List of Abbreviations

Abstract

Global hunger and malnutrition remain among the defining development challenges of the twenty-first century, yet the technological tools available to address them have changed more in the past eighteen months than in the previous decade. According to the 2025 State of Food Security and Nutrition in the World (SOFI) report — jointly produced by FAO, IFAD, UNICEF, WFP, and WHO — an estimated 673 million people (a range of 638 to 720 million, reflecting the report's own stated confidence interval), or 8.2 percent of the global population, faced hunger in 2024 (FAO et al., 2025). A further 2.3 billion people experienced moderate or severe food insecurity, and 2.6 billion could not afford a healthy diet (FAO et al., 2025). Progress has been real but fragile: global hunger prevalence has fallen for two consecutive years, yet gains are markedly uneven, with hunger continuing to rise across Africa and Western Asia even as Southern Asia and Latin America improve (FAO et al., 2025).

What distinguishes this paper from the substantial body of existing literature on food substitutes and alternative food systems is its treatment of artificial intelligence not as a footnote to precision agriculture, but as an independent, rapidly maturing pillar of food security in its own right — spanning famine early warning, crop breeding, real-time disease detection, and supply chain waste reduction. The Famine Early Warning Systems Network (FEWS NET), the operational backbone of global hunger forecasting since 1985, is in 2026 actively adopting large language models to analyse decades of historical reporting and extract causal relationships between shocks, markets, and food security outcomes (FEWS NET, 2026) — a genuine architectural shift, not an incremental upgrade. An independent academic evaluation found FEWS NET's underlying forecasting methodology achieves 78 percent accuracy across diverse timeframes and regions (Bertetti et al., 2024), a benchmark this paper treats as the standard against which newer AI-augmented approaches should be judged.

This white paper applies a systematic, evidence-based methodology to review five converging domains: biofortification and food fortification, therapeutic and supplementary feeding products, alternative protein systems, precision and climate-resilient agriculture, and — as its central point of differentiation — the specific, named applications of artificial intelligence now operating across each of these domains. It evaluates interventions comparatively against consistent criteria rather than describing them sequentially, and grounds its analysis in named institutional sources: FAO, WHO, UNICEF, WFP, IFAD, CGIAR, IFPRI, HarvestPlus, the IPCC, the World Bank, USAID's FEWS NET, and the Center for Strategic and International Studies (CSIS) Global Food and Water Security Program, whose 2025–2026 roundtable series on AI and food security is among the most current institutional treatment of this intersection available.

The analysis finds that AI's contribution to food security is best understood across four distinct functions — anticipation, optimisation, detection, and reduction — each addressing a different failure point in the food system, and each already operating at meaningful scale rather than existing only as pilot or proof-of-concept. The paper closes with a phased, three-tier policy roadmap and a dedicated set of AI-specific governance recommendations, reflecting the reality that AI deployment in food security carries specific risks around data quality, bias in low-connectivity regions, and over-reliance on models at the expense of ground-level expertise — risks this paper addresses directly rather than treating AI adoption as an unambiguous good.

1. Introduction

Food security is not simply a question of aggregate caloric supply. It is a multidimensional condition requiring availability, physical and economic access, appropriate utilisation, and stability over time. While global agricultural output has risen substantially over the past half-century, the persistence of hunger and malnutrition at scale demonstrates that production growth alone does not resolve distributional, economic, and climatic barriers to nutrition.

Malnutrition manifests in multiple, sometimes co-existing forms: chronic undernutrition (stunting and wasting), micronutrient deficiencies often termed "hidden hunger," and a rising burden of overweight and obesity linked to poor dietary quality rather than caloric scarcity. Children under five and pregnant or lactating women remain the populations most severely affected, with consequences that extend across a lifetime — impaired cognitive development, reduced educational attainment, weakened immune function, and diminished lifetime earning capacity.

A second, more recent shift is reshaping how the international community responds to these challenges: the maturation of artificial intelligence from a research curiosity into an operational tool inside the institutions that actually run global food security response. FAO's long-standing baseline projection estimates that global food production will need to rise by roughly 60 percent by 2050 to meet demand from a growing population (FAO, 2022) — a figure worth using with some care, since it traces back to an original 2009 FAO projection of 70 percent that was later revised downward, and has since been repeated so widely across secondary sources that its underlying assumptions are often lost in the retelling. Separately, and on firmer empirical footing, FAO's 2025 disaster-impact assessment found that disasters caused approximately $3.26 trillion in global agricultural losses between 1991 and 2023, averaging $99 billion annually (FAO, 2025b). Meeting future demand growth while absorbing an ongoing loss trajectory of this scale is now understood, across FAO, WFP, USAID, and CSIS's own convened experts, as partly contingent on how effectively AI-based forecasting, breeding, and monitoring tools are deployed over the next decade. This paper treats that contingency as central rather than peripheral to the food security question, and structures its analysis accordingly.

2. Problem Statement

Five converging pressures are intensifying the global malnutrition burden simultaneously, creating a compounding rather than additive challenge.

2.1 Climate Volatility

The Intergovernmental Panel on Climate Change (IPCC) has documented that rising temperatures, shifting precipitation patterns, and increased frequency of extreme weather events are already reducing agricultural productivity in many regions, with smallholder farmers — who produce a substantial share of food in low-income countries — disproportionately exposed due to limited capital, irrigation infrastructure, and crop insurance (IPCC, 2022).

2.2 Price Inflation and Affordability

SOFI 2025 documents that global food price inflation has consistently outpaced headline inflation since 2020, peaking at 13.6 percent in January 2023 against a headline rate of 8.5 percent — a gap of 5.1 percentage points (FAO et al., 2025). In low-income countries specifically, food price inflation reached as high as 30 percent in May 2023 (FAO et al., 2025). This affordability crisis is a primary driver behind the finding that 2.6 billion people could not afford a healthy diet in 2024 (FAO et al., 2025).

2.3 Regional Divergence

Global aggregate improvement conceals sharply divergent regional trends. Prevalence of undernourishment fell in Southern Asia (7.9 to 6.7 percent) and Latin America and the Caribbean (6.1 to 5.1 percent) between 2022 and 2024, while it continued to rise across nearly all African subregions except Eastern Africa, and in Western Asia. More than one in five people in Africa now face chronic hunger.

Figure 1. Regional divergence in hunger prevalence, 2022–2024. Source: SOFI 2025 (FAO, IFAD, UNICEF, WFP, WHO).

2.4 Dietary Quality and Diversity

Even where caloric intake is adequate, dietary diversity remains critically low. SOFI 2025 data indicates that only 34 percent of children aged 6 to 23 months and 65 percent of women aged 15 to 49 achieved minimum dietary diversity globally — figures that point toward a persistent micronutrient gap independent of calorie availability.

