Descubre Pandipedia

AI Adoption in Manufacturing and Services in Emerging Markets

Bottom line: The available evidence points to a generally faster, more digitally concentrated adoption pattern in services than in manufacturing, but it does not provide a reliable, economy-wide quantitative comparison for emerging and developing countries. Services therefore face more immediate exposure in some occupations, while manufacturing policy must focus more on foundational digital capability, infrastructure, data quality, and small-firm readiness. These conclusions should be treated as qualified because much of the comparative evidence comes from OECD economies or advanced-economy firms.^[1][2][3]

For policymakers, the practical distinction is not simply “services versus manufacturing.” Adoption and workforce effects vary by task, skill level, firm size, and digital readiness. ICT, finance, and professional services are unlike retail or food services, just as high-skill electronics manufacturing is unlike less digitally connected production.^[4]

1. Adoption speed and sectoral exposure

In the economies covered by the research, services generally adopt AI earlier than manufacturing. ICT, information and communication, finance, and professional, scientific, and technical services are among the leading adopters. OECD evidence reports adoption rates of 44% in information and communication services and 26% in professional, scientific, and technical services in 2024, while manufacturing trailed the leading service sectors.^[5][6]

The available trend evidence also indicates faster adoption growth in ICT and professional services between 2021 and 2024. Manufacturing adoption increased substantially, by roughly 70% in the cited OECD comparison, but from a lower base and with a smaller percentage-point increase than the leading service sectors.^[7][8] This supports a services-first pattern, not a universal rule for every emerging market.

2. Capital, infrastructure, and data barriers

The research does not establish that manufacturing or services systematically faces greater financing barriers across emerging markets. It does show that small firms generally have less investment capacity and that financial constraints can delay long-term AI and digital-transformation investment, without proving a sector-specific lending gap.^[19][20][21]

The nature of the investment need differs. For services, the evidence more clearly identifies shortages of computing capacity, data centres, connectivity, reliable power, and cooling capacity as constraints on AI workloads. Financing and regulatory reform must make these infrastructure projects bankable.^[22][23] For manufacturing, the strongest evidence concerns smaller firms’ weak digital readiness, poor-quality data, difficulty finding suitable vendors, and difficulty adapting business models.^[24][25]

This means that “capital barriers” should be interpreted broadly. Services may require investment in shared digital infrastructure and data capacity, while manufacturing firms may need finance and technical assistance for basic digitalisation, usable data, and complementary capabilities. The sources do not support a claim that one sector is universally more capital-intensive or financially constrained than the other.^[26][27]

3. Workforce effects

Services face more immediate and uneven workforce exposure because some activities are routine, language-based, or administrative. The IZA evidence specifically identifies the Philippines’ business-process-outsourcing sector as exposed to chatbot substitution, despite its relatively small employment share and substantial economic contribution.^[28] At the same time, high-skill services such as medicine may benefit from AI-enabled worker augmentation rather than straightforward replacement.^[29]

For manufacturing, the sources do not establish a distinct manufacturing-specific employment effect in developing economies. They do indicate that less digitally connected employment is likely to face slower initial generative-AI disruption because fewer tasks are currently exposed. Evidence from manufacturing firms outside the developing-economy context also cautions against equating AI adoption with immediate factory-floor job loss: reported use was concentrated mainly in functions such as sales and marketing rather than directly in production.^[30][31]

The appropriate workforce concern is therefore different by sector. In exposed service industries, the risk is concentrated displacement or rapid changes in work practices. In manufacturing and other lower-exposure activities, the nearer-term challenge may be unequal access to productivity-enhancing technology, alongside the need to build workers’ digital and technical capabilities. These are evidence-based interpretations, not precise forecasts of employment change for all emerging economies.^[32][33]

4. Distinct policy priorities

A common national AI strategy is unlikely to be sufficient. The sectoral evidence supports differentiated measures that address both adoption capacity and worker transition.

Conclusion and evidence boundary

The best-supported comparison is qualified: services generally adopt AI earlier and expose more digitally intensive, language-based, and administrative work, while manufacturing adoption is slower overall and more uneven, with especially important capability gaps among small firms. Capital constraints affect both sectors, but the evidence points to different bottlenecks: computing, connectivity, data-centre capacity, energy, and data governance are more clearly documented for services, while manufacturing evidence centres on digital readiness, data quality, vendors, and business-model adaptation.^{[43][44][45][46]}

The evidence does not justify a precise emerging-market estimate of the manufacturing-services adoption gap, nor a universal claim that AI will reduce manufacturing employment or replace service workers. Policy should therefore combine sector-specific support with better country-level measurement, especially of small firms, informal work, task exposure, and worker transitions.

QKD vs. Post-Quantum Cryptography: Which Should You Deploy?

