largely bypassed this problem: rather than calculating feed from population, systems now use hydroacoustic sensors 7 to detect whether animals are actively eating, and feed on that basis. The expert suggested that they would now expect further gains to come from broader adoption of existing technology, rather than improvements to current technologies. The expert noted that Population counting and estimation tools were ranked second by Expert 3 for use in shrimp. These include: In addition to context from our expert interviews, the following section draws on a Claude-assisted review of 16 academic papers published between 2023 and 2026 four primary empirical studies and 12 literature reviews , of which 13 focus primarily on aquaculture and three draw on broader agricultural or animal farming contexts. See the Appendix https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640- zdx9fecjpg53 for more information. Cost and unclear return on investment are the most consistently cited barriers to AI adoption in aquaculture across the papers that address this question directly Georgopoulos et al., 2023 https://doi.org/10.3390/su152316385 ; A further barrier is data availability, quality, and willingness to share. Our experts note that records are typically held in disparate formats, and that farmers are reluctant to share data they consider proprietary, limiting the volume of data that AI tool developers can train their models on. The literature points to two related problems: First, producing large, well-annotated datasets is costly Wu et al., 2025 https://www.mdpi.com/2071-1050/17/11/5084 , Second, proprietary data concerns can deter adoption altogether , particularly where it is unclear who controls and benefits from farm data Manning, 2024 https://doi.org/10.3390/challe15020032 ; Expert 1 noted that a related challenge is product maturity: AI companies often release products before they are sufficiently mature, requiring producers to be patient while tools develop. One such example described was an automated feeding system that turned itself on unsupervised in the early hours and dispensed 11 tons of feed into a single cage. Some companies are attempting to address contextual variability through structured on-farm calibration periods—in one case requiring three full grow-out cycles of data collection before the tool is fully operational Fletcher, 2025 https://perma.cc/H3RB-BYW2 —though the extent to which this resolves the underlying data limitations is unclear. Skill gaps are a consistently cited barrier to AI adoption, primarily reflecting low digital and AI literacy among farm operators and staff Georgopoulos et al., 2023 https://doi.org/10.3390/su152316385 ; Several governments have established dedicated aquaculture R&D and innovation funding programs —among them Norway Norwegian Ministry of Trade, Industry and Fisheries, 2023 https://perma.cc/3WY3-ZBTW , Scotland Public investment can also be particularly important where the commercial case for AI-adoption is weak. Expert 1 reported that the weaker return on investment case for lower-value species discourages both technology development by suppliers and adoption by farmers . The expert suggested that China has lagged for this reason, though government investment in university research is reported to be accelerating the development of AI tools suited to China’s specific species and production systems. In addition, effective May 2026, China’s revised Fisheries Law explicitly promotes a shift from extensive to intensive and environmentally friendly aquaculture production Godfrey, 2026 https://archive.ph/Z9cMU , Regulatory frameworks can also act as a bottleneck to AI adoption . Norwegian rules currently require manual lice counting of at least 20 fish per site weekly, and farms wishing to use automated camera systems in place of manual counts must obtain individual government dispensation to do so. The Norwegian Food Safety Authority has granted a large number of exemptions and is actively preparing formal regulatory requirements for automated methods, with Standards Norway developing minimum technical specifications Ministry of Trade, Industry and Fisheries, 2025 https://www.regjeringen.no/no/dokumenter/meld.-st.