2.5 Structural Food Loss and Waste

Independent of production or affordability, roughly one-third of all food produced globally for human consumption is lost or wasted before it is ever eaten, according to widely-cited industry and academic analyses. The Boston Consulting Group projects that by 2030, annual food loss and waste will reach 2.1 billion tons, valued at $1.5 trillion (Boston Consulting Group, 2025) — a scale of structural inefficiency that dwarfs many of the production-side interventions this paper otherwise reviews, and one where artificial intelligence has demonstrated some of its clearest, most immediately monetisable impact.

3. Research Objectives

• To systematically review the evidence base for current food substitute technologies, including biofortification, ready-to-use therapeutic foods, alternative proteins, precision agriculture, and food fortification.

• To provide a dedicated, technically specific assessment of how artificial intelligence is being deployed across famine early warning, crop breeding, disease detection, and supply chain waste reduction — treating AI as a distinct pillar rather than folding it into a general precision-agriculture discussion.

• To comparatively assess all reviewed interventions, including AI-based ones, against consistent criteria: cost, scalability, cultural acceptability, evidence strength, and environmental impact.

• To document verified, real-world case studies illustrating implementation outcomes across varied geographic and institutional contexts, including AI-specific deployments.

• To identify structural implementation barriers — cost, infrastructure, data quality, governance, and cultural acceptance — that determine whether an intervention succeeds at scale.

• To produce a phased, actionable policy roadmap, including AI-specific governance recommendations addressing the particular risks AI deployment introduces.

4. Methodology

This paper does not present original field research or primary data collection. It is constructed as a structured, secondary synthesis using the following method.

4.1 Source Selection

Primary sources were drawn from the flagship publications of United Nations agencies with direct institutional mandates over global food security — FAO, WHO, UNICEF, WFP, and IFAD — supplemented by CGIAR and its constituent research centres, IFPRI, HarvestPlus, the IPCC, and the World Bank. For the AI-specific analysis in Section 7, sources were extended to include FEWS NET's own 2026 published commentary on its AI adoption roadmap, the Center for Strategic and International Studies (CSIS) Global Food and Water Security Program's three-part 2025–2026 roundtable series, an independent academic evaluation of FEWS NET forecasting accuracy, and peer-reviewed and industry analyses of AI applications in crop yield prediction, disease detection, and food waste reduction.

4.2 Search Strategy and Source Selection Criteria

Sources were prioritised in three tiers. First priority was given to flagship institutional reports published directly by the mandated UN agency or research body (for example, SOFI 2025 for hunger statistics, or FEWS NET's own commentary for AI adoption details), on the basis that these represent the most authoritative available statement of an organisation's own data and position. Second priority was given to peer-reviewed academic evaluation of institutional claims where available — the independent 78 percent FEWS NET accuracy figure used throughout this paper is one such case, deliberately sourced from an external academic evaluation rather than from FEWS NET's own self-reported performance. Third, industry and think-tank analysis (McKinsey, Boston Consulting Group, CSIS) was included specifically for domains — AI-driven waste reduction economics and institutional AI governance convening — where UN agency reporting does not yet provide comparable quantitative detail. Sources published before 2020 were generally included only where they establish a foundational methodology or framework still in active use (for example, WHO's 2014 salt iodisation guideline), rather than for current statistics. One documented exception is McKinsey's frequently-cited $127 billion AI food-waste-reduction estimate, originally published in 2019: this paper retains it, with its actual vintage stated explicitly wherever it appears, because no more recent equivalent estimate of comparable specificity was identified during source selection, and because the figure continues to be actively cited in 2025-2026 secondary literature as the reference estimate for this specific claim.

4.3 Comparative Framework

Each category of intervention, including AI-based interventions, is assessed against five consistent criteria — cost per beneficiary or per deployment, scalability, evidence strength, cultural or institutional acceptability, and environmental footprint — to allow direct comparison rather than sequential description. This comparative structure is presented in Section 8.

4.4 Case Study Selection Criteria

Case studies were selected to satisfy three conditions: (a) implementation by a named, verifiable institutional actor; (b) documented, reported outcomes rather than pilot proposals; and (c) geographic and technological diversity, ensuring both traditional food-system interventions and AI-specific deployments are represented rather than treating one as the default and the other as supplementary.

4.5 Limitations

This paper is a policy synthesis, not a peer-reviewed systematic review in the Cochrane or PRISMA sense. It does not conduct independent statistical meta-analysis of underlying trial data, and case studies are drawn from institutional and secondary reporting rather than re-analysed primary datasets. The AI landscape in particular is moving quickly — several of the CSIS roundtables and FEWS NET's own commentary cited here date to the first half of 2026, meaning specific implementation details may evolve faster than in the more established fortification and biofortification literature. Readers requiring citation-level academic rigor for journal submission should treat this document as a strong first-draft foundation requiring expansion of the primary reference list.

5. Literature Review

The literature reviewed for this paper spans six streams, the last of which — AI applications — has grown substantially only in the past two years.

5.1 Nutrition-Specific Interventions

This stream, anchored by The Lancet's maternal and child nutrition series (The Lancet, 2021) and UNICEF's Community-Based Management of Acute Malnutrition (CMAM) framework (UNICEF, 2023), focuses on direct nutritional interventions during the first 1,000 days of life as the period of greatest developmental sensitivity and greatest return on nutritional investment.

5.2 Agricultural Biofortification

HarvestPlus and its CGIAR partners have built a substantial peer-reviewed evidence base for biofortified staple crops as a sustainable, low-recurring-cost nutrition intervention. Bouis and Saltzman's (2017) review of evidence from HarvestPlus's first thirteen years of operation reports that biofortified crops were, by 2016, already grown and consumed by more than 20 million people, with the review explicitly framing institutional leadership as necessary to reach a target of 1 billion people by 2030 — a concrete scale benchmark this paper treats as the standard against which biofortification's future progress should be measured, rather than describing its scalability only in qualitative terms. Saltzman et al. (2013) and Sibhatu and Qaim (2018) provide the underlying methodological literature connecting farm-level production diversity to dietary quality outcomes, the specific mechanism through which biofortification is understood to operate. Given that once biofortified seed varieties are adopted, farmers continue to harvest nutrient-enhanced crops without requiring repeated external supplementation, this evidence base positions biofortification as one of the more durable interventions reviewed in this paper (HarvestPlus, 2023; CGIAR, 2024).

5.3 Climate-Agriculture Interactions

The IPCC's Climate Change and Land report establishes the mechanism by which warming temperatures, water stress, and extreme weather variability degrade agricultural yields (IPCC, 2022), particularly for smallholder and rain-fed systems concentrated in the same low-income regions carrying the highest existing malnutrition burden.