For most organizations, deploy software-based post-quantum cryptography (PQC) as the default migration path. Reserve hardware-based quantum key distribution (QKD) for narrowly defined, high-assurance links where its theoretical security model justifies specialized optical infrastructure, operational complexity, and network constraints. The practical decision is not simply “quantum security versus non-quantum security”: both approaches still require authentication, trusted endpoints, secure key management, implementation assurance, and availability protections.

This comparison covers QKD against software-only lattice-based and hash-based schemes, focusing on security assumptions, cost, scalability, interoperability, and adoption conditions. The available research supports qualitative guidance, not an apples-to-apples total-cost or performance estimate.

Executive comparison

Security and threat-model differences

QKD’s strongest claim applies to an idealized key-establishment protocol against an attacker with unlimited computational power. That claim is conditional: it depends on correct protocol operation, parameter estimation, device behavior, the operating environment, and implementation security. It should not be interpreted as unconditional security for the entire communications system.^[28][29]

QKD does not authenticate communicating parties or classical messages by itself. Its classical sifting, error correction, and privacy-amplification steps require an authenticated classical channel; without authentication, a man-in-the-middle can impersonate the endpoints. QKD also does not automatically protect applications, private keys, key-management systems, or trusted relay nodes.^{[30][31][32][33]}

PQC instead assumes that selected mathematical problems or hash functions remain hard for classical and quantum attackers. This is a computational, not information-theoretic, security model. It can protect against “harvest now, decrypt later” attacks on stored ciphertext, but future cryptanalytic discoveries could weaken a scheme, and a compromised endpoint can still expose private keys or plaintext.^{[34][35][36][37]}

The implementation risks differ rather than disappear. QKD requires characterization of optical and detector behavior, including photon statistics, timing, spectral behavior, detection efficiency, dark counts, and after-pulsing. PQC deployments must address ordinary software and systems risks such as faulty randomness, parameter-handling errors, timing leakage, identity-binding failures, certificate problems, and private-key compromise. The supplied research does not provide a scheme-specific side-channel assessment for every ML-KEM, ML-DSA, or SLH-DSA implementation, so these should be treated as engineering threat-model considerations rather than universal findings.^{[38][39][40][41]}

Cost, scalability, and interoperability

Cost: QKD is usually the more infrastructure-intensive choice because it adds specialized quantum-optical equipment, suitable physical links, key-management components, and potentially trusted relays. PQC can generally reuse existing platforms and networks, although larger keys or signatures may increase bandwidth, memory, latency, retransmission, or constrained-device costs. The available evidence does not justify a universal price premium or a precise total-cost comparison.^{[42][43][44][45]}

Scalability: QKD is naturally constrained by the physical network. Optical loss and key-generation rates limit distance and capacity, and larger deployments may require hop-by-hop distribution, central sites, or trusted nodes. PQC avoids a QKD-specific physical topology and can be integrated across conventional enterprise, cloud, internet-facing, TLS, IPsec, and application environments, though cryptographic migration remains a substantial systems project.^{[46][47][48][49][50]}

Interoperability and lifecycle: QKD requires coordination among quantum links, conventional networks, key managers, application interfaces, and encryption systems. The reviewed material reports proprietary key managers, vendor-specific integration, multiple protocol families, and continuing integration difficulty. PQC is more compatible with software-based migration and hybrid designs, but organizations still need cryptographic discovery, application changes, certificate and PKI work, and crypto-agility so algorithms can be replaced if assumptions change.^{[51][52][53][54]}

The supplied maturity search was insufficient to support a detailed, current ranking of ML-KEM, ML-DSA, and SLH-DSA implementation maturity or standardization status. The evidence does establish their broad roles, but it does not justify unsupported claims about universal vendor readiness, comparative deployment maturity, or cross-vendor interoperability.

Deployment guidance

Use software PQC as the default when the objective is broad protection across existing networks, cloud services, applications, certificates, or stored data. Begin with cryptographic inventory and data-lifetime analysis, prioritize long-lived sensitive information, design for algorithm replacement, and use authenticated key establishment plus signatures appropriate to the protocol. A hybrid migration can provide continuity while systems and counterparties are upgraded, but the exact hybrid design should be validated for each protocol and implementation.

Bottom line

For a normal enterprise or public-sector migration, deploy software PQC first because it scales through existing infrastructure and avoids QKD’s specialized physical network and operational burden. QKD is a targeted defense for exceptional links, not a general replacement for authenticated, software-integrated cryptography. The final choice should follow the threat model: QKD may be justified when physical-link assurance and an information-theoretic objective dominate, while PQC is the practical default when coverage, scalability, interoperability, and lifecycle flexibility matter most.

Direct answer

The ISO standard number most commonly associated with collaborative robot safety is ISO 10218. ISO 10218-1 covers robot safety requirements, while ISO 10218-2 covers system integration and collaborative applications.^[1]

Current versus legacy context

Historically, professionals also searched for ISO/TS 15066:2016, whose collaborative-robot requirements were incorporated into ISO 10218-2:2025.^[2]