The industrial AI divide

How AI leaders are pulling ahead across transportation, logistics, and defense.

Aerial view of a container ship with AI data overlays illustrating the industrial AI divide in transportation, logistics, and defense.

Executive summary

Artificial intelligence is becoming the next competitive divide in transportation, logistics, and defense (TLD). The shift is already visible in the numbers: 12% of transportation, aviation, and automotive CEOs report AI-enabled revenue gains above 20%, the highest share of any industry surveyed in the Oliver Wyman Forum and the New York Stock Exchange’s 2026 CEO Agenda. As AI moves from pilots and productivity tools into core operations, a small group of companies is pulling away — improving asset utilization, accelerating engineering cycles, reducing downtime, and building operating advantages that compound with every deployment.

The shift is already reshaping how these industries, from aviation and rail to automotive and logistics, are designed, built, operated, and monetized. The companies embedding AI into their operating models are creating learning loops and competitive edges across the organization: Each deployment generates operational data, process knowledge, new edge cases, and organizational confidence to make the next deployment stronger.

Meanwhile, many legacy players remain uncommitted and risk waking up to a gap that is increasingly difficult and expensive to close. In asset-heavy, safety-critical, and regulated industries, advantage will come from industrializing AI, not simply deploying more use cases. That means building the operating discipline to scale AI safely across physical assets, frontline workflows, and mission-critical decisions.

To better understand the growing divide and how late movers can take practical steps to scale AI, the Oliver Wyman Forum and the University of California, Berkeley, worked with an advisory panel of academic and corporate leaders to interview more than 50 transportation and industrial experts. This report draws on those interviews and on recent Oliver Wyman Forum surveys of chief executive officers and chief financial officers, consumer research, and market analysis across aerospace, automotive, aviation, defense, industrial goods, infrastructure, logistics, and rail.

While AI is constantly advancing, the technology available today is not a constraint to creating immediate value. Industry players already should be deploying AI technologies such as pattern learning and optimization, virtual simulation and digital twins, AI agents, physical automation and robotics, and autonomous mobility.

And yet, successful AI adoption is uneven. Firms now fall into one of three archetypes in how they scale AI: A small group of leaders that embed AI into operating models, products, and management systems; adopters that use AI to improve existing core processes and see significant impacts, but have not yet turned AI into enterprise-wide advantage; and late movers that remain slow to scale AI.

In the next five to 10 years, AI will enable TLD firms to move toward entirely different operating models, such as rail systems that self-optimize and absorb disruptions autonomously, or self-dispatching freight networks.

To unlock that potential, firms must build the right conditions for AI to scale rather than just add more pilots or technology investments. Firms that scale AI throughout their business will multiply their edge: Better operations generate better data; better data improves models; better models improve decisions; and better decisions further strengthen the operating model.

This report explains what transportation, logistics, and defense CEOs and boards must do to avoid falling behind, and identifies where value pools are shifting.

Several key lessons emerge from our analysis and the experts’ perspectives:

1

Scaling AI is a company-wide, vision-driven transformation that must be led at the board and C-suite levels. Companies need a multiyear roadmap, but one adapted to the fast pace of the AI age. Execution should be governed through quarterly or even monthly milestones, and supported by agile approaches with investment in AI’s foundational capabilities in data, infrastructure, governance, talent, and change management.

For legacy players, scaling AI is structurally harder because they must manage costly brownfield assets and fragmented technology environments. These firms must rethink operating models entirely rather than rely on incremental progress, an approach that won’t close a compounding gap against the AI leaders.

Making that transition requires massive investments and a focus on a select number of high-impact use cases that can help deliver ROI, such as engineering cycles or fleet maintenance. These investments must be supported by short- and medium-term gains, such as improved productivity, faster product development, or better asset utilization. Note that CEOs who scale AI across two or more use cases report roughly twice the ROI in cost savings and revenue gains compared with those who don’t, according to the Oliver Wyman Forum and New York Stock Exchange’s 2026 CEO survey.

2

AI transformation in TLD is too often delegated to the IT department when it should be owned by the business. AI is not a mere toolset for the IT department. It will enable agents, robots, and new core business tools to integrate into many elements of a company’s operating model. In time, AI will also reshape or even replace parts of existing IT environments and business management software. Treating AI as just another technology to integrate, rather than as a catalyst for operating model transformation, is one of the most important reasons that TLD companies fail to scale it.

3

AI technology is ready to reshape operating models — firms that wait to implement it risk falling behind. New AI capabilities are constantly emerging, but the AI technologies that can make a difference today are already available, from predictive maintenance to route optimization. Legacy players that wait for the next AI models or for strong ROI use cases may find it harder to catch up to their competitors, as their existing IT infrastructure, governance, and operating models become more misaligned with AI’s capabilities and what effective deployment requires. More importantly, they lose the learning cycles needed to uncover weaknesses in areas like data management, talent models, and governance.

The cost of catching up increases with each missed development cycle. Consider, for example, that 58% of aviation maintenance, repair, and overhaul professionals said that their value expectations from AI investments are being met or exceeded (up from 20% in 2024).

4

Data must be treated as the most critical asset to fuel an AI-powered organization. For many firms, the first step to catching up with competitors is fundamentally reconsidering the way data is collected, governed, and used. Without quality data, there is no scalable AI. Data generated from fleets, flight routes, or passenger experience is a strategic asset. Making data governance a recurring agenda item for executive committees, with a chief data officer accountable for it, can formalize that process.

5

The shift toward smart TLD networks and human-machine collaboration makes a new talent strategy critical. In industries heavily reliant on factories and manufacturing, leaders must anticipate which activities will be performed by AI agents, robots, or autonomous equipment, which human roles and capabilities will become more valuable, and how humans and AI can collaborate on the factory floor or in the control room. That anticipation will give firms an advantage in redesigning next-generation workflows.

Companies that wait risk encountering AI talent shortages in critical areas such as engineering or robotics.

6

Guardrails against AI’s constraints and risks must be embedded in the operating model, ensuring best practices across safety, cybersecurity, and regulatory compliance. Regulation varies significantly across geographies and markets, including laws that dictate what AI-enabled equipment can do, where liability resides, and what oversight is required. The reputational consequences of an accident in transportation or in warehouse operations also heighten the risk. Scaling AI inherently creates new security and bias challenges that require safeguards and clear parameters across the full execution chain.

7

Companies must monitor the market for AI-native disruptors and consider targeted acquisitions. As with drones and driverless vehicles of the last decade, AI poses a threat to companies that fail to adapt. New entrants are entering TLD sectors unburdened by legacy assets or outdated ways of working, and they could pose an existential risk to legacy players. Identifying these disruptors early, adapting quickly, and making the right acquisitions will be critical to competing through this transition to a new industrial era.

Aerial view of urban traffic with AI data overlays illustrating how AI is reshaping transportation, logistics, and defense competitiveness.

AI

AI is opening a structural and competitive divide across transportation, logistics, and defense (TLD)

TLD’s competitive AI divide is real and tangible. Only 12% of transportation, aviation, and automotive CEOs, for example, report AI revenue gains above 20%, the highest of any industry. Driven by technology, proprietary data, operational capability, and organizational readiness, companies that are benefiting from AI are building a strategic moat around themselves.

Two groups trail behind the AI leaders that are embedding AI at scale across company-wide operations: AI adopters, which are using AI to improve existing processes but have not yet turned it into enterprise-wide advantage, and AI late movers, those with limited or poor integration, which remain slow to commit capital, build data foundations, redesign workflows, or establish the governance needed to embed AI safely in operational environments.

The automotive industry shows how this lead multiplies: AI-native competitors are accumulating proprietary training data and iterating on software that increases vehicle efficiencies or provides innovative add-ons, while incumbents are still modernizing the data infrastructure required to begin. Consider also where investment funding is flowing: Recent AI advances provided the connected vehicle and self-driving sectors the biggest funding growth among all mobility sectors in 2024, with global funding reaching $18.2 billion — double from 2023, according to Oliver Wyman analysis.

More than half of transportation executives see AI as an opportunity

Questions: “What are the top 3 threats for your business?,” “What are the top 3 opportunities for your business?,” “What are your top 3 priorities for increasing shareholder value in the next 1 to 2 years?”
Source: Oliver Wyman Forum x New York Stock Exchange CEO Survey 2026 (N=415)

For late-moving rivals, what makes this dynamic existential rather than merely competitive is that AI learning compounds in both directions. Early movers improve their models with every deployment cycle and failure. A near-miss in a validation run, an edge case in a maintenance prediction, or an anomaly in a production line are all data points that make the next model better.

Late movers will have more difficulty catching up to this group, as they’re starting without accumulated data, lessons, and operational experience.

Transportation outpaces other industries in AI revenue realization, yet meaningful monetization remains out of reach

Share of CEOs declaring AI-related revenue generation greater than 20%

Question: “What is the financial impact of your AI initiatives as of today as a percentage of your company’s total revenue generation?”
Source: Oliver Wyman Forum x New York Stock Exchange CEO Survey 2026 (N=415)

Data and organizational readiness are widening the gap in AI competitiveness

Many firms are still slow to adopt AI. Less than a third of CEOs in industrials, aerospace, and defense are even piloting AI, versus 49% across all CEOs, and only 8% of CEOs from these sectors report using AI at scale, versus 23% of CEOs overall.

An intensifying penalty for waiting takes different forms depending on the business. AI-native companies must compete in every development cycle to ensure that their models don’t fall behind in performance, proprietary training data, and product capability. For incumbents integrating AI into operations, the penalty is organizational: Late movers must modernize data infrastructure, processes, and skills simultaneously, under competitive pressure, with fewer safe cycles to experiment. In both cases, early movers absorb disruption incrementally while latecomers face a compressed, higher-risk transformation.