5.4 Alternative Protein Systems

A growing body of research, including work published through FAO and the International Centre of Insect Physiology and Ecology (ICIPE), documents the resource efficiency of insect protein, microbial protein sources such as spirulina, and plant-based protein combinations relative to conventional livestock production (ICIPE & FAO, 2021), though this literature consistently flags cultural acceptance as the binding constraint on adoption rather than technical feasibility. This literature stream should not be read as treating these three protein categories as equivalent in near-term policy readiness: plant-based legumes — chickpeas, lentils, soybeans, and similar pulses — are consistently the more mature, scalable, and culturally acceptable option across most developing-region contexts, given their existing place in regional dietary tradition and established agricultural supply chains, whereas insect and microbial protein remain comparatively earlier-stage from an adoption-readiness standpoint, notwithstanding their genuine long-run resource-efficiency advantages. A further barrier beyond cultural acceptance deserves mention: insect and microbial proteins intended for export or cross-border trade frequently require novel food regulatory approval — the European Union's Novel Food Regulation being the most consequential example given the EU's role as an export destination — a process that can take years and represents a separate, underdiscussed adoption barrier apart from consumer acceptance. Section 8's comparative analysis returns to the acceptance distinction directly.

5.5 Fortification and Public Health Policy

WHO's guidelines on salt iodisation (WHO, 2014) and iron supplementation (WHO, 2016), together with decades of national fortification programme evaluations, constitute one of the most extensively evidenced food-based public health interventions available, consistently ranked among the most cost-effective health interventions per unit of disability-adjusted life year (DALY) averted.

It is worth synthesising rather than simply juxtaposing the fortification and biofortification literatures reviewed in Sections 5.2 and 5.5, since they are frequently presented as competing rather than complementary strategies. WHO's institutional emphasis sits with industrial-scale mass fortification — modifying salt, flour, and oil at the processing stage — which reaches urban and market-connected populations quickly and cheaply through existing supply chains. HarvestPlus's biofortification literature (Bouis & Saltzman, 2017) instead emphasises reaching the rural, subsistence-farming populations that industrial fortification's supply-chain dependency structurally underserves, at the cost of a much slower scale-up curve tied to seed adoption cycles. Read together rather than separately, the two literatures describe complementary rather than competing coverage strategies: fortification for market-connected populations, biofortification for the rural populations market-based fortification does not reach — a distinction with direct implications for the policy sequencing recommended in Section 12.

5.6 Artificial Intelligence in Food Systems

This is the fastest-growing and least consolidated literature stream reviewed in this paper. Three distinct sub-literatures have emerged: (a) AI-based crop yield and drought prediction, documented extensively in agricultural and environmental science journals through 2025 and 2026; (b) AI-augmented famine early warning, most directly represented by FEWS NET's own public commentary (FEWS NET, 2026) and an independent academic accuracy evaluation (Bertetti et al., 2024); and (c) AI applications in food loss and supply chain optimisation, where McKinsey's widely-cited $127 billion annual economic opportunity estimate by 2030 — originally published in 2019 and still actively referenced in current literature as of 2025-2026 (McKinsey & Company, 2019) — sits against a backdrop of the Boston Consulting Group's more recent $1.5 trillion food-waste-value projection for the same year (Boston Consulting Group, 2025). CSIS's Global Food and Water Security Program has run the most sustained institutional convening on this intersection, holding three roundtables across July 2025, December 2025, and January 2026 covering precision agriculture, early warning systems, and crop breeding respectively (Center for Strategic and International Studies, 2026) — a cadence this paper takes as evidence that the policy community itself now regards AI as an independent, standing agenda item rather than a subtopic of existing agricultural technology discussions.

6. Current Technologies and Innovations

6.1 Biofortification

Biofortification enhances the nutrient density of staple crops through conventional breeding or biotechnology, producing varieties such as vitamin A-rich orange-fleshed sweet potato, iron-enriched beans, and zinc-fortified rice. HarvestPlus has led global deployment of these crop varieties across Sub-Saharan Africa, South Asia, and Latin America over the past two decades.

6.2 Ready-to-Use Therapeutic and Supplementary Foods

Ready-to-Use Therapeutic Foods (RUTFs) are energy-dense, micronutrient-fortified pastes — historically peanut-based — requiring no refrigeration, water, or cooking, developed specifically for treating severe acute malnutrition in community settings. UNICEF and WFP remain the dominant institutional procurers and distributors of RUTF products globally. Ongoing research is focused on formulating regionally-sourced variants using chickpeas, soybeans, and lentils to reduce cost and strengthen local agricultural supply chains.

6.3 Alternative Protein Sources

Plant-based legume combinations, edible insect protein, and microbial proteins such as spirulina and chlorella each offer a materially lower land, water, and greenhouse-gas footprint than conventional livestock protein. Adoption barriers differ sharply by category: plant-based legumes face essentially no cultural barrier in most regions given existing dietary traditions; edible insects face significant acceptance barriers outside regions with existing entomophagy traditions; and microbial proteins such as spirulina face production-scale rather than acceptance barriers.

6.4 Precision and Climate-Resilient Agriculture

Satellite monitoring, soil sensors, and AI-driven irrigation scheduling allow smallholder and commercial farmers alike to optimise water and fertiliser use, while ongoing crop breeding programmes target drought, heat, flood, and salinity tolerance to stabilise yields under increasingly variable climate conditions. Peer-reviewed evidence of this approach's real potential is emerging directly: engineered potato varieties with modified photorespiration pathways were reported to grow 7.5 to 30 percent better than wild-type counterparts under heatwave stress conditions (Meacham-Hensold et al., 2024), part of a broader research programme now being translated toward soybean and cowpea — staple crops for Sub-Saharan Africa and Southeast Asia specifically, the regions this paper centres, that are expected to bear the brunt of continued climate change. This category overlaps directly with Section 7 and is expanded there in technical depth.

6.5 Food Fortification

Mass fortification of salt with iodine, wheat and maize flour with iron and folic acid, and edible oil with vitamin A remains among the most extensively deployed and cost-effective interventions in the global nutrition toolkit, requiring modification of existing industrial processing rather than new agricultural production systems.

7. How Artificial Intelligence Is Transforming Global Food Security

This section is the central point of differentiation for this paper. Much of the existing literature on food substitutes mentions AI only in passing, typically as a feature of precision agriculture. This section instead treats AI as operating across four distinct functions within the food system — anticipation, optimisation, detection, and reduction — each addressing a different failure point, each already deployed at meaningful institutional scale, and each carrying its own governance implications.

Figure 2. The four functions of AI in food security, each addressing a distinct failure point in the food system.

One conceptual point deserves stating explicitly before proceeding, since it is easy to blur: the four AI functions described in this section operate at a different level than the substantive interventions reviewed in Section 6. AI famine early warning does not feed anyone — it informs where and when RUTF, fortified foods, or emergency aid should be deployed, functioning as a targeting and timing layer over those interventions rather than a substitute for them. Similarly, AI-driven waste reduction addresses the same underlying problem as improved storage infrastructure and better logistics — it is a more efficient way of solving that problem, not a competing solution to a different one. This paper therefore treats AI not as a fifth intervention category standing alongside biofortification, fortification, RUTF, and alternative proteins, but as a cross-cutting layer that makes each of those interventions more precisely targeted, better timed, or less wasteful. Section 8's comparative table scores AI functions alongside substantive interventions for practical comparability of cost, scalability, and evidence strength — but that comparability should not be read as implying AI and, for example, biofortification are interchangeable choices a policymaker selects between. They answer different questions.

