Why the Future Economy Still Depends on Basic Digital Foundations

The future economy may be powered by algorithms, but it is still built on networks. As governments race to fund AI strategies and digital-finance ecosystems, a new OECD study warns that the strongest economic signal comes from something far less fashionable: digital infrastructure.

Published in Economies, "Financial Technology, Artificial Intelligence and Machine Learning: Is the Digital Economy the New Economy?" by George Gkouskos, George Galanos and Charalampos Agiropoulos examines which digital pillar matters most, and finds that infrastructure does most of the heavy lifting.

Infrastructure Is the Load-Bearing Layer

Digital infrastructure, measured through internet use, fixed broadband, mobile subscriptions and secure servers, explains about 24% of within-country variation in the study's clean New Economy Index. The traditional-economy block explains essentially none, while the combined three-pillar Digital Architecture Index explains only about 9%.

Composite indices often appear more comprehensive because they combine multiple dimensions. Here, however, combining infrastructure, FinTech and AI/ML capability weakens rather than strengthens the fit. The three pillars are highly correlated, and infrastructure appears to carry most of the measurable signal.

The implication is not that digital finance or AI are economically unimportant. Both depend on the foundational systems that allow data, transactions and applications to move securely and at scale. FinTech requires reliable connectivity and interoperable networks. AI requires computing access, data flows, cloud capacity and secure digital environments. Infrastructure is therefore not merely one pillar beside the others; it is the layer on which the others operate.

The researchers describe the digital economy as a layered system in which traditional functions, production, finance and decision-making, are increasingly reorganized through digital channels. Infrastructure forms the foundation, FinTech operates through that base, and AI and machine-learning capabilities cut across the wider system.

For policymakers, this reframes the sequencing problem. Economies may be tempted to pursue advanced AI strategies or digital-finance ecosystems before fixing weak broadband, insecure networks or low-quality digital access. The study suggests that such strategies risk building sophisticated applications on an unstable base.

The AI–FinTech "Super-Effect" Does Not Survive Scrutiny

The researchers also test whether infrastructure, FinTech and AI/ML capability create a distinct joint effect beyond their separate contributions. The evidence is weak. In the baseline model, the three-way interaction is positive but not statistically significant under the preferred country-clustered errors. It becomes significant under some alternative error structures but loses robustness when the model is tested more strictly.

The joint association disappears when the analysis uses only actual Global Findex survey years. It turns negative in a dynamic specification, changes sign when a broader outcome is used and is matched by a future-dated placebo. Those failures matter because they suggest that the apparent relationship may reflect slow-moving common trends rather than a distinct economic mechanism.

The study does not convert suggestive correlations into a triumphant technology narrative. Instead, it treats the fragility of the result as a finding in itself. The restraint is especially relevant amid the global rush to frame AI, digital payments and data systems as mutually reinforcing engines of growth. The study does not disprove complementarity. It shows that the available OECD panel data cannot establish it reliably.

The AI measure itself also requires care. Because long-run, comparable cross-country data on enterprise AI adoption remain limited, the authors use R&D spending, patents and scientific publications as a proxy for AI/ML capability. This measures a country's knowledge base, not whether firms are actually deploying AI.

The proxy is nevertheless validated against Eurostat enterprise-AI adoption data. Across 24 European economies, the correlation reaches 0.80 for 2021 and 0.81 against the 2024 wave. That is strong evidence that the index captures something real, but it does not erase the gap between technological potential and productive use.

Digital Leaders Are Pulling Ahead, but Measurement Still Lags

The study's country comparisons show a familiar divide. Nordic economies, South Korea and the Netherlands rank near the top of the Digital Architecture Index, while several emerging OECD members remain below the sample average.

The digital-minus-traditional gap also widens over time. This suggests that digital systems are becoming more central to economic organisation. But the authors warn that the gap is partly definitional because the indices are built to move in opposite directions. It is best read as a benchmark of structural divergence, not as proof of a causal transition.

The same caution applies to the "new economy" outcome. The preferred index combines GDP per person employed with new business density. That captures productivity and entrepreneurial dynamism, but not digital-sector output, consumer welfare, platform activity or the distribution of gains.

The United States is also excluded from the preferred clean-outcome model because the relevant World Bank series lacks new-business-density data. The authors test a Census-based proxy and obtain broadly similar results, but the omission still matters because the world's largest frontier digital economy is absent from the headline specification.

FinTech data present another limitation. The Global Findex series contains only a handful of true survey waves. Annual values between those waves are interpolated, while later years are extrapolated. That creates smoother annual variation than the observed data can genuinely support and may understate uncertainty.

These measurement constraints are not peripheral. They point to a larger governance problem: digital policy is moving faster than official statistics. Governments are making major decisions on AI readiness, digital finance and platform regulation while cross-country data remain fragmented, delayed or conceptually inconsistent.

The study's open-data framework is therefore valuable even where its causal claims are limited. It provides a transparent method that statistical agencies, researchers and international organisations can rebuild, test and improve.

The Policy Lesson: Build the Rails Before Buying the Trains

Governments should not interpret the findings as proof that every broadband investment will raise productivity. The infrastructure coefficient is not precisely estimated under the strictest error structure, and a future-dated infrastructure placebo also loads positively. Common trends may therefore explain part of the association.

Even so, digital infrastructure remains the clearest descriptive counterpart of new-economy performance in the study. This makes it a priority for monitoring, especially when compared with fashionable but weakly measured claims about AI–FinTech synergy.

For advanced economies, the agenda includes network quality, cybersecurity, cloud access, secure servers, competition and affordability. For developing countries and Global South stakeholders, the stakes may be even higher. Scarce public funds can be diverted toward prestige AI initiatives while basic connectivity, reliable electricity, digital identification and interoperable payment systems remain incomplete.

The broader development lesson is that technologies generate returns through systems, not isolated projects. Digital infrastructure can support SDG 9 on industry, innovation and infrastructure, SDG 8 on productivity and entrepreneurship, and SDG 10 if access is distributed widely enough to reduce rather than reinforce inequality. But infrastructure alone is not sufficient. Skills, institutions, competition policy, data governance and firm-level adoption determine whether digital capacity becomes inclusive economic value.

Future research needs to move from cross-country co-movement toward causal identification. Policy shocks, broadband rollouts, regulatory changes and natural experiments could help isolate effects. Firm- and sector-level data could reveal where infrastructure, FinTech and AI actually raise productivity, and who captures the gains. Better official AI-adoption statistics are particularly urgent.