In addition to data, the compounding barrier also spans talent, governance, and organizational confidence, which are dependent on each other for development, and cannot be built overnight. Embedding AI technologies and applications into the operating model should not be treated as a one-off initiative but as a major multiyear transformation.

Integrating AI into TLD is a transformative change that takes time and careful planning

Not all AI use cases have the same time constraints: Copilots and workflow agents can often be trialed within months, but the highest-impact value pools in transportation, defense, or logistics are a different category entirely. Autonomy, robotics, and safety-critical assurance require redesigned operating models, end-to-end data pipelines, validation frameworks, human oversight architecture, and regulatory confidence — which take years to build and cannot be compressed regardless of how capable the underlying models become.

Organizations and regulators need the right ecosystem to deploy this technology, which often requires long and complex transformations. Every year spent not building that ecosystem is a year of irreversible delay; unlike model access, operational maturity has to be grown rather than simply procured when needed. The AI gap is becoming self-reinforcing, and governance is where late movers are most exposed. Competitive advantage in AI is shaped by two distinct sets of constraints: external pressures that every company must juggle — like AI developers or macroeconomic pressures — and internal governance gaps that every company must close.

Transportation lags industry average across all AI deployment categories, with more leaders reporting no AI plans

Share of CEOs who report having no AI plans across business operations

Question: “What stage has your company reached in deploying AI?”
Source: Oliver Wyman Forum x New York Stock Exchange CEO Survey 2026 (N=415)

AI integration hinges on computing, energy, and infrastructure access

Every TLD company, whether it is building AI-native products or deploying AI into existing operations, now depends on a small number of hyperscalers and foundation model providers for compute, infrastructure, and model access.

These providers set the terms, and the unit economics of AI — such as the cost per query of an AI chatbot — must be stress-tested before scaling, since every use case should survive a scenario where compute is more expensive, not less.

Organizations that have not established their cloud and AI procurement architecture will find themselves making consequential commitments under time pressure, with less room to negotiate.

Energy costs and chip availability add pressures that individual companies don’t control. The growing electrification of heavy industry and transportation, in parallel with the infrastructure demands of data centers, is expected to increase global electricity consumption by 3.6% annually through 2030. Pressures like these are causing 57% of global CEOs to cite macroeconomic conditions as the most prominent threat in 2026, with 41% naming geopolitics and trade policy.

Compute and energy availability are geopolitically constrained and must be treated the same way: secured early, diversified where possible, never assumed. Aligning strategies between these two factors will provide structurally better AI economics than treating resources as infinite.

Firms looking to secure access to computing power should look to moves within the industry, where some Korean and Taiwanese semiconductor players are investing in American cities for improved market access.

AI advantage depends on executive and board ownership of technology, risk, and operating model change

Internal governance structures are equally binding and more directly addressable. Board involvement in AI strategy is lower in the transportation sector than the average of all industries, with CFO engagement particularly low. Consider also that boards at all companies reporting a high ROI on AI (more than 10% total cost savings or revenue generation) have increased their involvement in AI strategy at nearly twice the average rate, according to global CEOs surveyed by the Oliver Wyman Forum.

Board involvement matters in both compliance and operations: Comprehensive AI deployment requires capital allocation decisions, risk tolerance frameworks, and accountability structures that cannot be delegated to technology teams. Without early board and CFO strategic engagement, firms will find themselves making consequential commitments — on vendors, architectures, and operating models — without the governance infrastructure to manage them.

The firms now accumulating AI capabilities are restructuring how strategic advantage is built and held in physical-economy industries. This is why scaling AI is best understood as a multiyear industrialization effort rather than a technology project. Moving from pilots into production requires concentrating on a limited set of high-impact opportunities, assigning clear P&L ownership, and treating capital allocation, risk tolerance, workforce redesign, and operating-model change as board and executive committee decisions.

That requires a startup mindset of continuous adaptation of talent models and skills, hiring specialized capabilities, and evolving decision-making processes. Asymmetries between companies and regions in their ability to attract, retain, and nurture top AI and data engineering talent are already shaping long-term competitive positions in ways that are difficult to reverse. Building institutional knowledge among employees takes time, and expertise is not something easily or quickly bought from outside talent at a later time. Executives who think they can plug-and-play AI technology within their organizations are making a similar mistake with talent.

Executives and boards have multiple priorities. First, security, reliability, and trust must be engineered from the outset: Unlike chatbot hallucinations, AI failures in industrial environments could spur production halts, safety incidents, and regulatory exposure. The organizations scaling AI in safety-critical environments have built control architectures, domain-specific guardrails, and organizational disciplines to operate AI within defined risk boundaries before deployment, not after.

Second, the regulatory and liability environment is changing faster than most deployment timelines anticipated, with AI increasingly treated as a regulated product globally whose outputs carry liability. Organizations that build compliance into architecture, documentation, and vendor selection from the start gain deployment speed. Firms that have not solved for these conditions will not scale AI in operational environments regardless of model quality or data foundations. They’ll instead be stuck in a recurring rework loop.

None of this is constrained by the maturity of the technology itself. The relevant AI capabilities — machine learning, simulation, computer vision, optimization, agents, robotics, and autonomy — are already mature enough to reshape physical operations in specific domains.

The more important question is how to integrate them into asset-heavy, safety-critical, and regulated systems where reliability, cost, data access, and human oversight determine whether value scales and which side of a competitive divide a firm will sit in.

That is why the next decade’s growth opportunities will center on interconnected transportation and industrial systems that learn, coordinate, and improve performance across fleets, networks, factories, logistics flows, and sites rather than individual AI technologies. The next section maps where those opportunities will concentrate over the coming decade.

Stacked shipping containers with AI data overlays representing future interconnected AI systems in transportation, logistics, and defense.

Next decade's growth

The next decade’s growth opportunities center on interconnected AI TLD systems

By 2035, AI will reshape transportation by cutting costs and improving how entire systems operate with continuous and proactive adaptation rather than reactive adjustments. Technology will be able to detect mechanical failures before they happen and plan accordingly, or power cobots that determine they’re needed on a different assembly task and proceed without manual reprogramming. Coordinated assets, simulated complex physical environments, and intelligence embedded into field operations will improve throughput and reliability.

The next frontiers are systems that learn, coordinate, and improve performance widely rather than automate in isolation — often in interaction with human operators, and within the constraints of regulation, data availability, and operational complexity. The expert perspectives interwoven throughout this section reinforce this shift, pointing to a future where advantage is defined less by individual technologies and more by how effectively organizations design and operate these intelligent, interconnected systems.

Smarter rail networks deliver more capacity from the same infrastructure

Rail’s most immediate AI opportunity is in optimizing existing infrastructure. Smarter scheduling, real-time delay management, and predictive maintenance can materially increase throughput and punctuality on corridors that have hit the limits of timetable-based control.

Rail networks are implementing some of these capabilities. Singapore, for example, partnered with a tech firm in April 2026 to deploy AI that unifies data distributed across standalone systems and creates a predictive maintenance analysis based on train performance, sensor readings, and asset lifecycle data. Elsewhere, New York’s public transit authority recently partnered with a different tech firm to retrofit smartphone sensors to subway cars to capture vibrations and sound patterns. AI and machine learning algorithms then analyze the sounds and vibrations to generate predictive maintenance insights.

Future vignettes

Rail: The resilient network

AI will make trains faster and more reliable

How passenger rail networks run without AI

Every disruption requires manual re-sequencing across dispatch, station operations, and maintenance.

Signaling faults, rolling-stock issues, and infrastructure defects are coordinated separately across traffic control, station managers, and planners, with decisions split across teams working in parallel.

Momentary delays at critical junctions snowball into hours across networks, reducing throughput and punctuality.

Track capacity, timetables, and maintenance are hard to coordinate in real time.

How AI will optimize passenger rail networks

Predict before it fails: Trackside sensors, interlocking logs, and onboard diagnostics will stream continuously; predictive models will detect early signaling degradation and forecast the impacts on block sections, junctions, and platform occupancy.

Simulate and replan instantly: A corridor-level digital twin will integrate timetables, rolling stock, crew diagrams, platform assignments, and safety headways, running thousands of recovery scenarios in seconds and selecting the optimal plan before disruptions occur.

Coordinate systems: AI agents will push updated timetables directly into dispatching systems, station controls, and passenger information systems in seconds.

Close the maintenance loop: Vision-enabled drones will confirm hardware degradation; maintenance windows will be inserted automatically, with crews, parts, and rolling stock aligned.

10-20%

More effective capacity on existing infrastructure

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Expert spotlight

The next frontier is self-optimizing networks

Andreas Malikopoulos, Professor, Cornell University

The next generation of AI will be defined by smarter systems, not individual agents. Interconnected AI systems that adapt over time will be more reliable and efficient for transportation firms. Yet today, transportation networks, energy systems, and logistics platforms are composed of many interacting units, each operating with limited information and local objectives. The core challenge is that decisions are often made in isolation, so systems remain locally optimized but globally inefficient.

Designing smarter AI systems requires a shift from optimizing components to orchestrating interactions. Each agent, whether a vehicle, robot, asset, infrastructure system, or human operator, must act on what it knows, while its decisions influence and are influenced by others.

Connectivity and information architecture are critical enablers. Better coordination does not require perfect information, full centralization, or full autonomy. It requires structuring how information is summarized, shared, and used. When agents rely on consistent representations of the system’s evolving state, they can make decisions that are locally feasible and globally aligned.

That architecture matters especially in dynamic and safety-critical environments. Systems that learn how to interpret the environment and those that make a decision based on the learning systems must be integrated and leverage the same data, despite being functionally distinct.

