Closing the Diffusion Gap

A Deployment Agenda for LAC Policymakers

The current technology cycle is unfolding in a polarized policy environment, where expectations oscillate between “instant productivity miracles” and “another speculative bubble.” In Carlota Perez’s framing, some economies appear to be in an installation phase: finance is abundant for experimentation and asset revaluation, while deployment—broad, practical use across firms, sectors, and public services—lags.

Measured productivity typically rises after diffusion and reorganization, not at the invention stage, and this lag is a consistent feature of past general-purpose technologies such as electricity and information and communication technologies. The political economy of bubbles matters because mis-timed policy can lock countries into unproductive trajectories: capital can flow toward speculative assets and narrow “frontier” winners while the complements needed for economy-wide diffusion—skills, standards, infrastructure, and management capability—remain underbuilt.

The stakes for Latin America and the Caribbean are acute. The region’s labour productivity per hour worked stands at just 33% of the OECD average (OECD, 2024)—a gap that has widened, not narrowed, since 1990, when it stood at 40%. Productivity growth across the region stagnated at an average of 0.06% annually between 1960 and 2019 (IDB), and ECLAC now reports four consecutive years of low growth, with GDP expanding at roughly 2.2–2.4% per year—rates that are insufficient to address the region’s pressing socioeconomic needs. This low-growth trap is shaped by weak investment and persistently low productivity, which reduces the room to absorb adjustment costs during major technological transitions.

The problem, therefore, is not whether new technologies exist, but whether LAC countries can time their institutional and financing choices to accelerate diffusion before incumbents and legacy constraints close the window of opportunity. This essay proceeds in three steps: it first documents why diffusion drives productivity outcomes; it then explains when late entry can be an advantage and when legacy constraints dominate; and it closes by showing why institutions—coordination capacity across regulation, skills, infrastructure, credit, and demand—are the binding constraint in LAC. Policymakers should treat diffusion capacity as a macroeconomic priority, because the largest losses in this cycle are likely to come from delayed deployment rather than from a lack of innovation.

The diffusion gap, not the invention gap, drives productivity.

Productivity gains from a general-purpose technology accelerate when a critical mass of firms and workers adopt the technology and reorganize production around it, not when the technology is invented. Paul David’s work on electrification found that large productivity gains arrived roughly four decades after the early commercial use of electricity, because factories needed redesign and new skills before electric power could be used as a flexible “unit drive” rather than as a retrofit for steam-era layouts. The same logic appears in the “Solow paradox” era of computing: U.S. productivity acceleration became visible only after the mid-1990s, when information technology investment density crossed a threshold, and complementary organizational changes were absorbed. A contemporary version of this pattern is now being documented for industrial artificial intelligence: there is an initial, measurable productivity decline among AI adopters, followed by stronger performance later, describing a “J‑curve” in which established firms face short‑run friction as they integrate AI into production and management systems.

For LAC, the diffusion mechanism is visible in sectors where adoption has been broad, and the complements were built in parallel. Brazil’s agricultural transformation is the clearest regional benchmark: productivity growth became sustained only after EMBRAPA research was coupled with extension, credit, logistics, and standards over a multi‑decade period, illustrating that diffusion infrastructure can matter as much as the underlying science. EMBRAPA has developed and recommended more than 8,000 technologies for Brazilian agriculture, reducing production costs and helping increase Brazil’s food supply. Chile’s renewable buildout also reflects diffusion and reorganization: it combined standardized auctions and contracting with rapid scaling of solar and wind, demonstrating that policy design can compress diffusion timelines by reducing transaction costs and increasing investor confidence.

Diffusion also explains why “frontier success” can coexist with weak aggregate productivity. Global frontier firms continued to post strong productivity growth even as overall productivity slowed, because diffusion to “middle‑tier” firms stalled; in LAC, the gap is reinforced by firm heterogeneity and informality, where many enterprises remain outside the reach of training, credit, and standards that enable productive adoption. The average informal employment rate across the region was 47.3% in 2023, with nearly 50% of SMEs operating informally and therefore structurally excluded from the extension services, financial instruments, and standards that enable productive adoption (OECD, 2024). The result is that measured productivity can remain flat even when technology use is visible in pockets: an economy can have world‑class adopters in mining, finance, or platform retail while most firms continue to operate with low management quality and weak process control. Management practices account for a substantial share of productivity differences across LAC firms, suggesting that the diffusion bottleneck is often organizational rather than technological. In practical terms, diffusion means building repeatable adoption pathways—procurement standards, interoperable platforms, skills pipelines, and extension services—that allow thousands of firms to reorganize, not just a handful to experiment. Without that breadth, the economy accumulates the costs of experimentation while the benefits remain concentrated and statistically small.