-24-20242025/id3097131/ , p. 76 . Expert 3 noted that mandatory automated lice counting is anticipated in the near term, which would accelerate uptake and eliminate the physical fish handling that manual counting requires. The United Nations’ Food and Agriculture Organization FAO projects global aquaculture production to grow over the coming decade FAO, 2024 https://openknowledge.fao.org/server/api/core/bitstreams/1273bc36-339b-43d2-8163-af4d805f2ad2/content/sofia/2024/fisheries-aquaculture-projections.html . Within this expanding industry, With time, we expect AI tools to become cheaper and easier to extend to more species. Training data requirements for feed optimization models are substantial: Expert 1 estimated that developing a reliable salmon feeding model required 14,000 cage days of data, though noted that applying an existing framework to a new species—such as sea bass or sea bream—may require only around a third of that. The reasons for this reduction were not established, though one possibility is transfer learning: a model trained on one species may encode general feeding-behavior patterns that require only limited additional data to adapt to a related species, rather than training from scratch. The literature cautions, however, that models do not always transfer readily across species or farming contexts, owing to differences in behavioral traits and environmental conditions Huang & Khabusi, 2025 https://doi.org/10.3390/pr13010073 ; This second installment of How AI is Affecting Farmed Aquatic Animals maps where the AI-enabled tools identified in Part 1 https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-1/ are being deployed in practice. We found that deployment is concentrated in Europe, Latin America, and Southeast Asia, with Norway and Scotland the largest exporters. In our dataset, Chile receives more deployments from foreign-headquartered companies than any other single country, while having no domestically-headquartered AI companies of its own. Deployment reaching lower- and middle-income producers across Latin America and Southeast Asia suggests that the welfare implications of AI in aquaculture could extend beyond the high-value, high-income operations that tend to dominate discussion of the sector. Of the species groups in our database https://docs.google.com/spreadsheets/d/1Jp-wd3rXnNpo ZheMrCKaDeNFNzsWRRb5DjtxCHCPhg/edit?gid=0 gid=0 , salmon and shrimp have the highest number of deployment instances in absolute terms, consistent with—though not independent of—their dominance at the innovation stage. Salmon AI deployment is more geographically concentrated than shrimp, particularly around Norway and Chile; shrimp AI deployment is more dispersed across Latin America and Southeast Asia. AI adoption currently remains limited: experts estimate approximately 15% of all salmon producers and <10% of shrimp producers use AI tools—a figure that rises to ~75% and ~25%, respectively, among top producers. The primary barrier to greater adoption is reported to be an unclear return on investment: tools with demonstrable returns, such as feed optimization and biomass estimation, have gained the broadest traction. In our final report, we will examine what these patterns mean for the welfare of the hundreds of billions of aquatic animals farmed each year. This report is a project of Rethink Priorities https://rethinkpriorities.org/ RP —a think-and-do tank dedicated to informing decisions made by high-impact organizations and funders across various cause areas. Sophie Williamson managed the database, did the research, conducted the expert interviews, and wrote the report. Hannah Moulange managed the project, and William McAuliffe oversaw the project. Thanks to Shane Coburn for copyediting, to Thais Jacomassi for bibliography support, and to Urszula Zarosa for publishing the report online and assisting with dissemination. A special thanks to Natasha Boyland for giving feedback on the draft. This report was produced by Rethink Priorities between January and June 2026. The project was supported by a Movement Grant from Animal Charity Evaluators ACE . The views expressed are those of the authors and do not necessarily reflect those of ACE. If you are interested in RP’s work, please visit our research database and subscribe to our newsletter. To ensure that our deployment analysis reflects AI adoption in aquaculture specifically, we classified companies by the degree to which they specialize in aquaculture. We excluded multi-sector companies from the deployment analysis, as deployment locations recorded for these companies could reflect AI use in non-aquaculture contexts rather than in aquaculture. The definitions applied are as follows: | |---| To assess where AI-aquaculture tools are being used in practice, we systematically searched for deployment evidence for each company in our database. The methods used and inclusion criteria applied are described below. | |---| Table 2 : Overview of the background and experience of the three experts interviewed for this report Expert 1 | Expert 2 | Expert 3 | | |---|---|---|---| Role type | Professor of aquaculture & CEO of an aquaculture technology consultancy | Former Chief Technology Officer at an aquaculture AI company & independent consultant | Advisor to an aquaculture technology company & co-founder of an aquaculture consultancy | Species expertise | Salmonids, shrimp, bluefin tuna | Salmonids | Shrimp | Years in aquaculture | ~30 | ~5 +~10 in AI | ~12 | Based in | Australia | Norway | Spain | We completed a timeboxed three-hour search to find and screen papers discussing barriers to AI adoption in aquaculture. We used Claude to identify relevant papers, manually