Data can continuously refine the system’s state, while decision rules operate on that state in a predictable and robust way. This allows systems to adapt over time without sacrificing reliability, interpretability, or performance guarantees.

AI-driven vehicle coordination across the network can reduce fuel consumption by up to about 45% while improving travel time by over 30%.

Connected and automated vehicles have the potential of optimizing the autonomy of all cars rather than individual autonomy. Vehicles can anticipate interactions, adjust trajectories, smooth traffic flow, coordinate through intersections, and align routing with real-time network conditions. These benefits do not require full automation: Even in mixed traffic, coordinated vehicles can influence surrounding human drivers and improve network performance.

The broader implication is a move toward self- optimizing networks: systems that continuously adapt through coordinated decision-making across many agents. Advantage will come less from improving individual assets and more from designing the decision architecture that allows assets to interact intelligently as part of an integrated whole.

AI enables new aircraft and airspace traffic flows

The winners of this space will be defined by speed and adaptability. Aircraft manufacturers must certify their aircraft the fastest and adapt operations to low-altitude airspace, while infrastructure manufacturers must satisfy regulators and ensure safety despite the increasing amount of traffic and diversity of fleets sharing that airspace.

Early air taxi movers can capture a sector expected to reach $7.6 billion in global revenues by 2035, up from $3.1 billion in 2030, according to an Oliver Wyman Forum analysis.

Future vignettes

Aviation: The scalable low-altitude airspace

The sky that can handle thousands of flights

How low-altitude airspace works without AI

Drones, helicopters, and electric vertical take-off and landing (e-VTOL) aircraft are managed at low density with instrument and visual flight rules; separation is self-managed, especially around non-towered airports.

Managing a high density of simultaneous low-altitude flights requires heavy manual coordination between operators and regulators, similar to the en-route airspace (e.g., sectors, jetways, corridors, capacity constraints).

Operators are required to file flight plans wait for approval, and comply with them (unlike much recreational general aviation).

There is no full framework to manage distinct vehicles (the coexistence of drones, future unmanned e-VTOLs, helicopters, and others).

How AI will revolutionize low-altitude airspace

Situational awareness: Sensor fusion will integrate more sensor data beyond the current existing ADS-B (such as lidar, radar, enhanced ADS-B, voice communications, and ground sensors) into a real-time 3D airspace model tracking vehicles, birds, and obstacles.

Prediction of conflicts: AI trajectory prediction models such as diffusion models will incorporate intent into vehicle motion forecasts over short horizons, forecasting conflicts before they occur.

Deconfliction at scale: Multi-agent coordination algorithms will increase safety and solve deconfliction problems in a decentralized fashion, with each aircraft negotiating space in real time.

Operation within airspace structure: Airspace will be designed to support higher capacity, low-altitude corridors with specific procedures.

Decision support: Onboard intelligence, in particular partial digitalization of communications, will enable increased and human-compatible voice coordination.

Winners will be aircraft manufacturers that are able to leverage AI fast enough to certify and start operations first; in the long term, they’ll automate some operations, including removing the pilot from the cockpit

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Autonomous software hastens automotive safety and performance testing

Automotive safety and performance design have been slowed by testing in physical environments, which are sequential, expensive, and reset with every design change.

AI shifts a significant share of that burden into simulation: World models trained on real fleet data can stress-test braking systems, edge-case scenarios, and control architectures at a breadth and speed that physical testing cannot match, while synthetic data fills the gaps that real-world collection cannot safely cover.

Many automotive R&D executives anticipate AI will disrupt the typical development process by leveraging libraries of products to design and test simultaneously, according to Oliver Wyman analysis.

Future vignettes

Automotive: The virtual development cycle

The car tested before it is built

How automotive development works without AI

Safety-critical systems (braking, advanced driver assistance, battery management) require physical testing across millions of operating conditions and use cases to meet approval standards.

Rare and dangerous edge cases (sensor degradation, extreme weather, unexpected driver behavior) are expensive, impractical, or unsafe to test physically.

Development follows a sequential iterative approach of prototyping and testing that extends over 30 months, with no ability to run design and validation in parallel.

Each design change invalidates prior test results, triggering costly re-testing campaigns.

How AI will revolutionize automotive development

Test before build: World models will train on fleet data and high-fidelity simulations, learning vehicle- environment interactions under real conditions.

Stress-test failure: Scenario generation algorithms will search for failures beyond predefined tests; millions of edge cases can be evaluated continuously.

Generative 3D design: While today’s LLMs generate text, future generative models will be able to generate 3D objects and drastically accelerate computer-aided design.

Closed loop: Models will optimize braking components, control parameters, and system architecture in parallel, feeding back into simulations and creating a closed system where design and validation co-evolve.

Automate certification: AI systems can map simulation results to regulatory requirements, producing traceable evidence packages, reducing the number of physical tests.

Augment with synthetic data: Rare events like sensor failure during emergency braking will be validated via synthetic data generation, expanding safety coverage beyond physical testing limits.

50%

Reduction in testing costs

50%

Reduction in time to market

Source: Expert interviews and Oliver Wyman analysis of publicly available information

AI turns asset-intensive sites into continuously optimized systems

Mining will move from manually coordinated operations to fully orchestrated, autonomous sites. AI can synchronize fleets, equipment, and extraction in real time, maximizing utilization, increasing output, and turning one of the most asset-intensive industries into a continuously optimized system.

Future vignettes

Mines: The self-orchestrating mine site

The mine runs as a real-time AI operating system

How mining operations work without AI

Every machine requires a skilled operator across 24/7 shift rotations; headcount is determined throughout.

Site decisions are made in silos, with geology, production, maintenance, and economics planned and updated independently, too slowly to respond to real-time conditions.

When equipment fails or conditions change, replanning requires hours of coordination across shift supervisors, planners, and dispatchers.

Downtime ripples unchecked across the production chain, with no unified system to absorb and rebalance.

Cost per ton, ore recovery, and throughput are challenging to optimize in real time without a single connected model.

How AI will revolutionize mining operations

Sense: Pit geometry will refresh daily via satellite, drones, and machine sensing, while haul roads and ore positions will update continuously.

Decide: A digital twin will fuse geology, weather, equipment health, ore grades, energy use, and parts supply into a single real-time model; AI will optimize production, maintenance, and economics simultaneously.

Delegate to human: Workforce will shift from operating machines to governing exceptions from remote-control centers. Humans will oversee an integrated system instead of individual assets.

Act: Routes, plans, and machines will re-optimize within minutes, and all metrics will update in real time. Autonomous haul trucks will operate continuously and ore will be routed automatically by grade, hardness, and moisture.

Close the loop: Fault codes will trigger maintenance, parts orders, and dispatch; ore routing will link to contracting and procurement; P&L will update continuously based on real-time data.

~30%

Higher production than manned operations

Source: Expert interviews and Oliver Wyman analysis of publicly available information

AI powers real-time logistics orchestration

Logistics will move from a network that coordinates reactively to one that dispatches, reroutes, and recovers autonomously. AI agents will manage capacity matching, exception handling, and last-mile sequencing in real time, while autonomous trucks eliminate the scheduling constraints that occur when drivers are not available.

Capturing this value at scale today remains the exception, not the rule. The gap between what is possible and what is realized is not driven by ambition or use cases, but by the difficulty of deploying AI in real-world, safety-critical, and legacy-heavy environments.

Future vignettes

Logistics: The self-dispatching freight network

The corridor that moves freight without a dispatcher

How freight networks run without AI

Capacity matching is manual and fragmented. Brokers negotiate loads across thousands of carriers by phone, email, and spreadsheet, with no shared view of available capacity across the network.

Routing and dispatch run on static plans that cannot adapt. When conditions change, delays worsen before coordinators can intervene.

Up to 35% of trucks run empty because shippers and carriers lack real-time visibility into where capacity exists and where demand is moving.

Exception management requires human coordinators to triage, reroute, and communicate disruptions one by one, creating bottlenecks that scale with volume.

How AI will improve freight networks

Dispatch without a dispatcher: Autonomous trucks will operate without drivers on designated highway corridors. AI will assign loads, sequence routes, and manage handoffs between autonomous and human-driven legs.

Predict and preempt: AI agents will monitor shipment status, traffic, weather, and carrier performance to continuously flag delays before they worsen and require recovery actions.

Match capacity in real time: AI-driven brokerage platforms will match available capacity to demand dynamically, considering pricing, selecting carriers, and confirming bookings in seconds rather than hours, eliminating the broker intermediary on standard lanes.

Orchestrate across the chain: Shared network model will coordinate handoffs between shippers, carriers, terminals, and last-mile operators, reducing dwell time at nodes and routing exceptions.

Close the financial loop: Proof of delivery, invoice generation, and payment approval will trigger on confirmed handoff, eliminating manual document processing across the freight lifecycle.

30%

Reduction in cost per shipment

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Expert spotlight

AI shifts value from products to outcomes — and changes who captures it

Michael Sharov, Partner, Oliver Wyman

AI is changing how industrial companies operate and where value sits in the value chain. In the past, much of the sector’s economics was anchored in product sales, parts, labor, and service events. For years, OEMs and service providers have moved toward services and software, managing performance outcomes continuously rather than reacting to failure after the fact.

AI did not initiate that change, but it drastically changes the speed, granularity, and scalability of that shift by making continuous optimization and intervention economically feasible. As a result, fleets, sites, and assets can be monitored, diagnosed, optimized, and reconfigured in near real time — and the economic center of gravity moves away from the machine alone and toward the intelligence layer that improves uptime, throughput, component life, technician effectiveness, energy use, and cost per ton.

