The hard part of innovation: How technologies diffuse – and why institutions matter

Few gatherings are as saturated with technological possibility as the World Economic Forum's Annual Meeting.

Artificial intelligence, advanced manufacturing, clean energy, biotechnology, and autonomous systems – each year, Davos brings fresh demonstrations of human ingenuity and accelerating technical capability. The dominant impression, reinforced by headlines and stagecraft alike, is one of momentum: innovation racing ahead, breakthroughs cascading from one sector to the next, a future arriving faster than institutions seem able to absorb.

In one sense, this moment reflects a vision long anticipated here. Nearly a decade ago, Klaus Schwab described a Fourth Industrial Revolution defined not by a single technology, but by the convergence of digital, physical, and biological systems – reshaping economies, societies, and governance all at once. That convergence is no longer theoretical. It is now visible across sectors and geographies, from AI-enabled manufacturing to gene-based medicine, from electrified transport to data-driven supply chains.

And yet, across Davos this year, a more sober and surprisingly consistent message emerged beneath the surface of frontier excitement. The hard part of innovation today is no longer invention. It is building the institutions, infrastructure, and trust needed to diffuse and deploy new technologies at scale – without amplifying risk, inequality, or systemic fragility.

This distinction matters. For much of the past two decades, innovation discourse has been dominated by the frontier: the next model, the next molecule, the next performance benchmark. Progress was measured by novelty and speed. But as Fourth Industrial Revolution technologies move from laboratories into workplaces, hospitals, grids, supply chains, and public services, a different set of constraints comes into view. Scaling innovation is no longer simply a matter of technical feasibility. It becomes a social, economic, and political challenge, shaped by governance capacity, capital allocation, workforce readiness, and public legitimacy.

Breakthroughs are arriving faster than the systems designed to deploy them. Electricity grids strain under new forms of demand. Health systems struggle to integrate digital tools without exacerbating access gaps. AI systems outperform expectations in controlled settings but encounter friction when embedded in real-world workflows. Even technologies widely seen as essential – clean power, nuclear energy, data-driven supply-chain oversight – face delays not because the science is uncertain, but because institutions are misaligned with the pace and scale required.

This is where responsibility enters the picture – not as an abstract ethical add-on, but as a practical requirement for diffusion and scale. Innovation deployed without adequate guardrails can magnify inequality, concentrate power and erode trust. Excessive caution or fragmented regulation, however, can stall adoption, leaving societies stuck with brittle systems ill-suited to current pressures. The challenge is not to choose between speed and safety, but to build pathways that allow innovation to spread broadly while remaining accountable and resilient.

Innovation falters not at the point of discovery, nor even at proof of concept, but in the messy middle between demonstration and broad adoption. This is where diffusion slows, costs rise, institutions hesitate, and responsibility becomes inseparable from scale.

Energy offers some of the clearest illustrations. Advances in generation technology are no longer the primary constraint. Wind, solar, nuclear, and storage technologies are well understood, and investment interest is substantial. Yet electricity grids remain underbuilt, permitting timelines stretch for years, and pricing structures often obscure real costs and incentives. As demand grows – from data centres, electric vehicles, and industrial electrification – systems designed for a different era strain to adapt. The challenge is not whether clean power can be produced, but whether institutions can coordinate infrastructure, finance, and public consent at the speed required.

Healthcare presents a similar picture. Digital tools, AI-assisted diagnostics, and data-driven systems promise measurable gains in efficiency and outcomes. In controlled settings, many already outperform existing approaches. But diffusion into real health systems exposes different bottlenecks: fragmented data, workforce constraints, reimbursement models that reward volume over outcomes, and persistent concerns about trust and accountability. Technologies designed to improve care can, if poorly deployed, widen access gaps or add administrative burden. Here again, responsibility is a prerequisite for adoption at scale.

Artificial intelligence, despite its prominence, follows the same logic. Much of the public conversation remains focused on frontier capabilities, but enterprise deployment tells a more restrained story.

AI systems often perform impressively in isolation, yet encounter friction when integrated into everyday workflows. Questions of reliability, oversight, liability, and user trust quickly dominate. In practice, adoption depends less on model performance than on organisational change: how work is redesigned, how outputs are validated, and how humans remain meaningfully in the loop. To be sure, there have already been instances of highly successful diffusion, delivering large and measurable gains at enterprise scale. Those cases underscore the same point: diffusion succeeds not by accident, but where institutions are able to absorb change deliberately and at pace.

Even in industrial manufacturing and logistics – sectors often seen as better suited to automation – scaling innovation exposes institutional seams. Advanced tools can raise productivity, but only where supply chains, standards, workforce skills, and financing align. Without those conditions, innovation remains uneven, concentrated among early adopters rather than diffusing across entire sectors.

What ties these cases together is not technological immaturity, but institutional lag. Systems built to manage incremental change struggle with technologies that evolve quickly and cut across traditional boundaries. Regulation is often sector-specific, while innovation is not. Capital is available, but patient capital is scarcer. Skills gaps widen faster than training systems adapt. Public trust, once lost, is difficult to rebuild.

These frictions help explain why diffusion has emerged as a central challenge. Scaling innovation responsibly is not simply about making technologies cheaper or faster. It is about aligning incentives, redesigning institutions, and building confidence that adoption will deliver durable benefits rather than short-term disruption.