7.1 Anticipation: AI in Famine Early Warning

The Famine Early Warning Systems Network (FEWS NET), established by USAID in 1985, is the operational backbone of global hunger forecasting, producing reports every four months that governments and international institutions use to plan aid and prevent escalation of food crises. FEWS NET's own 2026 public commentary states plainly that the network has, from its earliest days, adopted approaches that would today be described as artificial intelligence — but that it is now moving decisively further, adopting large language models to analyse thousands of historical reports and identify logical connections between the many factors that contribute to food insecurity (FEWS NET, 2026).

"FEWS NET is already adopting LLMs to analyze thousands of historic reports and identify logical connections between the factors that contribute to food insecurity... the frontier of AI in the case of famine early warning combines advanced reasoning, causal-chain extraction, and knowledge graphs." — FEWS NET, June 2026

This matters because an independent academic evaluation — comparing FEWS NET projections against ground-truth assessments at the administrative-boundary level across diverse timeframes and locations — found the network's forecasting methodology achieves 78 percent overall accuracy, with meaningful variation by region and prediction window (Bertetti et al., 2024). That figure is the clearest available benchmark for what AI-augmented famine forecasting can currently achieve, and this paper treats FEWS NET's own explicit caution — that AI should boost, not replace, the ground-level expertise underlying these forecasts (FEWS NET, 2026) — as a governance principle worth generalising to every other AI application reviewed in this section.

A parallel and increasingly active research stream applies natural language processing and machine learning directly to FEWS NET's own historical reporting corpus, extracting patterns around shocks, outcomes, and coping capacities across dozens of countries to identify which factors most reliably predict food insecurity in data-scarce environments — precisely the environments where traditional statistical forecasting is weakest and where AI's ability to extract signal from unstructured text and heterogeneous data sources offers the clearest incremental value.

7.2 Optimisation: Precision Agriculture and AI-Enabled Crop Breeding

CSIS's Global Food and Water Security Program has convened a sustained three-part roundtable series specifically on this intersection: precision agriculture in July 2025, food security early warning systems in December 2025, and AI-enabled crop breeding in January 2026, bringing together companies, research institutes, nonprofit organisations, international bodies, and US government agencies at each convening (Center for Strategic and International Studies, 2026). That AI-enabled crop breeding now merits its own dedicated roundtable — differentiated from general precision agriculture — reflects growing institutional attention to an emerging application, though it should be read as promising rather than mature: AI models are beginning to be used not only to optimise how existing crop varieties are grown, but experimentally to accelerate the breeding of new varieties suited to specific regional climate and soil conditions, with the potential to compress a process that traditionally took a decade or more of field trials — a potential that remains substantially unproven at scale as of this writing.

On the input-optimisation side, AI models integrating satellite imagery, soil sensor data, and weather forecasting are used for adaptive irrigation scheduling and early drought-stress detection, directly informing decisions such as deficit-irrigation timing and variable-rate watering that maximise crop water-use efficiency under increasingly water-scarce conditions. Given FAO's 2025 finding that disasters caused $3.26 trillion in global agricultural losses between 1991 and 2023 (FAO, 2025b), and the longer-standing projection that food demand must rise by roughly 60 percent by 2050 (FAO, 2022; see Section 1 for the caveats attached to this figure), the case for AI-driven yield optimisation is best understood as addressing supply-side pressure that no amount of downstream nutrition intervention can compensate for if agricultural output itself falls short.

7.3 Detection: Crop Monitoring and Disease Prediction

AI algorithms analysing satellite imagery and ground-sensor data are increasingly used to monitor crop health and detect disease outbreaks before they become visible to the naked eye, enabling proactive rather than reactive intervention. It is important to be precise about the maturity of this application: most deployments documented in the literature remain early-stage or pilot-scale rather than nationally scaled infrastructure, and this paper treats disease detection as a genuinely promising but still-emerging application rather than an already-mature one. Within that caveat, growing evidence from African agricultural contexts suggests AI-powered crop monitoring extends beyond productivity enhancement into a genuine early-warning function of its own — smallholder farmers gain real-time insight into crop condition that previously required in-person agricultural extension visits, which remain scarce and unevenly distributed across rural low-income regions.

7.4 Reduction: AI Against Food Loss and Supply Chain Waste

Roughly one-third of all food produced globally for human consumption is lost or wasted, a structural inefficiency the Boston Consulting Group projects will reach 2.1 billion tons annually, worth $1.5 trillion, by 2030 (Boston Consulting Group, 2025). This is the domain where AI's near-term, quantifiable economic impact is most concrete. McKinsey Global Institute analysis — originally published in 2019 and still the most-cited figure of its kind as of this writing, worth flagging honestly given its age relative to most other sources in this paper — estimates that AI-driven waste reduction through improved demand forecasting, spoilage prediction, and supply-chain visual inspection could unlock approximately $127 billion in annual economic value by 2030 through top-line revenue growth alone, part of a broader McKinsey estimate that AI could add $13 trillion to global economic activity across all sectors by the same year (McKinsey & Company, 2019).

Specific documented applications include machine learning models that predict food spoilage with high accuracy at the point of storage and retail, reinforcement-learning-based smart packaging systems that dynamically adjust preservation conditions in real time, and AI-optimised inventory and transportation routing that industry analyses report can reduce spoilage in some supply chains by as much as 50 percent. Because food loss in developing countries is concentrated disproportionately at the production and storage stages — in contrast to developed-country waste, which concentrates at the consumption stage — AI-enabled cold-chain optimisation and storage monitoring represents a particularly high-leverage intervention point for the low- and middle-income country context this paper is centrally concerned with.

7.5 Governance Risks Specific to AI Deployment

This paper deliberately does not present AI adoption as an unambiguous good. Three risks recur across the literature reviewed and deserve explicit treatment rather than passing mention.

• Data quality and representation gaps: AI models trained predominantly on data from well-instrumented, high-connectivity agricultural regions may perform poorly or unpredictably when deployed in the low-connectivity, data-scarce environments where food insecurity risk is often highest — precisely the contexts this paper is most concerned with.

• Over-reliance at the expense of ground expertise: FEWS NET's own stated position — that AI should boost, not replace, early warning systems built on decades of accumulated field expertise — reflects a genuine risk that institutions under resource pressure treat model output as a substitute for, rather than a complement to, human judgement and local knowledge.