That shift expands what can be sold. When AI is integrated in off-road and industrial equipment, the first step is often consulting, advising, and focused monitoring of, say, a machine that is the bottleneck in production, which help the customer improve operations rather than a contract with large fees. Over time, those offers can become embedded analytics, remote support, software-enabled services, and bundled performance offers. Eventually, the most credible leaders will move toward outcome-based models, because once AI is shaping day-to-day decisions, customers will increasingly pay for operating results, not just for equipment and repair events.

This also changes who has power. Control will increasingly accrue to whomever owns the customer workflow and the decision layer: the operating context, data, models, recommendations, and interventions that shape what happens every day in the field. That may be a large software player if OEMs allow it.

At the same time, AI raises the importance of the dealer and field-service layer, but changes what that layer sells. The dealer’s role shifts from reactive service and parts fulfillment toward performance support: diagnostic consulting, remote monitoring, technician enablement, uptime management, and site- level intervention. The winners will be those who translate AI insight into physical action and customer outcomes rather than simply digitizing the old model. Beyond that, AI may enable a further shift: selling both outcomes and AI-managed operating services that run parts of the customer workflow on the customer’s behalf.

AI should be viewed as more than just a productivity tool, with benefits to commercial and ecosystem facets as well. The winners will use AI to lower costs, redesign offers, capture more of the value they create, and strengthen their position in the customer workflow where the next generation of industrial economics will be decided.

Autonomous mining haul trucks with AI data overlays illustrating industrial AI applications in transportation, logistics, and defense available today.

AI applications driving operational impact

Existing AI applications are driving impact

For TLD, the AI applications that matter most for value creation are already sufficiently mature and deployed by leading companies that are winning measurable gains to cost, throughput, reliability, and safety.

Machine learning, optimization, computer vision, and simulation have been delivering measurable impact in real operating environments for years. Recent advances have made AI more visible and more accessible, but they have not fundamentally changed what is possible.

AI is often treated as a single technology — a perception reinforced by the rapid rise of generative AI and large language models. AI is, in practice, a stack of diverse technologies with different roles, maturity levels, and deployment requirements. Reducing AI to chatbot models risks misdiagnosing its relevance, either by overestimating its immediate impact or by underinvesting in the capabilities that drive operations.

These capabilities include systems that optimize operations from data, simulate complex environments, automate decision workflows, enable autonomous operations, and extend physical automation. The same technologies also define the product itself for a growing set of AI-native companies: autonomous vehicle platforms, intelligent aircraft systems, and software-defined industrial equipment where AI capability is the primary source of operational differentiation.

Pattern learning and optimization

These AI systems analyze large volumes of operational data to detect patterns and predict outcomes, and can continuously optimize decisions such as routing, scheduling, maintenance, and resource allocation. This enables organizations to move from reactive to predictive operations that improve reliability, utilization, and cost efficiency in real time.

Machine learning is the core engine; clean, well-structured operational data is the prerequisite that determines whether outputs are trustworthy enough to act on. Deployments range from predictive maintenance in industrial facilities — where one industrial technology leader cut unplanned downtime by 85% and freed 250 operator hours per month — to dynamic scheduling and resource allocation across logistics networks. The technology is mature, and third-party solutions are widely available, making this the most accessible entry point for companies earlier in their AI journey. Companies not yet using it are leaving measurable efficiency gains on the table.

Case study

Industrial AI enhances manufacturing competitiveness

A leading automotive original equipment manufacturer adopted a centralized AI-at-scale approach, combining connected plant data, digital twins, and industrial AI applications to improve operations

Multiple operations streams. AI deployed in manufacturing, logistics, engineering, and purchasing

Investment planning. AI investment coordinated through stream owners and central technology governance to avoid fragmentation

Improved production efficiency. Connected plant data, digital twins, and AI-enabled quality systems

Business challenge

More digitalized OEMs compete with stronger digital foundations in manufacturing processes

Operating in complex legacy environments limits the ability to redesign operations and systems across different plants

Value is difficult to unlock from fragmented industrial data spread across manufacturing, logistics, and engineering data

AI deployment

Purchasing: Support in RFQ writing, supplier- response screening, proposal evaluation, and contract-design activities

Manufacturing (indirect functions only): Application to production planning, layout design, and manufacturing support activities

Engineering: Application to improve design of components and systems

Logistics: Quality digital twin for inspection, simulation, and operations management

Impact and achievements

Paving the way for AI-enabled manufacturing competitiveness

Industrial intelligence layer. Connected factory data creates a real-time operational view of manufacturing, logistics, and quality performance

Digital twin ecosystem. Virtual representation of products, assets, and processes enables simulation and optimization

Reactive to predictive operations. AI systems identify bottlenecks, process deviations, and quality issues beforehand, supporting proactive operational decision-making

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Case study

AI-accelerated engineering transforms systems development

A developer of advanced defense technology across air, land, sea, space, and cyber domains uses generative AI to accelerate systems engineering, software development, and enterprise transformation

Data and AI enablement acts as a catalyst for business growth by creating a common data product

Generative AI used to improve productivity, efficiency, quality, collaboration, and innovation

Model-based enterprise foundation integrates digital threads across the design-buy-build-sustain lifecycle

Business challenge

Engineers rely on manual workflows to develop requirements

Initial requirements are difficult to create when no prior requirements baseline exist

Building a complete and exhaustive requirements set is time-consuming and prone to gaps

Requirements vary across system engineers and still need human review

AI deployment

Requirements generator/navigator: Generates requirements from existing systems of record and internet resources

Agentic requirements framework: Combines process models, requirement procedures, examples, vendor documentation, structured inputs/outputs, and human-in-the-loop review

Lifecycle-wide generative AI: Experiments span concept, requirements, business process flows, functional specs, technical specs, system design, test cases, software development, test automation, and automated deployment

Impact and achievements

67% cost savings potential indicated by four generative AI prototypes developed through a six-week tiger-team effort

70% span reduction indicated versus traditional delivery efforts

7% quality increase through prototype comparison

Sequential Meta Performance Learning (SEMPL) productionized with dedicated resources, product roadmap development, and demand-management focus

Paving the way for AI-accelerated engineering

AI agents across the delivery lifecycle. Roadmap scales AI agent support across requirements and design, development, testing, and deployment, including requirements definition, specs drafting, test-case development, defect detection, test traceability, and deployment automation

Governance-led scale-up. Continued focus on AI governance, transformation culture, and adoption and results tracking to manage risks and sustain value creation

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Virtual simulation and digital twins

AI-powered simulation and digital twins create high-fidelity virtual representations of assets, processes, and environments, allowing companies to test scenarios, validate designs, and optimize operations without physical constraints.

This compresses development cycles and reduces testing costs, while enabling decisions that would be too dangerous or expensive to test in the real world.

Deep learning, machine learning, and computer vision give these environments the fidelity to reflect how real systems behave, but building simulation infrastructure with sufficient accuracy to be operationally trusted requires significant data investment and domain expertise.

The highest-impact deployments are in asset-heavy industries with long development cycles: Automotive companies are cutting testing costs by over 50% and time-to-market by similar margins. Companies that have not yet built simulation capability are absorbing those costs in physical testing and experiencing slower iteration and delayed validation. The next evolution — AI systems that learn physical rules from data and generalize to environments they have never experienced — is the key idea around world models, covered in more depth in the section on frontier AI.

AI agents

AI agents automate complex, multistep workflows by combining reasoning, data access, and system integration. They support or execute decisions across functions such as planning, maintenance, supply chain, and customer operations, reducing coordination effort and increasing execution speed across the enterprise.

Agents must understand language, plan multistep tasks, act within systems, retain context, and operate safely in production. Their capabilities span a spectrum from decision support to full autonomous execution, with each step demanding deeper embedding into the operating model. A digital freight broker has deployed more than 30 agents across its logistics lifecycle, cutting scheduling times by up to 38% and cutting delay durations by nearly 80%. The technology is broadly mature, but production deployment remains uneven. The constraint is process redesign and integration depth, not model capability.

Case study

Agentic AI reduces logistics market fragmentation

A digital freight brokerage platform connects shippers and carriers by leveraging agentic AI to coordinate logistics tasks across a fragmented trucking market

Proprietary data. Models are trained on massive, real-world logistics data from shipments worth billions

Real-world deployment. Dozens of AI agents are performing real business tasks

AI integration. AI is embedded across the entire workflow, not siloed features

Business challenge

Limited real-time visibility into load status and availability makes efficient scheduling difficult

Logistics teams are consumed by manual, administrative work, leaving little capacity for strategic priorities (for example, cost control)

AI deployment

Agentic AI logistics management: Digital agents analyze live logistics information to run end-to- end logistics processes

Scheduling and tracking agent: Agent monitors shipments in real time, handling updates and escalating to human operators when needed

Impact and achievements

Deployed 30+ AI agents to conduct key logistics tasks across the freight lifecycle

Cut scheduling times by up to 38%

33% reduction in costly reschedules

Reduced overdue load statuses by 15% and cut delay durations by nearly 80%

AI agents for freight rate negotiations

Voice-based AI agents. Agents manage inbound and outbound calls, track shipments, handle reschedules, secure capacity, and can negotiate better rates

Voice agents for rate negotiations are cutting driver hold times by up to 98%, but humans are kept in the loop for final verification

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Expert spotlight

The governance challenge shifts when agents stop waiting for instructions

Emilio Frazzoli, Professor, ETH Zurich

Most current deployments treat AI agents as sophisticated execution tools: They receive a task, complete it, and return a result. A human remains in the loop at every meaningful decision point. That model is changing — the next generation of agents will not wait for instructions. In transport and industrial operations, how these agents make decisions and the governance models surrounding them must be addressed before deployment.