checked titles for relevance, and retained 16 papers for which the full text was available. Of these, 13 focused primarily on aquaculture while three drew on wider agricultural or animal farming contexts. Of the 16 papers, 12 are literature reviews four systematic reviews using documented search protocols, and eight narrative syntheses , and four are empirical studies collecting data directly from farmers, industry professionals, or stakeholders from Greece, the Netherlands, Canada, and China . The depth of focus on adoption barriers varies considerably across the set, and studies examining adoption from the farmer or operator perspective remain sparse. We asked Claude to extract the main barrier themes from each paper, noting whether a theme was addressed and, where it was, how prominently it featured as a discouraging factor for example, a numerical ranking, a percentage share of cited barriers, or explicit characterization such as “the most commonly cited barrier” . We then compared these findings against the views expressed by our expert interviewees, and selected the themes with the strongest consensus or most substantive discussion across both sources for inclusion in this report. It is important to note that the themes highlighted here may not fully represent the barriers faced by practitioners implementing AI in aquaculture. The papers in this selection vary considerably in their framing, objectives, and methodology. Of the 16 papers, only eight treat barriers as a central analytical focus, four address them as a secondary concern, and two mention them only incidentally. The lists below are non-exhaustive. We should indicate if and how this exclusion would most distort the picture we are painting i.e., if AI products from multi-sector companies have a different pattern of deployment than their single-sector counterparts . We acknowledge potential biases from other limitations in data, but not this one. Classification of aquaculture-only, aquaculture-primary, and multi-sector companies Logging of deployment locations Limitations in what the database covers Barriers to adoption The barriers to adoption identified in the literature and by experts may not capture the full picture. One mechanism that could further suppress adoption—but which we are not aware of being documented—concerns regulatory disincentives. While AI tools could help farms monitor fish health and minimize stock losses, certain voluntary regulatory frameworks may inadvertently discourage their adoption. For example, farmers might discontinue using AI systems that provide more accurate data if that transparency leads to adverse consequences, such as: Akram, W., Muhayy Ud Din, Lyes Saad Saoud, & Irfan Hussain. 2026 . A review of generative AI in aquaculture: Applications, case studies and challenges for smart and sustainable farming. Aquacultural Engineering , 112 . https://doi.org/10.1016/j.aquaeng.2025.102637 Areeba, N., Khalid, A., Rabia, M., & Nayab, H. 2025 . AI-Driven Aquafarming: Recent Innovations and Challenges to Enhance Sustainable Seafood Practices . Progress in Aqua Farming and Marina Biology. https://academicstrive.com/PAFMB/PAFMB180046.pdf Cao, K., Ping Wang, Siyu Kong, & Chunzhen Zhang. 2026 . Driving factors of agricultural artificial intelligence adoption intention: An empirical study in Shandong province based on innovation characteristics, technology commitment, and individual heterogeneity. Frontiers . https://doi.org/10.3389/frai.2026.1630717 Chowdhury, A., Khondokar H. Kabir, Michael McQuire, & Dominique P. Bureau c. 2025 . The dynamics of digital technology adoption in rainbow trout aquaculture: Exploring multi-stakeholder perceptions in Ontario using Q methodology and the theory of planned behaviour. Aquaculture , 594 . https://doi.org/10.1016/j.aquaculture.2024.741460 D’Agaro, E. 2025 . Fish Farming 5.0: Advanced Tools for a Smart Aquaculture Management. Applied Sciences , 15 23 . https://doi.org/10.3390/app152312638 Dessler, G. 2017 . Human Resource Management . Pearson Education. FlipHTML5. https://online.fliphtml5.com/xgsob/mdft/ p=1 European Commission. 2025, April . Digital technologies and energy transition in fisheries and/or aquaculture . https://cordis.europa.eu/programme/id/HORIZON HORIZON-MISS-2025-03-OCEAN-03 Fish Farming Expert. 2026, March . £7m deal ensures future of Scotland’s aquaculture innovation centre . https://www.fishfarmingexpert.com/aquaculture-crown-estate-scotland-innovation/7m-deal-ensures-future-of-scotlands-aquaculture-innovation-centre/2093355 Fletcher, R. 2025, January . Using sonar to shed new light on the murky world of shrimp ponds. The Fish Site . https://thefishsite.com/articles/minnowtech-using-sonar-to-shed-new-light-on-the-murky-world-of-shrimp-ponds Food and Agriculture Organization of the United Nations. 2024 . Fisheries and aquaculture projections, 2022–2032 . https://openknowledge.fao.org/server/api/core/bitstreams/1273bc36-339b-43d2-8163-af4d805f2ad2/content/sofia/2024/fisheries-aquaculture-projections.html Georgopoulos, V. P., Dimitris C. Gkikas, & John A. Theodorou. 