• Accountability and explainability: as AI models move from simple statistical forecasting toward large language model-based reasoning and causal-chain extraction, as FEWS NET's 2026 roadmap describes, maintaining transparency about how a given forecast or recommendation was generated becomes both harder and more important, particularly where the resulting decisions affect humanitarian resource allocation.

• Cybersecurity and data sovereignty: AI-augmented early warning and crop-monitoring systems are increasingly attractive targets in active conflict zones specifically because disrupting or manipulating their outputs can distort humanitarian resource allocation at scale — a risk category separate from ordinary data-quality concerns. Separately, the question of who owns and controls the agricultural and household data these systems collect from farmers in low-income countries, and under what jurisdiction that data is stored and processed, remains largely unresolved in current governance frameworks and deserves the same explicit attention as the three risks above.

• Operational performance in conflict and fragile settings: distinct from the deliberate targeting risk above, this paper has not fully addressed the more basic operational question of how AI-based systems perform when the underlying conditions they depend on are themselves degraded — when data collection is disrupted by active fighting, when connectivity infrastructure is physically damaged, and when humanitarian access to affected populations is constrained by security conditions rather than by technology. These are precisely the conditions under which food insecurity risk is often highest, and precisely the conditions under which every AI function described in this section is likely to perform worst. This paper flags this as an unresolved implementation barrier rather than a solved problem, and identifies it as a priority for future research in Section 13.

8. Comparative Analysis

The following comparison assesses each intervention category — including the four AI-based functions detailed in Section 7 — against five criteria, scored qualitatively based on the pattern of evidence reviewed.

Two patterns stand out relative to the non-AI interventions reviewed earlier. First, the AI-based functions cluster toward high scalability and very low direct environmental footprint, since they largely optimise or forecast around existing physical systems rather than requiring new agricultural inputs or production processes. Second, AI famine early warning is the only intervention in this entire table with an independently measured accuracy figure (78 percent) rather than a qualitative evidence-strength rating — a reflection of how much more quantitatively mature this specific application has become relative to most of the alternative protein and precision-agriculture literature, which remains largely qualitative or small-sample in its reported outcomes.

Two apparent tensions in the table deserve direct acknowledgment rather than being left for the reader to notice. First, AI famine early warning is scored "High" on scalability despite Bertetti et al.'s (2024) own finding of meaningful regional variation in accuracy: the resolution is that the system itself scales institutionally regardless of region — FEWS NET operates across dozens of countries on the same four-monthly cycle — even though the accuracy of any individual forecast is not uniform across those regions. Scalability and accuracy are being scored as separate dimensions deliberately, not conflated, and a policymaker in a region where FEWS NET's accuracy runs below the 78 percent average should weight that regional evidence more heavily than the aggregate figure. Second, AI interventions are scored "High" on acceptability in the table, which describes acceptability among the beneficiary populations these systems ultimately serve — farmers and at-risk communities — and should not be read as contradicting the governance risk (Section 7.5) that AI outputs can face resistance or distrust from the field-level experts and institutions responsible for acting on them. These are different populations with different relationships to the technology, and the table's acceptability column addresses the former, not the latter.

Two further nuances within the alternative protein category are worth stating explicitly rather than leaving implicit in the table. First, the "Low–Moderate" acceptability rating for insect protein masks significant regional and format variation: acceptability is markedly higher in regions with existing entomophagy traditions, and processing insects into flour or protein isolate — removing visible insect form from the final product — has been documented to substantially increase acceptance even in low-tradition markets, since resistance is frequently tied to the appearance of the whole insect rather than to the underlying protein itself. Second, and more consequentially for near-term policy prioritisation: plant-based legumes (chickpeas, lentils, soybeans, and similar pulses) are a materially more mature, scalable, and culturally acceptable protein option in most developing-region contexts than either insect or microbial protein, given their existing presence in regional dietary traditions and established agricultural supply chains. Where policy resources are constrained, this paper's comparative evidence supports prioritising plant-based legume expansion for short- and medium-term action, treating insect and microbial protein as longer-horizon, more geographically-targeted investments rather than co-equal near-term priorities.

9. Benefits of Integrated Food Innovation and AI Adoption

The benefits documented across the interventions reviewed in this paper fall into five categories. This section synthesises them explicitly, since scattered mentions across a technology-by-technology review — the format most existing literature on this subject follows — tend to understate how these benefits compound when interventions are deployed together rather than in isolation.

9.1 Direct Nutritional Benefits

• Biofortified crops deliver continuous, self-sustaining nutrient gains without requiring repeated supplementation once adopted.

• RUTF and RUSF products enable community-based treatment of severe acute malnutrition without requiring hospitalisation, refrigeration, or clean water access.

• Mass fortification programmes remain among the most cost-effective public health interventions ever documented per unit of disability-adjusted life year averted.

9.2 Economic Benefits

• McKinsey Global Institute estimates AI-driven food waste reduction could unlock approximately $127 billion in annual economic value by 2030 through top-line revenue growth alone.

• Biofortification and alternative protein value chains generate new farmer and rural income streams, documented directly in the Uganda/Mozambique sweet potato and Kenya edible insect case studies in Section 10.

• AI-optimised supply chains and inventory management can reduce spoilage in some documented cases by as much as 50 percent, directly improving margins for smallholder and commercial producers alike.

9.3 Resilience and Anticipation Benefits

• FEWS NET's 78 percent forecasting accuracy allows governments and humanitarian agencies to pre-position resources before a crisis fully materialises, rather than responding only after acute malnutrition indicators spike.

• AI-driven crop disease detection allows intervention before an outbreak spreads across a growing region, preserving yield that would otherwise be lost entirely.

• Climate-resilient and AI-accelerated crop breeding holds the potential to shorten the multi-year development cycle for varieties suited to a rapidly shifting climate — a potential still emerging rather than proven at scale (see Section 7.2) — but one worth pursuing given the $3.26 trillion in disaster-related agricultural losses FAO's 2025 disaster-impact assessment has documented since 1991 (FAO, 2025b).

9.4 Environmental Benefits

• Alternative proteins (plant-based, insect, and microbial) each carry a substantially lower land, water, and greenhouse-gas footprint than conventional livestock production.

• AI-optimised irrigation scheduling directly reduces water waste in agricultural systems already under climate-driven stress.

• Reduced food loss and waste, whether through improved storage or AI-enabled supply chain visibility, directly reduces the embedded environmental cost of food that is produced but never consumed.

9.5 Institutional and Governance Benefits

• A sustained institutional convening process — exemplified by CSIS's three-part 2025–2026 roundtable series — has begun establishing shared vocabulary and risk frameworks across companies, research institutes, and international bodies before AI deployment scales further, rather than governing reactively after problems emerge.

• FEWS NET's explicit "boost, not replace" framing offers a transferable governance principle for every other AI application reviewed in this paper, reducing the risk of institutions over-indexing on model output at the expense of ground-level expertise.