These agents will pursue objectives autonomously, interact with external systems, negotiate with other agents, and in some cases develop operating logic optimized for machine- to-machine communication rather than human oversight. A logistics agent that books freight capacity, reroutes shipments, and renegotiates contracts in real time is not a chatbot. It is an actor in the operational environment, making decisions that affect third parties, creating contractual obligations and carrying liability.

The governance challenge this creates is qualitatively different from anything in current AI risk frameworks. The question is whether the boundaries of a task are defined precisely enough to prevent the agent from taking consequential actions its operators did not anticipate — and whether humans retain enough visibility into agent behavior to intervene when those boundaries are crossed.

In safety-critical and regulated environments, the answer today is often no. Agents operating across fragmented IT and OT systems, interacting with supplier platforms, regulatory interfaces, and other automated systems can accumulate decision authority faster than governance structures can track. The risk is gradual opacity rather than a dramatic failure: systems that function, deliver value, and become load-bearing in operations before anyone has established who is accountable for what they decide.

The organizations best positioned to capture the value of autonomous agents are those that treat governance architecture as a precondition for deployment, not an afterthought. That means defining agent permissions explicitly, building mandatory human review points into high- consequence workflows, and maintaining the ability to audit — and override — agent decisions at any point in the chain.

Physical automation and robotics

AI extends into the physical world through robots that can perceive, adapt, and perform tasks in dynamic environments. This allows automation to be flexible and move beyond fixed, repetitive processes, increasing throughput, and safety in areas such as manufacturing and infrastructure operations.

The enabling technology closely mirrors autonomous mobility but shifts the emphasis from navigation to manipulation and physical interaction. Safety verification carries particular importance, as robots operating alongside human workers require certified behavioral guarantees before deployment is permissible. Early movers are already operating autonomous fleets at scale in mining and running multi-robot fulfilment systems across global warehouse networks — with productivity and cost impact that is beginning to redefine what human-only operations can competitively justify.

Expert spotlight

Deployment will unlock data — a key to robotics

Ken Goldberg, Professor, University of California, Berkeley

Many predict a robotics revolution in the near future, but it may be much further away than expected.

The common thesis is that large-scale data will “solve” robotics, just as it did for vision and human speech. But robotics starts from a very different place: far higher-dimensional problems and far less training data. Modern language models are trained on hundreds of millions to billions of hours of data, while robotics datasets are measured in thousands.

Yet data volume alone does not explain performance. While end-to-end models try to learn everything from raw data, hybrid systems in robotics instead combine machine learning with explicit physical models and modular task design. The result is stronger performance with less data.

The real constraint is data quality at scale. Data gathered from simulations or internet videos are imperfect, while remote control of robots is expensive. The best data comes from deploying real robots in real production environments.

This creates the real flywheel for robotics. Engineering and hybrid models make narrow working systems good enough to deploy. That deployment generates production data, which then improves model performance for future deployments. However, it remains essential to set realistic timelines to avoid overpromising and underdelivering.

The winners in robotics will deploy narrow working systems now and use production data to improve them over time, rather than wait for massive datasets. Data is key to win in robotics, but the path runs through engineering discipline rather than data volume alone.

Case study

Autonomous systems orchestration transforms mine operations

A manufacturer of heavy equipment used in construction, mining, and energy successfully scaled AI to real-world deployment in mining operations.

Proven deployment scale. AI systems are fully operational in harsh, real-world environments, including remote mines

Data advantage. Ecosystem comprises 1.6 million connected assets and cloud platform of high-quality data

History of innovation. Early prototypes of this technology date back over 30 years

Business challenge

Skilled operators are required for every machine, limiting 24/7 utilization and increasing cost

Remote, hazardous environments expose workers to safety risks

Equipment downtime and reactive maintenance cascades across the production chain

Decisions are split across separate systems and teams

AI deployment

Autonomous haulage and operations: LiDAR, radar, GPS, and cameras enable autonomous fleets of trucks, dozers, and drills with minimal human intervention

360° digital mine view: Multi-sensor fusion creates a real-time digital view, enabling safe navigation and machine-to-machine coordination

Site-level orchestration: AI covers mine planning, scheduling, simulation, and geological modeling

AI assistant in the cab: Multimodal, agent-like system can diagnose issues, schedule maintenance, and manage critical health alerts

Impact and achievements

~700 autonomous trucks operational globally, one of the first to deliver Level 4 autonomy at scale

11 billion+ tons of material safely moved; 380 million+ kilometers traveled autonomously

3x autonomous truck fleet targeted by 2030 (~2,000+ trucks)

Revolutionizing other industries

Mine-to-quarry progression. Real-world quarry deployment demonstrated that mining autonomy can transfer into smaller, lower-complexity environments

Potential expansion to construction. The next frontier is moving from controlled, predictable mine environments into dynamic, human-heavy construction sites across excavators, loaders, haul trucks, dozers, and compactors

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Expert spotlight

AI scales physics

Samer Madanat, Dean of Engineering, NYU Abu Dhabi; Global Professor of Civil and Urban Engineering, NYU Tandon

AI makes simulations of the physical world usable at scale.

Transport systems have physical constraints. Traffic flow, congestion, and network dynamics follow well-understood principles that purely data-driven models do not capture. The computing speeds of traditional approaches such as microsimulation and dynamic traffic assignments are often too slow for large-scale or real-time use.

On the other side, AI can provide the likely output of any input and act as a fast surrogate for repeated scenario testing. AI can compress a 115-day physics simulation workload into 16 seconds, while retaining more than 95% fidelity to the original model.

Performance remains tied to domain-specific data and engineering capability. Models trained in one environment require calibration in another, though few-shot learning — an AI capability where a model can learn a new task using only a small sample of data — can reduce the amount of new data required when teams understand how to adapt models to new physical and operational contexts.

The bottleneck is not only high-quality data but the ability to structure, interpret, and operationalize it. Organizations that combine proprietary data with the engineering skills needed to integrate physics and AI into a scalable, transferable modeling stack will benefit the most.

Autonomous mobility

AI enables vehicles and mobile assets to perceive their environment and operate with limited or no human intervention. This can increase efficiency, safety, and asset utilization across transport, logistics, and industrial operations — particularly in structured or semi-structured environments where operational conditions are predictable enough to certify.

Safe autonomous operation requires vision, localization, planning, and control to work reliably in real time. The technology is proven in bounded environments: Fully autonomous mining fleets are operational, and robo-taxi services are completing hundreds of thousands of paid rides weekly. Scaling beyond these environments depends less on model capability than on regulation, liability frameworks, and integration into existing operational workflows. The constraints are the conditions under which society and regulators will allow autonomous systems.

Expert spotlight

AI creates value by coordinating humans and machines

Alexandre Bayen, Professor, University of California, Berkeley

In transportation and industrial systems, the challenge is not to automate a single asset but to improve the performance of a larger system of vehicles, infrastructure, operators, and customers. In that context, full autonomy is not the core problem. The harder task is optimizing systems where humans and AI make decisions together.

Mixed autonomy is therefore not a transition phase but the steady state of real-world systems. In most operations, some decisions remain human, whether for safety, service, or because the human remains the ultimate customer and part of the product itself. These models are already in development: In May 2026, Singapore announced collaborations with several AI, mobility, and logistics players to pilot robotics services like cleaning, security, and food and parcel delivery in its Punggol District. These AI systems will complement existing human operations.

The difficulty is not only technical. While AI must handle physical models, scene semantics, human intent, and unpredictable behavior, the real bottleneck is uncoordinated human behavior at scale. Humans act as independent agents, with fragmented information, limited visibility into system objectives, and incentives often misaligned with overall performance.

This shifts AI’s role from optimizing assets to coordinating interactions. It can improve processes such as trajectory optimization, support human work through autopilots, optimize fleets, and outperform humans where coordination must happen comprehensively. Its value lies in shaping how decisions interact across the system, not replacing them.

Performance therefore depends on interaction design. Systems must guide, assist, and improve behavior, not dictate it, and embed AI into workflows, environments, and safety-critical decision contexts with real-time feedback, intuitive interfaces, and aligned incentives.

Progress will be uneven. Structured environments such as freeway driving enable near-term gains; more complex domains, such as electric vertical take-off and landing aircraft, require new models and policy frameworks; human-robot collaboration remains longer term due to less structured environments and higher complexity.

The payoff is nonlinear. Benefits emerge only when enough participants follow system guidance, making compliance and social acceptance central. The frontier is not autonomy alone but coordination capability: orchestrating human and machine behavior so the system performs better as a whole.

Frontier AI

Frontier AI — the most advanced, cutting-edge AI systems today that can perform many different tasks at an advanced level — is not yet institutionalized, but it will shape the next ceiling of AI performance in TLD. The prior applications show where AI already is creating value today. Technologies like world models and quantum computing address physical complexity, data scarcity, combinatorial optimization, and assurance limits.

World models simulate physical environments and predict how systems evolve in situations never before encountered, making them highly relevant for autonomy validation, robotics training, synthetic data generation, and operational digital twins. Their current value is domain-specific rather than general-purpose, but this is already enough to matter greatly in TLD systems, where proprietary physical data can become a strategic moat.

Quantum computing will enable faster computations of AI algorithms — all priority areas in asset-intensive, networked industries. The technology remains early-stage, so the executive action is not broad deployment but targeted readiness: partnerships, talent, cybersecurity, and identification of workloads where traditional methods hit hard limits.

Leaders should treat these technologies as an optionality portfolio. World models warrant targeted experimentation now where physical data and simulation workflows already exist. Quantum computing warrants monitoring, selective partnerships, and risk preparation. The winning posture is disciplined readiness linked to real operational bottlenecks.