If innovation breaks down most often in the transition from demonstration to adoption, the obvious question is what distinguishes the cases where diffusion does work. Davos offered no single blueprint, but it did surface a set of recurring conditions that make large-scale, responsible deployment more likely.

Successful diffusion depends first on institutional ownership, not just technical sponsorship. Technologies that spread beyond pilots tend to have clear executive backing and defined accountability for outcomes. Diffusion stalls when responsibility is fragmented across innovation units, IT departments, regulators, and frontline operators. It accelerates when deployment is treated as a core operational priority rather than an experimental add-on.

Second, diffusion requires integration into existing systems, not parallel ones. Many technologies perform well in isolation but struggle when layered onto legacy infrastructure or entrenched workflows. Where diffusion succeeds, organisations invest early in interoperability, data integration, and process redesign. This work is unglamorous, but it determines whether new tools reduce friction or add to it.

Third, scale depends on workforce readiness and redesign, not substitution alone. Across sectors, durable gains came not from removing humans from the loop, but from redefining their role within it. Technologies diffused more quickly where workers were trained to interpret outputs, exercise judgment, and intervene when systems failed. Where adoption was framed as replacement, resistance grew and diffusion slowed.

Fourth, successful diffusion relies on patient capital and longer time horizons. Many of the gains discussed at Davos accrued over years, not quarters. They required upfront investment in data, infrastructure, and training before returns became visible. Short-term performance pressures can be fatal to diffusion, particularly in sectors where reliability and trust matter as much as speed.

Finally, diffusion depends on trust, both internal and external. Internally, users must trust that systems are reliable and aligned with organisational goals. Externally, customers, regulators, and the public must trust that deployment will not compromise safety or accountability. Trust is built not through assurances, but through design choices that make systems auditable, explainable, and responsive to failure.

Taken together, these conditions help explain why diffusion is uneven. They also clarify why some organisations have already realised large, measurable gains from new technologies, while others remain stuck in pilot mode. Success is rarely accidental. It reflects sustained effort to align technology with institutions capable of absorbing change without breaking.

If diffusion is difficult but not impossible, what distinguishes the cases where it has worked? This year’s Annual Meeting offered a small number of unusually clear examples, notable less for novelty than for the conditions under which technologies were deployed.

One of the most concrete came from Amin Nasser, president and chief executive of Saudi Aramco, who described how artificial intelligence has been integrated across core operations at the company. Rather than treating AI as a standalone capability, the company embedded it into upstream and downstream systems – from reservoir modelling and geosteering to predictive maintenance, logistics, and supply-chain optimisation. Aramco tracks what it calls “technology realised value,” reflecting gains already captured in operations rather than projected savings. Nasser said that in 2023 and 2024, this figure reached roughly $6 billion, with about half attributable to AI, and that additional multi-billion-dollar gains were expected in 2025, pending third-party verification.

Much of that value has come from AI embedded directly into capital-intensive, safety-critical processes. AI-enabled earth models and real-time analytics allow drill bits to be steered more precisely underground, anticipating equipment stress and identifying problems before failure occurs. The ability to pull equipment out of wells pre-emptively, rather than repairing it in situ, has reduced downtime and improved productivity in some wells by 30–40 percent. The significance of the example lies not in the technology alone, but in the fact that AI has reshaped operational decision-making at scale.

A similar logic emerged in the session on autonomous systems, where Shao Tianlan of Mech-Mind Robotics described the large-scale deployment of task-specific industrial robotics. By focusing on narrowly defined use cases and prioritising reliability, the company has been able to scale deployments into the thousands, shortening learning cycles and delivering measurable productivity gains. Systems were built with fallback mechanisms, teleoperation, and clear performance thresholds, allowing autonomy to expand incrementally without eroding trust among users.

These cases help clarify remarks made elsewhere at Davos by Jensen Huang, the chief executive of Nvidia, and Satya Nadella, chief executive of Microsoft, both of whom emphasised that the value of advanced technologies depends less on frontier performance than on infrastructure, integration, and organisational change. Compute, power, data, and capital are necessary, but insufficient on their own. What ultimately determines impact is whether institutions are capable of absorbing new tools into everyday operations.

Taken together, these examples reinforce a broader point. Successful diffusion is rarely accidental, and it is rarely driven by technology alone. Where institutions combine scale with accountability, patience with discipline, and ambition with operational realism, innovation can move from promise to practice at speed.

The picture that emerged at Davos this year was not one of technological stagnation, nor of unbridled acceleration. Innovation is advancing rapidly across domains. What lags is the capacity of institutions to absorb that innovation at scale without destabilising the systems on which economic and social life depend.

This is the institutional reckoning of the Fourth Industrial Revolution. Technologies that once operated at the margins are now embedded in core systems. At this scale, diffusion is not a technical afterthought. It is the main event.

Responsibility, in this light, is not a brake on innovation. It is the condition that makes innovation durable. Responsible diffusion means designing systems that can fail safely, adapt quickly, and earn trust through performance rather than promise.

For leaders, the implication is straightforward but demanding. Supporting innovation is no longer enough. The harder work lies in funding the unglamorous foundations that allow innovation to spread: infrastructure, skills, governance, and trust. These investments rarely command attention, but they increasingly determine whether innovation delivers lasting value.

If earlier waves of technological change were defined by disruption, the promise of this one lies in durability. The task ahead is not to choose between speed and responsibility, but to recognise that, at scale, the two are inseparable.

Watch the full session 'Scaling AI: Now Comes the Hard Part':