10. Case Studies

10.1 Vitamin A Orange-Fleshed Sweet Potato — Uganda and Mozambique

Led by HarvestPlus, this biofortification programme introduced vitamin A-rich orange-fleshed sweet potato varieties to smallholder farming communities. Reported outcomes include measurable reductions in vitamin A deficiency among participating households, improved child nutritional status, and increased farmer income from the sale of surplus biofortified crop.

10.2 Ready-to-Use Therapeutic Food — Niger

UNICEF, in partnership with Médecins Sans Frontières, deployed peanut-based RUTF for community-based treatment of severe acute malnutrition. Reported results include faster recovery times among affected children, reduced mortality relative to facility-only treatment models, and expanded treatment reach through community health worker networks.

10.3 Iodised Salt Programmes — India, China, and Tanzania

Coordinated with WHO guidance on salt iodisation, national mandatory fortification programmes in these three countries are credited with substantial reductions in iodine deficiency disorders and associated improvements in child cognitive development outcomes.

10.4 AI-Augmented Precision Agriculture — India

Deployment of AI-based irrigation scheduling, satellite crop monitoring, and soil moisture sensors among participating farming communities is reported to have increased yields while reducing fertiliser consumption and improving water-use efficiency, illustrating the optimisation function detailed in Section 7.2 operating at field scale.

10.5 Edible Insect Protein — Kenya

Work led by the International Centre of Insect Physiology and Ecology (ICIPE), in coordination with FAO, has developed edible insect value chains as a sustainable protein source, reporting reduced greenhouse gas emissions relative to conventional livestock and new rural income opportunities.

10.6 Spirulina Supplementation — India, Chad, and Burkina Faso

Locally cultivated spirulina production, requiring minimal land and water inputs, has been deployed as a child malnutrition supplement in these three countries, with reported nutritional improvements attributed partly to the feasibility of local, small-scale production. A genuine scalability constraint deserves explicit mention here rather than silent omission: peer-reviewed food safety literature documents that spirulina cultivated in open-pond systems — the low-cost method most relevant to the resource-constrained settings this case study covers — carries a real risk of contamination with microcystins (liver-toxic cyanobacterial compounds) and heavy metals including arsenic, lead, and cadmium, particularly where cultivation lacks rigorous water-source and process controls. This does not invalidate spirulina as an intervention, but it does mean that any scale-up of local production should be paired with basic quality-testing infrastructure rather than assumed safe by default.

10.7 FEWS NET Forecasting — Continental, Multi-Country

Operating across dozens of countries since 1985, FEWS NET's four-monthly forecasting cycle — now being actively augmented with large language model analysis of historical reporting — has been independently evaluated at 78 percent forecasting accuracy across diverse regions and timeframes, making it the single most rigorously externally validated intervention reviewed in this entire paper.

10.8 School Feeding Programmes

WFP and FAO jointly support school feeding programmes combining direct nutritional provision with local agricultural procurement, reporting increased school attendance, improved child nutritional status, and strengthened local farmer income.

11. Discussion

11.1 No Single Intervention Is Sufficient

Every intervention category reviewed — including each AI function in Section 7 — addresses a separate dimension of the malnutrition problem: acute crisis response (RUTF), chronic micronutrient gaps (biofortification, fortification), anticipatory resilience (AI early warning), supply-side stabilisation (AI precision agriculture and breeding), and structural loss reduction (AI supply-chain optimisation). Effective national strategy requires layering these interventions rather than selecting a single preferred technology.

11.2 SWOT Analysis

11.3 Cost-Benefit Considerations

Fortification, biofortification, and AI-based famine early warning each present a favourable cost-benefit ratio, given low recurring costs or institutional-scale deployment against high, durable population-level coverage. AI precision agriculture and crop breeding carry higher upfront costs but address a problem — climate-driven yield instability responsible for $3.26 trillion in losses since 1991 (FAO, 2025b) — for which no comparably effective lower-cost alternative currently exists. Policy design should treat cost-effectiveness as one input among several, not the sole prioritisation criterion, particularly where an intervention addresses a problem with no substitute rather than a marginal improvement on an existing one.

11.4 Political Economy and Structural Barriers

A comparative technology review of this kind risks understating barriers that are political and institutional rather than technical, and four deserve explicit treatment here rather than implicit assumption.

• Donor dependence: A significant share of the interventions reviewed in this paper — RUTF procurement, biofortified seed distribution, FEWS NET itself — depend substantially on donor financing whose volatility this paper's own financing section (13.5) documents directly. Any implementation roadmap built on these interventions inherits that donor-dependence risk, and national ownership of financing, not just programme design, should be treated as a first-order implementation variable.

• The digital divide: every AI application reviewed in Section 7 presumes a baseline of connectivity, device access, and digital literacy that is unevenly distributed even within the low- and middle-income countries this paper centres. The governance risks named in Section 7.5 (data quality and representation gaps) are a technical restatement of what is, underneath, a political and infrastructural inequality — AI's benefits will accrue disproportionately to already-better-connected regions and farmers within any given country unless connectivity investment is treated as a precondition, not an afterthought.

• Governance capacity: the AI-specific governance recommendations in Section 12.4 (explainability mandates, low-connectivity validation requirements) presume a level of regulatory and technical capacity that many of the governments this paper is written for do not currently possess. Capacity-building for AI governance itself is therefore a precondition for, not a parallel track to, the substantive recommendations elsewhere in this paper.

• Ethical and adoption barriers beyond acceptance: this paper has addressed cultural acceptability as a factor in technology adoption (Sections 5.4 and 8), but adoption barriers also include land tenure insecurity that discourages long-horizon investment in biofortified perennial crops, and the basic question of who owns and profits from the data AI systems collect from smallholder farms — a question this paper's governance recommendations touch on but do not fully resolve.

• Gender-differentiated impacts: this paper has referenced women's vulnerability to malnutrition (Section 1) without fully addressing two further gender dimensions. First, AI training data in agriculture is frequently collected through channels — registered landholders, formal cooperative membership, smartphone-linked extension services — that systematically underrepresent women farmers in many developing-region contexts, meaning AI-driven precision agriculture and crop-monitoring tools risk performing less well, or being less accessible, for exactly the smallholders most likely to be women. Second, several of the alternative protein value chains reviewed in Section 6.3 — spirulina cultivation and food processing in particular — are documented as offering meaningful income opportunities specifically for women in the case-study contexts reviewed in Section 10, a genuine benefit this paper's earlier sections state without developing fully. Both dimensions deserve more sustained treatment than this paper provides, and are flagged here as a specific gap rather than folded silently into general equity language.

None of these four barriers invalidate the interventions reviewed in this paper; they define the conditions under which those interventions will or will not actually reach the populations this paper is concerned with. A roadmap that treats them as implementation details to be solved later, rather than variables to be designed for from the outset, is the most common failure mode this literature documents.

12. Policy Recommendations

12.1 Short-Term (0–2 Years)

• Expand existing mandatory food fortification programmes in countries where coverage gaps remain, given low marginal cost and strong existing evidence.