TLD leaders need more than one AI strategy

The applications described in this section span a wide maturity range — from predictive maintenance that delivers returns today to autonomous systems still being validated for broad deployment. That spectrum reflects a structural split in how AI creates value in these sectors: a set of workflow and optimization tools that are broadly accessible and can be deployed with relatively low barriers, and a set of mission-critical capabilities — autonomy, robotics, physical control systems — that require years of investment in data infrastructure, safety engineering, and operational integration before they move the P&L.

The two are not in competition, and leading companies are pursuing both. But they require different timelines, governance, and definitions of success. Treating them as a single AI program is one of the most common ways organizations misallocate capital and misread how far behind they are. Understanding the distinction is necessary, but it is not sufficient. Across both curves, the organizations that have moved beyond pilots share one experience: Deployment in these sectors is harder, slower, and more expensive than anticipated, not because the technology is immature but because of the complexity in reliably analyzing data from a broad number of sensors operating in different conditions and across multiple equipment and regulations.

Real-world use cases of world modeling show simulation and data generation capabilities

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Real-world emerging quantum computing use cases show sensor, simulation, and cybersecurity capabilities

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Expert spotlight

AI adoption in transportation is splitting into two curves

Michael Sharov, Partner, Oliver Wyman

Many companies still talk about “AI adoption” as if it were one, monolithic thing. In transportation and industrials, it is not. The sector is splitting into two very different adoption curves.

The first is democratized AI: copilots, code generation, operational agents, engineering assistants, and other workflow tools that are increasingly easy to trial. These systems can create real value across R&D, engineering, sourcing, procurement, service, and back- office operations with relatively low barriers to experimentation.

The hard part is usually not proving the technology works, but giving capable people access, embedding the tools into real workflows, setting guardrails, and learning fast enough to separate value from theater. That is why the early winners in this curve will be those with the fastest learning loops, clearest workflow ownership, and the discipline to scale what works, rather than simply those with the biggest platforms.

The second is mission-critical industrial AI:The second is mission-critical industrial AI: autonomy, robotics, AI-native validation, physical control systems, and real-time orchestration of machines, fleets, and sites. These applications are not simply “harder AI.” They require a different stack and a different commitment: years of investment in sensing, data infrastructure, control systems, hardware integration, safety engineering, validation, and operational know- how. In these domains, value is often proven well before adoption is broad, because deployment is gated by assurance, integration, and operating discipline rather than model access alone.

The two curves are distinct, but they are not isolated. Democratized AI will increasingly accelerate the mission-critical curve by improving coding, simulation, engineering productivity, documentation, and evidence generation. But the convergence should not obscure the difference. A company can be strong in autonomy or site- level optimization and still lag in enterprise gen AI adoption. Another can move quickly on copilots and agents while remaining far from leadership in mission-critical industrial AI.

That is why top performers need two adoption playbooks, not one AI strategy. Democratized AI should be driven by access, experimentation, workflow redesign, and learning velocity — the last one essential in the sense that learning outcomes become an asset in themselves. Mission-critical industrial AI should be governed as a longer-cycle investment in safety, reliability, integration, and operating advantage. Treating them as the same race is one of the easiest ways to misallocate capital and misread competitive position.

Six forthcoming challenges

Source: Expert interviews and Oliver Wyman analysis of publicly available information

Expert spotlight

Controllability, reliability, and trust

Sonia Vanier, Professor, École Polytechnique

Transport systems operate under some of the most stringent regulatory and safety frameworks of any industry, where the consequences of failure are measured in both human and financial terms. The question facing operators today is whether AI systems are controllable enough rather than capable enough.

General-purpose AI models present structural failure risks incompatible with transport operations: uncertain outputs, recommendations that cannot be explained to domain experts or certified by safety authorities, and cybersecurity vulnerabilities that become operational threats when AI connects to control-command and signaling systems.

The decisive shift is from isolated models to networks of collaborating AI agents, grounded in world models and with explicit modeling of the network and its operational rules. That can enable agents to simulate decisions, quantify uncertainty, and coordinate coherently. The domains most relevant include delay prediction, train traffic optimization, maintenance planning, and integrated operational optimization. But this power introduces critical risks: cascading errors through agent pipelines, extended attack surfaces on essential infrastructure, sensitive data exposure, and the challenge of maintaining meaningful human oversight in distributed decision-making systems.

The most robust technical foundation combines operations research, control over AI operations with explicit instructions and check points, deep and reinforced learning, and world models, producing solutions that are simultaneously adaptive, optimal, explainable, and formally verifiable.

Competitive advantage will go to the operators who do three things simultaneously: invest in collaborative research with peers, academia, and cross-industry associations to develop solutions adapted to their processes and constraints, train their teams to integrate AI critically and responsibly, and demand that their systems be robust, explainable, and secure by design.

Public and private sectors can work together to encourage R&D initiatives that align with government AI aspirations in addition to ensuring liability and compliance at the start. Some nations have specific AI research programs driven by dedicated agencies. In the US, for example, the Defense Advanced Research Projects Agency is driving defense-related research while other “moonshot” programs push energy and mobility innovations.

Expert spotlight

Fragmented governance and legal risk across markets

Mauricio Paez and Olivier Haas, Partners, Jones Day

AI has moved beyond a technology race to become a license to operate challenge, where deployment at scale depends as much on governance and regulatory alignment as on model performance. For business leaders, the strategic question is whether AI can be deployed at scale reliably and safely across jurisdictions, not whether AI works.

Legal complexity is not a constraint to be managed after the fact — it is a core design parameter of industrialized AI systems. The current landscape is marked by regulatory fragmentation across major markets.

In the European Union, comprehensive, risk- based regulation is reshaping market access by imposing stringent obligations on high-risk AI systems, including transparency, traceability, conformity assessments, and lifecycle monitoring. At the same time, expanded liability frameworks extend accountability to software and continuously evolving systems, raising the bar for documentation and assurance. The US presents a more decentralized model, where enforcement actions and guidance from authorities like the FTC and NHTSA, as well as litigation risk for deployers, collectively define the regulatory perimeter. This approach creates flexibility but also uncertainty in liability exposure and standards, requiring companies to self-govern while navigating divergent state rules and evolving federal priorities. China, by contrast, combines rapid deployment incentives with state-driven oversight, emphasizing algorithm governance, labeling, and data control. This regime enables speed and experimentation but introduces distinct requirements around content compliance, security reviews, and cross-border data governance.

The result is a structurally fragmented environment in which global companies must design AI systems and operating models capable of functioning across multiple, sometimes conflicting, legal regimes. In this reality, risk management and governance must be embedded from the outset. Industrial AI systems — particularly in safety-critical and engineered asset contexts — carry real-world consequences. Failures can result in operational disruption, regulatory enforcement, or liability exposure. As such, organizations must move from traditional compliance approaches to a more sophisticated governance by design, integrating legal, regulatory, and ethical requirements directly into system architecture, development lifecycles, and operational controls.

As a result, four interrelated domains demand executive attention to enable scalable and compliant AI deployment.

First, data sovereignty has become a first-order design constraint. Organizations must maintain clear visibility over where data resides, how it flows across borders, and which legal regimes apply. This directly shapes infrastructure strategy, including cloud architecture, localization, and interoperability decisions.

Second, third-party risk management is just as central. AI systems increasingly depend on a complex ecosystem of vendors — model providers, data suppliers, and cloud platforms — introducing dependencies that create legal and operational exposure. Effective governance requires robust contractual safeguards, auditability, and continuous oversight across the supply chain.

Third, asset lifecycle management must also evolve. Unlike traditional software, AI systems are dynamic models that degrade over time, retrain, and interact with changing environments and needs. This necessitates end-to-end lifecycle governance, including validation, monitoring, version control, and decommissioning, to ensure safety, reliability, and regulatory compliance over time.

Finally, data governance underpins the entire system. High quality, well managed data is not only essential for performance but also for explainability, auditability, and compliance. It enables the cross-functional coordination — across legal, technical, and operational teams — that increasingly defines value creation in industrial AI.

The bottom line is clear: regulatory complexity should be treated as a strategic input into system design, not as an external constraint. Organizations that align legal strategy, technical architecture, and operational governance early will be better positioned to scale AI safely and efficiently. As the gap widens between technological capability and organizational readiness, governance excellence is emerging as a decisive competitive advantage. Those that wait will fall behind and face increased liability exposure.

Overhead view of a commercial aircraft with AI data overlays symbolizing how CEOs can implement AI in transportation, logistics, and aviation.

The C-suite checklist

C-suite needs a clear checklist to implement AI

Closing the gap between AI potential and real impact is an operational challenge in development, implementation, and adoption. While many companies have access to similar tools, only a few have translated AI into widespread operational and financial outcomes.

Successfully deploying AI goes beyond the IT department to a comprehensive business transformation that requires active ownership from the highest levels of the organization. Without a clear mandate and strong enforcement from the board and executive committee, AI initiatives risk remaining fragmented, underfunded, or limited to isolated pilots.

Leading organizations move beyond isolated use cases and instead align strategy, technology, governance, and the operating model around a small number of high-impact priorities like virtual simulations or physical robotics.

Based on interviews and our advisory panel of 13 members, the following sections outline how these front-runners consistently overcome the barriers that stall most organizations, and how they build the capabilities required to turn AI from experimentation into enterprise-wide value.

The C-suite checklist to implement AI in transportation, logistics, and defense

Source: Oliver Wyman Forum analysis

Strategy and governance

Most companies fragment resources across too many AI initiatives. Instead of concentrating effort on a few transformative bets, they spread budgets and talent across pilots that never reach production or move the P&L. The AI technologies that can reshape operating models are available today. Legacy players that wait for the next phase of AI models or strong ROI use cases may find it harder to catch up to their competitors.