Today’s LAC payment systems offer a high‑frequency diffusion example with unusually granular evidence. Brazil’s Pix system was designed as a public interoperability platform, and the Banco Central do Brasil reports that Pix has reached more than 170 million individuals and processed over 7 billion transactions in January 2026, with an official daily record above 313 million transactions in December 2025. These numbers indicate diffusion at scale, achieved through institutional design choices—mandatory participation and standardized interfaces—rather than through pioneering technology. While the macroeconomic productivity effects of Pix are still being studied and should be treated cautiously, the adoption curve itself illustrates the core argument: when rules make adoption easy, and network benefits are universal, diffusion can be faster than in legacy-heavy payment ecosystems. The policy inference is that productivity improvements are most likely to appear after such systems are embedded in business processes—such as inventory management, invoicing, credit scoring, and tax compliance—not merely after the payment rail is launched.

When late entry helps—and when incumbents prevail

Gerschenkron’s argument that “economic backwardness” can create advantages remains relevant because latecomers can adopt mature technologies at lower cost than pioneers, provided switching costs are low, and incumbents cannot block entry. Domains where legacy systems were weak or fragmented, such as digital payments and renewables, are prominent in adoption because they benefit from modular technologies and strong network effects once a common standard is adopted. Brazil’s Pix illustrates a latecomer advantage in a large economy with legacy banking; by choosing a centralized, interoperable design, the central bank reduced the value of proprietary payment networks and accelerated adoption. The Banco Central do Brasil’s Pix statistics, showing very large user and transaction counts, support the claim that institutional mandates can substitute for decades of incremental market building in network industries. Similarly, Chile’s solar leap is a case in which mature global technology could be deployed quickly once contracting and risk allocation were standardized. The latecomer mechanism here is not inventing panels or turbines but importing mature equipment and accelerating deployment through credible rules and bankable contracts. The implication for policymakers is that “late entry” is best understood as a governance opportunity: it lowers technological uncertainty while raising the returns to institutional design.

This latecomer logic extends to artificial intelligence, where LAC’s position is revealing. The region accounts for 14% of global visits to AI tools despite representing only 11% of global internet users—suggesting demand and curiosity are not the binding constraint. Yet despite representing 6.6% of global GDP, the region attracts only 1.12% of global AI investment (ECLAC’s Latin American Artificial Intelligence Index, ILIA 2025). The gap between user appetite and investment flows illustrates precisely the latecomer’s dilemma: access to mature tools is available, but the institutional scaffolding—governance frameworks, talent pipelines, and risk-sharing mechanisms—that would convert that access into productive diffusion remains thin.

However, the latecomer advantage operates unevenly in LAC because legacy constraints are often political rather than technical. Where sunk assets are large and politically connected, incumbents can slow diffusion even when technologies are commercially viable. Mexico’s electricity reforms—legal changes requiring at least 54% of electricity to be dispatched by the state utility and dissolving independent regulators in favor of a centralized commission reporting to the executive—provide a recent illustration of how rules can reallocate market access and thereby shape diffusion speed. Whatever the stated objectives, such changes can increase regulatory uncertainty for private deployment in generation and grid services, and uncertainty is a known deterrent to capital-intensive diffusion that relies on long-horizon revenue recovery. The Mexico situation aligns with the broader point that the constraint is not hardware availability—solar modules and storage systems are globally tradable—but the institutional pathway that determines who can connect, sell, and recover costs. In sectors like energy and transport, where investment is long-lived and politically salient, latecomer advantage is therefore conditional on whether incumbents are constrained and whether rules remain stable across political cycles.

Chile’s experience also shows that latecomer deployment creates new constraints that must be addressed quickly to prevent a second, integration-related lag. Renewable curtailment in Chile reached 6,084 GWh in 2025, reflecting transmission congestion and daytime oversupply as the generation mix becomes increasingly renewable. The transmission challenge is not evidence against renewable diffusion; it is evidence that the binding constraint has shifted from investment to system integration—transmission, storage, and demand-side electrification. In Gerschenkron’s terms, latecomers can import mature generation technologies, but they still must build the complementary institutions and infrastructure that allow the technology to be used productively. The broader lesson for LAC is that late entry compresses time: it can speed initial deployment, but it also forces governments to coordinate integration investments earlier in the diffusion curve than pioneers did. Where that coordination fails, latecomer advantage is converted into stranded assets, higher system costs, and political backlash that can slow subsequent waves of deployment.

Coordination capacity: the real policy bottleneck

The operational bottleneck in deployment is the capacity to coordinate institutional change across multiple domains simultaneously: regulation, skills formation, infrastructure, credit, and demand creation. The historical benchmarks make the same point in different technological languages. Electrification required factory redesign, standardization of equipment, worker retraining, and reliable grids; computing required new business processes, enterprise software, and data systems; and today’s AI tools require governance over data access, cybersecurity, procurement, and workforce adaptation. Looking at industrial AI is instructive because early productivity declines after adoption imply that integration frictions—training, workflow redesign, and managerial learning—dominate in the short run before benefits scale. Management practices are a major source of productivity differences, reinforcing the point that “soft” complements can be as binding as physical infrastructure. This observation is policy-relevant because it shifts attention from subsidizing tools toward building the capabilities to use tools well at scale.