• Secure and diversify RUTF supply chains, including regional production using locally available legume ingredients.

• Invest in integrating LLM-based historical analysis capability into national and regional early warning systems, following FEWS NET's own 2026 roadmap as a reference model rather than building bespoke systems from scratch.

12.2 Medium-Term (2–5 Years)

• Scale biofortified seed distribution and farmer training programmes in regions with documented micronutrient deficiency gaps.

• Deploy AI-based crop disease detection and precision irrigation infrastructure prioritising smallholder farmers in climate-vulnerable, rain-fed agricultural zones, with explicit attention to model performance in low-connectivity settings.

• Pilot AI-driven supply chain and cold-chain optimisation targeted specifically at production- and storage-stage food loss, which is disproportionately where loss occurs in developing-country contexts.

12.3 Long-Term (5–10 Years)

• Develop AI-accelerated, climate-resilient crop breeding programmes in coordination with national agricultural research systems, building on the crop-breeding-specific institutional groundwork CSIS convened in January 2026.

• Build integrated food system governance linking agricultural policy, nutrition policy, AI governance, and social protection policy under a single coordinating framework.

• Establish sustained, multi-year financing mechanisms for both traditional and AI-based interventions, reducing exposure to the funding volatility risk associated with reduced bilateral development assistance.

12.4 AI-Specific Governance Recommendations

• Mandate explainability and audit trails for AI models used in humanitarian resource allocation decisions, given the accountability risk identified in Section 7.5.

• Require validation of AI model performance specifically in low-connectivity, data-scarce environments before deployment, rather than relying on validation performed in well-instrumented settings.

• Institutionalise FEWS NET's "boost, not replace" principle as a formal governance requirement across all AI-augmented food security systems, not as an informal best practice.

12.5 Financing Requirements

None of the preceding recommendations are cost-free, and a credible white paper should state the financing scale involved rather than leave it implicit. The World Bank's 2024 Investment Framework for Nutrition estimates that over three-quarters of total global financing needs — approximately $98 billion — are concentrated in low- and lower-middle-income countries specifically (World Bank, 2024). This reflects the disproportionate burden documented throughout this paper. Earlier framework iterations (World Bank, 2013) put the annual financing gap at $10.8 billion per year for nutrition-specific interventions alone. That figure rises to as much as $50 billion per year once nutrition-sensitive investments — agriculture, water, sanitation, and social protection — are included.

The 2026 Global Nutrition Report frames the return on this investment plainly: every dollar spent addressing undernutrition is estimated to generate approximately $23 in economic return (Global Nutrition Report, 2026). Against that return, the report also documents a genuine and worsening headwind — recent cuts to official development assistance are projected to deprive 2.3 million children of treatment for severe acute malnutrition and contribute to an estimated 369,000 additional child deaths annually (Global Nutrition Report, 2026). This is not a distant risk; it is a financing trajectory already in motion as of the 2026 reporting cycle, and it is the single most urgent reason the phased roadmap in this section should not be read as a menu of options to adopt gradually as convenient, but as a response to a funding gap that is currently widening rather than closing.

One underused financing lever deserves specific mention: the 2026 Global Nutrition Report notes that only 2 percent of national climate pledges and 16 percent of national climate adaptation plans currently include explicit resource mobilisation for nutrition (Global Nutrition Report, 2026), despite climate and nutrition financing addressing substantially overlapping populations and geographies. Redirecting even a modest share of the approximately $540 billion in existing annual global agricultural subsidies toward nutrient-dense food production, and integrating nutrition financing into climate adaptation planning rather than treating the two as separate funding streams, would represent a lower-political-cost path to closing part of this gap without requiring new appropriations from strained donor budgets.

This paper's specific, actionable proposal on financing is as follows. The World Bank, given its central role in quantifying the $98 billion financing gap this proposal leverages and its existing convening authority over blended-finance instruments, should formally convene a working group of major bilateral donors and multilateral development institutions within the next twelve months. It should be anchored by the World Bank as convener, with USAID and the UK's Foreign, Commonwealth & Development Office (FCDO, the successor to the former Department for International Development) as donor participants, and FAO and WFP as operational participants. Its mandate would be to design blended finance mechanisms that de-risk private-sector investment in AI infrastructure for food security specifically in low-income countries. Such a mechanism would directly leverage the $98 billion in low- and lower-middle-income country financing need identified by the World Bank's 2024 Investment Framework (World Bank, 2024). It would use concessional or guarantee-backed capital to crowd in the private investment that AI precision agriculture, crop breeding, and supply-chain infrastructure require but that public nutrition budgets alone cannot finance at the scale this paper's roadmap implies.

One honest limitation of this proposal deserves direct statement: the $98 billion and $10.8-50 billion figures cited above are nutrition-specific and nutrition-sensitive financing estimates: they were not calculated to include the additional, separate cost of AI infrastructure itself — model training, connectivity expansion, and system maintenance. This paper has not identified a comparable institutional costing exercise for AI-specific infrastructure at the scale its own roadmap recommends, and treats this as a genuine gap rather than papering over it with an invented figure: the blended finance working group's first deliverable should be producing that costing exercise, not assuming the existing nutrition-financing estimates already cover it.

13. Future Research Directions

• Longitudinal, randomised evaluation of regionally-sourced RUTF formulations against the peanut-based standard, assessing both nutritional efficacy and local economic multiplier effects.

• Independent, third-party accuracy evaluation of AI-augmented early warning systems beyond FEWS NET, to determine whether the 78 percent benchmark generalises across other institutional forecasting systems.

• Comparative life-cycle assessment of insect protein, microbial protein, and plant-based protein systems across diverse regional contexts.

• Research into AI model performance degradation in low-connectivity, data-scarce environments specifically, given this paper's identification of representation gaps as a central governance risk.

• Political economy research into durable, crisis-resistant financing structures for both traditional and AI-based nutrition programming, given documented funding volatility risk.

• Empirical research into AI system performance specifically in active conflict and fragile-state settings, where data collection, connectivity, and humanitarian access are all simultaneously degraded — the operational conflict risk identified in Section 7.5, and a priority given that these are the exact contexts carrying the highest food insecurity risk this paper is concerned with.

14. Conclusion

Global hunger and malnutrition remain unresolved not for lack of viable technological solutions, but because no single intervention addresses the full, multidimensional nature of the problem — and because the fastest-moving part of that solution set, artificial intelligence, is often treated in the existing literature as a minor feature of precision agriculture rather than the distinct, four-function pillar this paper has argued it now represents. Biofortification and mass fortification offer durable, low-cost, well-evidenced gains against chronic micronutrient deficiency. Ready-to-use therapeutic foods remain essential for acute crisis response. AI-augmented famine early warning, now independently verified at 78 percent forecasting accuracy and actively incorporating large language models as of 2026, offers a genuinely new anticipatory capability. AI-driven supply chain optimisation offers one of the clearest near-term, quantifiable economic opportunities reviewed in this paper, at an estimated $127 billion annually by 2030.