AI transformation is too often delegated to the IT department when it should be owned by the business. Without clear board and executive ownership, organizations default to low-risk experimentation at the edges, while the applications that matter most often require strategic commitment before near-term ROI is visible.

What AI leaders do differently

Elevate AI at the board and C-suite levels; start with a few big bets

Data and IT readiness

AI late movers often do not have the necessary data and IT to support AI in live operations. Their data is difficult to access in real time, poorly organized, and fragmented and disconnected across systems. As a result, promising use cases work on isolated datasets but stall in production. The issue is whether the underlying data can be accessed, connected, governed, and used reliably across the operating environment as it becomes among the most critical assets.

What AI leaders do differently

Approach data as a strategic asset

Workflow redesign

AI that is deployed alongside work rather than embedded into it often causes employees to revert to existing processes or use personal workarounds. AI may generate insight, but decisions and behaviors remain unchanged.

While a machine can be augmented with AI relatively easily, a factory, plant, or network operation is much harder. These are human-enabled systems with established routines, informal workarounds, multiple handoffs, and many failure modes. Unless AI is built into workflows, adoption stalls. Consider that CEOs of AI deployment leading firms — those scaling AI in at least two categories of use cases and receiving notable ROI — are significantly more likely to be redesigning workflows (49%) than the global average (38%), according to Oliver Wyman Forum analysis.

Finally, guardrails against AI’s constraints and risks must be embedded in the operating model, ensuring best practices across safety, cybersecurity, and regulatory compliance.

What AI leaders do differently

Ensure safety, cybersecurity, and regulatory compliance

Change management

Leading companies anticipate which activities will be performed by AI agents, robots, or autonomous equipment — and which human roles and capabilities will become more valuable. They treat workforce transformation as part of the AI initiative from the start. These firms redesign roles, evolve the talent model to integrate human and non-human collaboration, train frontline and managerial staff, and recruit targeted talent for specific business units. Those who don’t engage in the right change management default to resistance, underuse, or superficial adoption. They also risk investor disapproval, with Mercer’s 2026 Global Talent Trends study finding that 97% of investors say their view of a firm lowers when its approach to agile, skills-based models falls short.

Serving employees’ desire for AI skills can help prevent that internal resistance. The desire for training to further develop skills among workers has seen a relative leap of 98% between 2021 and 2025, according to a five-year Oliver Wyman Forum study of how 300,000 workers feel across 20 countries.

What AI leaders do differently

Act not, not when perfect; rethink talent strategy for new human-machine collaboration

Only about a third of employees are confident they are getting the skills they need, but those that do are far more assured and engaged. Their job satisfaction has held steady between 88% to 90% since 2021, while the general population has swung between 68% and 71%.

Leaders also need to navigate a fragmented workforce holding varied views on AI. Consider Gen Z, which has adapted to AI faster than any other generation. Younger workers are 1.7 times likelier to attend AI training, twice as likely to report skill gains, and 2.3 times more likely to say AI improves their output. While frequent use has surged 65% among Gen Z since 2023, 45% of boomers have stopped using AI at work altogether.

Indeed, our research for this report has found that senior workforces are slower in both adoption and acceptance, while those at the manager level need to become accustomed to managing younger workforces armed with a proliferating technology in AI.

Expert spotlight

AI transformation is a workforce strategy

Konstantinos Varsos, Partner, Oliver Wyman

Artificial intelligence is entering corporate life faster than earlier technologies did for most organizations. Executives are under pressure to act quickly, but a central question is often overlooked: How will AI’s value actually be realized through the workforce?

Organizations typically frame AI in one of two ways: as an efficiency lever that reduces headcount or consolidates roles, or as a growth enabler that frees capacity for higher- value work. Both views are valid, but neither fully captures the scale of change underway. For employees, however, the impact is more immediate and uncertain. They are being asked to adopt new tools, adjust long- standing workflows, and take on changing responsibilities, often without clarity on how their roles, pay, or career paths will evolve. That uncertainty is central to AI transformation.

As with any major change, speed can create resistance. This resistance is often quiet, showing up as slower adoption, continued reliance on familiar processes, or workarounds. And because AI systems still have limitations — such as hallucinations, inconsistent outputs, and dependence on data quality — employees have little incentive to fully trust them, often maintaining fallback methods that dilute value at scale. This helps explain why many organizations struggle to capture full returns from AI. Leaders often expect a direct line from deployment to performance gains, but adoption is uneven, behavior changes lag, and organizational complexity slows progress. When multiple roles change at once without coordination, transformation overload can reduce effectiveness rather than improve it.

The answer is to treat AI transformation as a workforce strategy, not just a technology initiative. The focus should shift from tools to roles: which tasks will be automated, which augmented, which elevated, and how work itself will change. That requires deliberate redesign of roles, workflows, and expectations, along with careful sequencing.

Trust is equally important. Employees are more likely to embrace AI when they understand its strengths and limits, see clear benefits in their daily work, and believe the organization is investing in their future. Transparency and reskilling are essential to aligning workforce behavior with business goals.

Ultimately, AI transformation is about redefining how work is done and value is created. Companies that integrate technology, operations, and human capital will be better positioned to capture AI’s potential. Those that do not risk underdelivering — not because the technology fails, but because the organization is not ready to evolve with it.

Impact capture into P&L

Most companies fail to capture the value generated by AI. Each deployment generates data and operational knowledge that improves the next one — but only if financial impact is measured and captured into the P&L.

AI often delivers measurable productivity gains — faster engineering cycles, fewer defects, reduced downtime — but those gains do not automatically translate into financial impact. Freed-up time is absorbed into existing activities, output increases are not monetized, and cost structures are not adjusted to capitalize on the efficiencies created. As a result, the P&L does not move.

What AI leaders do differently

Define how value will be captured upfront

Technology sourcing

TLD leaders must choose the right AI technology for the job. Many organizations overinvest in the most advanced models without aligning technology choices to actual use case requirements. This leads to unnecessary cost, architectural complexity, and limited scalability. At the same time, lengthy and unclear build versus buy decisions create fragmentation, and long-term lock-in risks. And even when technology solutions are fully procured, organizations must still invest in upskilling its workforce so that employees can effectively operate and manage them.

Firms must also monitor the market for new, AI-powered disruptors in the market and adapt quickly. Making the right acquisitions will be critical to compete.

What AI leaders do differently

Monitor the market for AI-native disruptors and consider M&A

Satellite communication dish with AI data overlays representing AI adoption challenges for transportation, logistics, and defense leaders.

The AI challenge

The AI challenge to TLD leaders is to act now or fall behind

AI in transportation, logistics, and defense has moved beyond promise, and already is delivering tangible gains for TLD firms. The technologies are available, the use cases are proven, and the value pools are available. The question now centers on which companies will capture that value and which will fall behind.

The defining challenge is the ability to deploy technology in complex, safety-critical, and legacy-heavy environments. This is where the gap between leaders and the rest of the market is widening. This group is building capabilities now rather than waiting for perfect conditions — embedding AI into their operations, reshaping their operating models, and multiplying the advantage with every deployment cycle.

Waiting is the highest-risk strategy. The organizations that act now — focusing on a few high-impact priorities, building the foundations for scale, and committing to transformation — will define the next generation of performance in TLD.

The implications depend on where a company stands today.

Key actions to deploy AI at scale

Source: Oliver Wyman Forum analysis

Companies that have not yet committed to AI deployment

The organizational readiness gap — data infrastructure, deployment capability, change management capacity — grows every quarter without committed investment, and cannot be closed quickly. The first decisions are about commitment and concentration, not capability building.

  • Choose two or three AI priorities that would materially improve cost, throughput, or product performance. Fund them with dedicated budget, headcount, and a delivery timeline tied to operational results
  • Put AI on the board agenda as a capital allocation decision, with executive committee accountability assigned to the priority applications and named business owners responsible for results
  • Set one live and operational milestone for the next 12 months. This should be a specific process that generates measurable output. A program review or proof-of-concept completion is not sufficient
  • Accept that data does not need to be perfect before deployment begins. For the priority use cases, deploy against the best available data and use live operations to improve labeling, quality, and coverage over time. Data readiness grows through use, not in preparation for it
  • Assess AI and talent gap now. Attracting and developing engineers capable of building and operating AI systems in industrial environments is among the longest readiness gaps to close and is a multiyear commitment that cannot be compressed. Start the assessment and hiring plan before it becomes a bottleneck

Companies scaling from pilots but not yet at enterprise deployment

The risk is proliferation: too many initiatives running in parallel, insufficient depth on any single application, and a growing distance between what AI delivers in controlled test environments and what shows up in the P&L. The trade-off that needs to be forced is breadth versus depth.

  • Assign every active AI initiative a P&L owner and a hard financial target integrated into the budget. Initiatives that cannot meet that target within the cycle get killed or paused
  • Separate the workflow AI from mission-critical, industrial AI. Copilots and planning agents require different investment horizons, validation standards, and definitions of success from robotics and safety-critical systems. Running them as a single program is one of the most common sources of capital misallocation at this stage
  • Build the foundations for mission-critical deployment now. Regulatory confidence, end-to-end data pipelines connecting IT and OT systems, human oversight architecture, and safety validation frameworks require multiyear investment that cannot be compressed by model capability alone. Make worker adoption a delivery condition for scaling. If operators or technicians are working with parallel processes to AI systems, deployment has not succeeded
  • Establish board and CFO-level governance over AI investment proportional to the capital being deployed. The decisions that shape long-term competitive position — operational technology stacks, data ownership, and control architectures — are strategic and irreversible technical choices
  • Secure compute and energy as a strategic input, not an assumed utility. As AI moves from enterprise software into operational environments, the cost and availability of compute becomes a structural cost input. Organizations that have not assessed edge versus cloud requirements for their specific operational use cases, or established their AI procurement architecture, will face higher costs and fewer options

Companies with AI embedded at scale

The risk is optimizing the current generation of deployments while the next value layer forms around them. For equipment manufacturers, operators, and industrial service providers, that next layer is not a technology upgrade. It is a commercial model shift and an ecosystem positioning decision — and both require action before the space is defined by others.