In LAC, institutional variation helps explain why some countries deploy faster than others, even with similar access to global technology. ECLAC’s Latin American Artificial Intelligence Index (ILIA 2025) classifies countries into three tiers: Chile, Brazil, and Uruguay lead as “pioneers,” followed by “adopters” such as Colombia, Ecuador, Costa Rica, and the Dominican Republic—while over one-third of LAC countries remain “explorers,” with limited capacity and early-stage ecosystems. The spread across these tiers tracks institutional investment more closely than income level alone. Diffusion agencies and mission-oriented institutions such as EMBRAPA and CORFO operate as bridges between technology and users by providing extension, de-risking, and coordination. Regulatory design can convert network effects into public goods, as in Pix’s mandated interoperability, or leave them as private tolls that fragment markets. The empirical pattern is consistent with the diagnosis of LAC in a low-growth trap: weak investment and limited policy space mean that countries cannot afford long periods in which adoption costs rise while productivity does not. This increases the value of institutions that can shorten the “learning and reorganization” period by providing predictable rules and rapid problem-solving capacity across agencies.

The institutional constraint is most visible when diffusion reaches system-wide thresholds and cross-sector dependencies become binding. Chile’s curtailment problem illustrates how deployment success creates a new coordination challenge: transmission planning, storage regulation, and demand creation must move together if renewable generation is to translate into lower system costs and broader productivity gains. Similar threshold effects can be expected in digital transitions: instant payments become productivity-relevant when integrated into invoicing, tax systems, and working-capital finance; broadband becomes productivity-relevant when firms adopt complementary software and skills; and AI becomes productivity-relevant when organizations redesign workflows and governance to use it reliably. For policymakers, the practical implication is that “technology policy” cannot sit in a single ministry. It requires delivery capability that can align regulators, finance ministries, training systems, and infrastructure agencies on a shared deployment path with clear milestones and feedback loops. Without that coordination, diffusion can still occur, but it will tend to be shallow, uneven, and slow to register in productivity statistics.

Building deployment states in LAC.

There is a simple but demanding conclusion: productivity is a lagging indicator because it reflects diffusion and reorganization, not the presence of new technology. The first takeaway for LAC policymakers is that the relevant unit of action is the adoption system, not the innovation event. Historical experience with electricity and information technology indicates that productivity gains arrive after complementary investments—skills, reorganization, standards, and infrastructure—become widespread. Policies that fund pilots and prototypes without building diffusion channels raise costs while leaving productivity unchanged, because the complements that turn tools into output remain scarce.

The second takeaway is that timing advantages exist for latecomers, but they are conditional. LAC countries can import mature technologies cheaply and deploy quickly when switching costs are low. Rules create open access, as the diffusion of Pix and the rapid scaling of renewables in parts of the region illustrate. Yet legacy political economy can still block or delay deployment, and recent regulatory changes in Mexico’s power sector show how market design can reshape incentives for private investment and system modernization. The third takeaway is that institutions, not access to technology, are the binding constraint. The countries that move fastest are those able to coordinate regulation, finance, skills, infrastructure, and demand creation simultaneously, and to adapt quickly when deployment creates new system constraints such as grid congestion or AI talent shortages.

A policy agenda consistent with these findings would treat diffusion capacity as economic infrastructure. That means investing in delivery institutions that can scale adoption across small and mid-sized firms—including the roughly 50% of LAC SMEs that currently operate informally and remain outside formal extension networks. It means designing interoperable standards where network effects matter and sequencing complementary investments so that skills and infrastructure expand alongside deployment rather than years later. It also implies setting realistic evaluation horizons: early-stage productivity metrics can understate impact during integration periods, so performance frameworks should track diffusion milestones—adoption rates, interoperability coverage, training completions, and system bottlenecks—alongside output measures. For LAC, where fiscal space is limited, and productivity has stagnated for decades, shortening the diffusion-to-productivity lag is not a technical preference but a macroeconomic necessity. The call to action is therefore administrative and fiscal rather than rhetorical: LAC governments should build “deployment states” capable of moving proven technologies into broad productive use, because that is the point in the cycle when productivity begins to rise and when latecomer advantages can be converted into durable structural transformation.

Key sources

OECD (2024), Informality and Households’ Vulnerabilities in Latin America; ECLAC, Latin American Artificial Intelligence Index (ILIA 2025); IDB, 2025 Latin American and Caribbean Macroeconomic Report; ECLAC, Latin American and Caribbean Economies — Growth Projections 2025–26; Banco Central do Brasil, Pix statistics (January 2026); IMF Working Paper “What Can Artificial Intelligence Do for Stagnant Productivity in Latin America and the Caribbean?” (October 2024); SCIELO, “Controversies about the Process of Technology Transfer from Public Research Institutions in Brazil: The Case of EMBRAPA” (2014).