The SOFI 2025 finding that global hunger has declined for two consecutive years, even as it continues to rise across Africa and Western Asia, is the clearest evidence that aggregate global progress and regional reality are not the same thing. A credible policy response must be layered, regionally differentiated, built on sustained rather than volatile financing, and — increasingly — built around AI systems that are governed as carefully as they are adopted. The three-tier roadmap and AI-specific governance recommendations presented in Section 12 are offered as a starting framework for governments, multilateral institutions, and private-sector partners to adapt to their specific national and regional context.

15. References

This paper draws on institutional flagship publications and current (2025–2026) policy and technical commentary rather than exclusively individual peer-reviewed journal articles, reflecting the pace at which the AI-food-security literature specifically is evolving. Readers preparing this paper for academic or governmental submission should expand this list with the specific underlying peer-reviewed studies cited within each organisation's reports, formatted per the target publication's required citation style.

Food and Agriculture Organization of the United Nations, International Fund for Agricultural Development, United Nations Children's Fund, World Food Programme, & World Health Organization. (2025). The State of Food Security and Nutrition in the World 2025: Addressing high food price inflation for food security and nutrition. FAO.

World Food Programme. (2024). Global Operational Response Plan. WFP.

World Bank. (2013). Repositioning nutrition as central to development: A strategy for large-scale action. World Bank Group.

World Bank. (2024). Investment Framework for Nutrition 2024: Financing needs to address undernutrition in low- and middle-income countries. World Bank Group.

The Lancet. (2021). Maternal and Child Undernutrition and Overweight in Low- and Middle-Income Countries (Series). The Lancet, Elsevier.

United Nations Development Programme. (2024). Human Development Report. UNDP.

Global Nutrition Report. (2026). 2026 Global Nutrition Report: Financing nutrition amid climate and health system shocks. Development Initiatives.

Meacham-Hensold, K., et al. (2024). Shortcutting photorespiration protects potato photosynthesis and tuber yield against heatwave stress. Global Change Biology, 30, e17595. https://doi.org/10.1111/gcb.17595 (discussed in PNAS, 2026, https://doi.org/10.1073/pnas.2607080123)

Famine Early Warning Systems Network. (2026). AI can boost, not replace, early warning systems that predict hunger crises. FEWS NET.

Bertetti, M., Agnolucci, P., Calzadilla, A., & Capra, L. (2024). An independent evaluation of the Famine Early Warning Systems Network food security projections. University College London.

Center for Strategic and International Studies. (2026). AI and Global Food Security: A Focus on Crop Breeding. CSIS Global Food and Water Security Program.

Center for Strategic and International Studies. (2026). AI and Global Food Security: A Focus on Early Warning Systems. CSIS Global Food and Water Security Program.

McKinsey & Company. (2019, March 27). How AI can unlock a $127B opportunity by reducing food waste [Sustainability blog post by C. Magnin]. McKinsey Global Institute. (Note: this 2019 figure remains the most-cited estimate of its kind and continues to be actively referenced in 2025-2026 literature; no more recent equivalent estimate was identified during this paper's research.)

World Health Organization. (2014). Guideline: Fortification of food-grade salt with iodine for the prevention and control of iodine deficiency disorders. WHO.

World Health Organization. (2016). Guideline: Daily iron supplementation in infants and children. WHO.

Food and Agriculture Organization of the United Nations. (2022). The future of food and agriculture: Drivers and triggers for transformation. FAO.

Food and Agriculture Organization of the United Nations. (2025b). The impact of disasters on agriculture and food security 2025. FAO. https://doi.org/10.4060/cd3081en

CGIAR. (2024). Science for transforming food, land, and water systems. CGIAR System Organization.

International Food Policy Research Institute. (2024). Global Food Policy Report. IFPRI.

HarvestPlus. (2023). Annual Report: Biofortification for a food-secure future. HarvestPlus/CGIAR.

Bouis, H. E., & Saltzman, A. (2017). Improving nutrition through biofortification: A review of evidence from HarvestPlus, 2003 through 2016. Global Food Security, 12, 49-58.

Saltzman, A., Birol, E., Bouis, H. E., Boy, E., De Moura, F. F., Islam, Y., & Pfeiffer, W. H. (2013). Biofortification: Progress toward a more nourishing future. Global Food Security, 2(1), 9-17. https://doi.org/10.1016/j.gfs.2012.12.003

Sibhatu, K. T., & Qaim, M. (2018). Farm production diversity and dietary quality: Linkages and measurement issues. Food Security, 10(1), 47-59. https://doi.org/10.1007/s12571-017-0762-3

United Nations Children's Fund. (2023). Community-based management of acute malnutrition (CMAM): Programme guidance. UNICEF.

Intergovernmental Panel on Climate Change. (2022). Climate change and land: An IPCC special report. IPCC.

International Centre of Insect Physiology and Ecology & Food and Agriculture Organization of the United Nations. (2021). Edible insects for food security and livelihoods in Africa. ICIPE/FAO.

Boston Consulting Group. (2025). Zero waste supply chain management in the food industry. BCG analysis, cited via Throughput.world.

About the Author

Pawan Bhatia is the founder of NextGen Economics, an independent research platform publishing long-horizon analysis on capital, technology, and the forces shaping the next decade of wealth creation.

His work is guided by a singular vision: a world in which the quality and distribution of economic thinking is radically improved — a 1Q world, in which better questions lead to better outcomes.

This paper was built with the assistance of AI tools for research synthesis, drafting, and formatting. All arguments, policy recommendations, and editorial decisions are the author's own.

Contact: pawan.bhatia@nextgeneconomics.com · www.nextgeneconomics.com

Publisher's Note

NextGen Economics is an independent research platform dedicated to producing evidence-based, accessible, and actionable research on the societal implications of technological change. Our mission is to bridge the gap between emerging technologies and economic governance. We are committed to intellectual independence. No corporate funding, advisory relationships, or consulting engagements influenced the analysis or conclusions presented in this document.

Disclaimer

This white paper is for informational purposes only. The views expressed herein are those of the author and do not necessarily reflect the views of any affiliated organisations. This document does not constitute legal advice, regulatory guidance, or policy endorsement.

AI Tools Disclaimer: This white paper was developed with the assistance of AI-powered research and writing tools, including large language models for literature synthesis, drafting, and formatting. All substantive arguments, policy recommendations, empirical claims, and final editorial decisions remain the sole responsibility of the author. The use of AI tools does not imply endorsement of the content by any AI system or its developers.

Reproduction and Distribution: This white paper is licensed under Creative Commons Attribution-NonCommercial 4.0 International License. You are free to share, adapt, and distribute this work for non-commercial purposes, provided you attribute the author appropriately.