  • Equipment manufacturers and industrial service providers must decide now who controls the decision layer between assets and customers’ operations — the data, models, and interventions shaping daily field decisions on uptime, throughput, energy use, and cost per unit
  • Factor AI-enabled results into the commercial roadmap now. Once AI makes continuous optimization economically feasible at fleet and site level, customers will increasingly pay for operating results rather than equipment and repair events. There will be a progression from condition monitoring to embedded analytics to outcome-based contracts
  • Operators like airlines, rail networks, logistics providers, and mining companies must decide which parts of the operating model become AI-orchestrated and which remain human-led. Those who act decisively will set the terms before vendors or system integrators do
  • Govern autonomous agents as operational actors before their decision authority expands further. Agents that manage freight capacity, adjust maintenance schedules, or optimize production in real time carry risk. These agents need established permissions, human-led audits, and override capabilities before agent capabilities exceed what governance structures can account for
  • Treat simulation capability and world models as proprietary assets. Companies that have proprietary operational sensor data and engineering depth can use simulation and world models to build their advantage. Those that treat it as a technology to monitor rather than an asset to invest in will find that gap increasingly difficult to close

The companies that concentrate on fewer but transformative bets, build foundations before urgency forces the issue, and move toward the next value layer before competitors define it will be in a structurally stronger position in the next technology development cycle. Each deployment generates proprietary operational data that improves models, better models improve operations, and stronger operations accelerate the next deployment. That cumulative dynamic is what makes the gap between early movers and the rest difficult to close, and more difficult with every quarter that passes. The cost of waiting is the accumulating lead of the companies that did not wait.

Aerial view of a passenger train with AI data overlays highlighting AI in rail transportation and logistics infrastructure.

Appendix

Appendix

Technology readiness

AI is often treated as a single technology — a perception reinforced by the rapid rise of generative AI and large language models. In practice, AI is not one capability, but a stack of diverse technologies with different roles, maturity levels, and deployment requirements. Reducing AI to chatbot models risks misdiagnosing its industrial relevance — either by overestimating their immediate impact or by underinvesting in the capabilities that actually drive operations.

The more important point for TLD is that the technologies that matter most for value creation today are already sufficiently mature. Machine learning, optimization, computer vision, and simulation have been delivering measurable impact in real operating environments for years. Recent advances have made AI more visible and more accessible, but they have not fundamentally changed what is possible.

AI architecture and ownership of mainstream technologies available today

Cluster What it does Examples Why it matters
AI Foundations Provides the data, compute, deployment, and control layer needed to run AI reliably at scale Chip architecture, model development platforms, toolchains, quantum computing Determines whether AI stays a pilot or becomes an operating capability
IT ownership
Business
ownership
AI Algorithms Uses data to predict, classify, optimize, and improve decisions Machine learning, natural language processing, graph neural networks Drives much of today’s measurable value in forecasting, optimization, and anomaly detection
Generative AI Creates or transforms text, code, images, and other content, making knowledge work faster and more accessible LLMs, multimodal models, fine-tuning, diffusion models, World models Expands the range of work AI can support across functions
AI Agents Plans steps, uses tools, and helps move work through workflows Function-calling, memory and reasoning, multi-agent coordination Extends AI from answering questions to supporting execution
Physical AI Helps machines sense, decide, and act in the real world Computer vision, sensor fusion, SLAM, motion planning Brings AI into operations, robotics, and autonomy

Source: Oliver Wyman Forum analysis

What matters instead is how it is applied

AI’s full potential to create value unfolds when it is infused with domain knowledge and steered by people who understand the systems it operates in. In TLD, this means engineering expertise, operational constraints, safety requirements, and workflow logic.

Generic models may be widely available, but meaningful performance depends on how well they are adapted to the realities of the operating core.

This is amplified by the structure of the sector: Value is concentrated in mission-critical processes, and deployment must happen in brownfield, fragmented IT/OT environments.

This is what makes AI fundamentally different from traditional enterprise IT. While IT systems support workflows, AI shapes decisions within them. In TLD, that shift brings AI directly into engineering, maintenance, production, and physical operations. The differentiator for firms lies in the quality of that integration into realworld systems and decisions.

As a result, ownership changes. While foundational capabilities remain largely IT-led, the application layer increasingly depends on business, engineering, and operations teams shaping how AI is used. These teams become co-designers of how intelligence is embedded into workflows. At the same time, firms are no longer limited to consuming predefined software solutions. With models, APIs, agent frameworks, and low-code tooling, they can increasingly build domain-specific intelligence layers on top of their own processes.

Leaders do not need to master the full technology stack. They need to understand where sufficiently mature AI can already create operational value and how to apply it within the operating core.

Current TRL levels per AI technology domain (2026)

Source: Oliver Wyman Forum analysis

Technology Readiness Level (TRL)

A nine-point scale providing a standardized measure of a technology’s maturity, from initial concept to demonstrated real-world performance

Source: Oliver Wyman Forum analysis

Methodology

This report was developed by the Oliver Wyman Forum and the University of California, Berkeley, based on in-depth research, interviews, and analysis. We interviewed more than 50 senior executives and experts across the industries covered in this report, including automotive, aviation, rail, manufacturing, transport and logistics, mining, aerospace, and defense. These interviews were used to test where AI is already creating value, where deployment is proving most difficult, and which capabilities distinguish early movers from the rest of the market.

The report draws on the Oliver Wyman Forum and New York Stock Exchange’s global CEO and CFO surveys, which provide a broader quantitative view of executive priorities, perceived threats, and AI adoption levels across all industries, leveraging insights from 415 CEOs and 494 CFOs. Survey findings were used to benchmark transportation, aviation, automotive, industrials, aerospace, and defense against other sectors, and to identify the gap between AI awareness, investment priority, and realized business impact.

The research team complemented the interviews and survey analysis with Oliver Wyman research and case-based market analysis on AI applications, technology readiness, operating model requirements, and deployment barriers in asset-intensive industries. This included assessment of application areas such as pattern learning and optimization, digital twins and simulation, digital agents, robotics, autonomous mobility, and frontier AI technologies.

Together, these sources allowed the team to build both a strategic and operational view of AI in TLD. The methodology was designed to move beyond technology hype and focus on where AI is already delivering measurable value, what prevents companies from scaling it, and how leading organizations are turning AI from isolated pilots into durable enterprise impact.

Acknowledgments

Acknowledgements

Authors

Jean-Pierre Cresci (Partner, Oliver Wyman), Alexandre Bayen (Associate Provost and Professor of Civil Environmental Engineering and Electrical Engineering, University of California, Berkeley), Michael Sharov (Partner, Oliver Wyman), Konstantinos Varsos (Partner and Head of the Digital Operations Practice, Oliver Wyman), Simon DeForni (Principal, Oliver Wyman), Thilo Grunwald-Henrich (Principal, Oliver Wyman), Fabrizio Maria Guidi (Principal, Oliver Wyman), Ludovic Cartigny (Manager, Oliver Wyman Forum).

Contributors

Ana Kreacic, Tom Stalnaker, Thomas Kautzsch, Simon Schnurrer, Tim Garnett, Douglas Carlucci, Bruce Spear, Joris D’Incà, Andrew Medland, Michael F. Wagner, Daniel Kronenwett, Romed Kelp, Nate Savona, Jessica Stansbury, Guillaume De Ranieri, Joseph Walsh, Brian Prentice, Alessandro Tricamo, Radhika Saini, Bashar Fairaq, Pooja Patel, Dan Kleinman, Jilian Mincer, Marissa Lynch, Dustin Irwin, Birgit Andersen, Chiara Azua, Richard Beadle, Georgia Bennett, Jocelyn Cao, Charlotte Fuller, Harry Hepburn, Sujin Lee, Ramona Pillai, Sangeetha V. Supramaniam, Karolina Lewandowska, Nick Liptak, Jens Lischka, Ashley McCoy, Kit Hoe Leong, Therese Palmere, Collin Koenig, Weronika Talaj, Cynthia Perez, Karolina Jaworska, Wai Leong Hoh.

Advisory panel

Olivier Camino (Deputy CEO, Foundever), Liam Cleaver (Partner and Senior Research Director, IBM Institute for Business Value), Daniel Eitler (Chief Data and AI Officer, Mercedes-Benz Group AG), Olivier Flous (Senior Vice President, Engineering and Digital Transformation, Thales), Emilio Frazzoli (Professor, ETH Zurich), Ken Goldberg (Professor, University of California, Berkeley), Stephen Goldsmith (Derek Bok Professor of the Practice of Urban Policy, Harvard Kennedy School); Mohamed Hussein Karmastaji (Chief Executive Officer, Q Mobility), Samer Madanat (Dean of Engineering, NYUAD; Global Professor of Civil and Urban Engineering, NYU Tandon), Andreas Malikopoulos (Professor, Cornell University), Mauricio Paez (Partner, Jones Day), Valeria Sandei (CEO, Almawave; Global Head of AI, Almaviva Group), Sonia Vanier (Professor, École Polytechnique).