2023 . Factors Influencing the Adoption of Artificial Intelligence Technologies in Agriculture, Livestock Farming and Aquaculture: A Systematic Literature Review Using PRISMA 2020. Sustainability , 15 23 . https://doi.org/10.3390/su152316385 Gkikas, D. C., Vasileios P. Georgopoulos, & John A. Theodorou. 2024 . Exploring Aquaculture Professionals’ Perceptions of Artificial Intelligence: Quantitative Insights into Mediterranean Fish Health Management. Water , 16 24 . https://doi.org/10.3390/w16243595 Godfrey, M. 2026, March . China revises fisheries law to crack down on IUU fishing, clean up aquaculture sector. Seafood Source . https://archive.ph/Z9cMU selection-909.1-922.1 Huang, Y.-P. & Simon Peter Khabusi. 2025 . Artificial Intelligence of Things AIoT Advances in Aquaculture: A Review. Processes . https://doi.org/10.3390/pr13010073 León-Balderrama, J. I., Ysser Enrique Monsiváis Neria, & Beatriz Arellano Grajales. 2026 . Technology adoption in aquaculture: A systematic review of international empirical evidence. Iberoamerican Journal of Development Studies . https://doi.org/10.26754/ojs ried/ijds.12366 Manning, L. 2024 . Innovating in an Uncertain World: Understanding the Social, Technical and Systemic Barriers to Farmers Adopting New Technologies. Challenges , 15 2 . https://doi.org/10.3390/challe15020032 Ministry of Trade, Industry and Fisheries. 2025, April . Report to the St. 24 2024–2025 . https://www.regjeringen.no/no/dokumenter/meld.-st.-24-20242025/id3097131/ Responsible Seafood Advocate. 2025, June . Canada backs AI-enabled aquaculture technology project to grow $159B ocean economy. Global Seafood Alliance . https://www.globalseafood.org/advocate/canada-backs-ai-enabled-aquaculture-technology-project-to-grow-159b-ocean-economy/ Ryan, M. & Vincent Blok. 2026 . Artificial intelligence and animal farming: A scenario of drivers, barriers, and impacts in 2032. Agriculture and Human Values , 43 26 . https://link.springer.com/article/10.1007/s10460-025-10806-0 SankeyMATIC. n.d. . SankeyMATIC . Retrieved June 11, 2026, from https://sankeymatic.com/ The Fish Site. 2011, November . Spotlight On Seafood: Investment Trends . https://thefishsite.com/articles/spotlight-on-seafood-investment-trends The Fish Site. 2025 . Manolin unveils AI harvest forecasting and sea lice treatment tools . https://thefishsite.com/articles/manolin-unveils-ai-harvest-forecasting-and-sea-lice-treatment-tools Tina, F. W., Nasrin Afsarimanesh, Anindya Nag, & Md Eshrat E. Alahi. 2025 . Integrating AIoT Technologies in Aquaculture: A Systematic Review. Future Internet , 17 5 . https://doi.org/10.3390/fi17050199 UNCTAD. 2006 . A Case Study of the Salmon Industry in Chile . United Nations. https://unctad.org/system/files/official-document/iteiit200512 en.pdf Williamson, S., William McAuliffe, & Hannah Moulange. 2025, December . How AI is affecting farmed aquatic animals. Part 1: Innovation. Rethink Priorities . https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-1/ Wu, A.-Q., Ke-Lei Li, Zi-Yu Song, Xiuhua Lou, Pingfan Hu, Weijun Yang, & Rui-Feng Wang. 2025 . Deep Learning for Sustainable Aquaculture: Opportunities and Challenges. Sustainability , 17 11 . https://doi.org/10.3390/su17115084 The total counts company-country pairs per headquarters location, so a company with dual headquarters counts each separately: e.g. a company headquartered in both Norway and the US deploying only to Norway contributes two pairs: one for the Norway deploying to Norway, and one for the US deploying to Norway. Only 5/66 companies analyzed have dual headquarters. This is consistent with our finding at the innovation stage that these species had the most companies targeting them. However, the two findings are not independent: species with more dedicated AI innovation would be expected to show higher deployment instances simply as a result. The deployment ranking across species should not therefore be read as evidence that any species has seen disproportionately high or low uptake of the tools developed for it. For F igure 4 https://rethinkpriorities.org/research-area/how-ai-is-affecting-farmed-aquatic-animals-2/ post-10640-fig salmon operations and We found one further salmon deployment instance in the UK where Scotland was not specified. The term “top producers” was not formally defined in the interviews, with experts variously referring to “top 10 producers” and “very large players.” We interpret this as referring to the largest aquaculture companies by production volume, though the precise definition was not clarified. See footnote above. Hydroacoustic sensors commonly hydrophones, which function like underwater microphones detect the clicking sounds shrimp make when eating, and use these to determine whether the animals are actively feeding, triggering feeders to dispense or withhold feed accordingly.