Please enable JavaScript to view the site.

Perspective:

Perspective

15 Sep 2025

9 minute read

AI trends 2025: Adoption barriers and updated predictions

Enterprise challenges for agentic AI, physical AI and sovereign AI

From compliance to workforce readiness, explore common organizational barriers to adopting agentic, physical and sovereign AI—and how the future of AI is evolving.

As organizations accelerate their adoption of advanced AI trends in 2025, the path forward is marked by both opportunity and complexity.

Imagine a global logistics provider where autonomous AI agents negotiate delivery routes, dynamically responding to weather-related delays and supply chain bottlenecks. Picture a hospital where robotic assistants collaborate seamlessly with clinicians and wearable health monitors continuously analyze patient data and alert staff to changes. Or consider a multinational bank leveraging sovereign AI practices to ensure sensitive customer data and proprietary models remain securely within national borders, adapting instantly to new regional requirements.

These scenarios aren’t distant possibilities. They are on the near horizon as organizations accelerate their AI adoption. Yet, turning these visions into reality demands more than just enthusiasm for new technology. Organizations must overcome technical limitations, manage operational complexities, address evolving regulatory and compliance requirements, and ensure their teams are ready for change.

This article explores the practical challenges organizations face when adopting agentic, physical, and sovereign AI. The range of challenges illustrates the multifaceted nature of AI adoption and the need for integrated strategies.

Following the methodology used in our previous survey on these same topics, Deloitte invited AI leaders and decision-makers from various industries—as well as a wider LinkedIn audience—to share their perspectives on the most significant barriers facing their organizations.

Here’s a summary of common adoption challenges for these AI trends in 2025, what people are saying and updated AI predictions for 2026.

Agentic AI: From autonomy to alignment

Agentic AI systems hold significant transformative potential with their abilities to adapt to changing environments, make complex decisions, and collaborate with humans and other agents. Yet adoption is still nascent.

What are the common organizational challenges around agentic AI?

Unclear use cases or business value:

Many organizations struggle to move agentic AI from theory to practical return on investment (ROI). Without well-defined applications, leaders risk investing in experiments that don’t scale or demonstrate return, slowing buy-in and funding.

Integration with legacy systems:

Agentic AI thrives in dynamic, connected environments, but many enterprises rely on legacy infrastructure that is often rigid, making it difficult for autonomous AI agents to plug in, adapt and orchestrate processes. Overcoming this requires platform modernization, API-driven integration and process re-engineering.

Governance, risk and compliance concerns:

Organizations are weighing the risks of delegating decision-making to AI at a time when no regulatory frameworks specific to agentic AI exist. Current rules address general AI safety, bias, privacy and explainability, but gaps remain for autonomous systems. Internal governance models, policies and safeguards for human-AI collaboration are critical for responsibly scaling agentic AI while regulation catches up.

Lack of technical expertise:

Successful deployment of agentic systems calls for deep technical capabilities in adaptive learning, agent orchestration, realistic simulation and enterprise integration. AI workforce transformation and talent readiness is becoming a strategic differentiator. Organizations without in-house expertise risk vendor dependence and slower adoption during upskilling phases.

What are people saying about agentic AI?

According to nearly 60% of the AI leaders and representatives surveyed, their organization’s primary challenges in adopting agentic AI are integrating with legacy systems and addressing risk and compliance concerns. These top barriers were followed closely by lack of technical expertise. While the LinkedIn community agreed that risk and compliance is a top challenge for agentic AI adoption, unclear use case/business value was the top answer. Poll participants noted that seemingly endless possibilities unlocked by agentic capabilities may leave organizations unsure where to start, while some organizations may struggle with shifting their mindset to rethink current work. This contrast highlights how AI leaders are perhaps more focused on operational and regulatory hurdles, whereas LinkedIn respondents are considering strategic uncertainty and impact on day-to-day work.

More details

Agentic AI

* Percentages may not total to 100% due to rounding

Physical AI: From labs to the real world

Physical AI—robots, autonomous vehicles, digital twins and smart health devices—represents the convergence of intelligence with the physical world. However, bringing these technologies to life is complex, as organizations must navigate significant technical, operational and regulatory hurdles throughout the implementation process.

What are the common organizational challenges around physical AI?

High investment and operational costs:

Deploying physical AI often requires significant up-front capital for hardware, integration and specialized infrastructure. Beyond deployment, organizations face ongoing maintenance and upgrade costs, making ROI timelines longer and more uncertain compared to software-only AI.

Safety and security concerns:

Unlike digital AI, physical AI interacts directly with people, machines and environments, raising risks around injury, property damage or operational disruption. Companies may establish rigorous safety protocols, cybersecurity safeguards and regulatory compliance checks to protect both users and bystanders.

Integration with infrastructure:

Many industries still operate with legacy machinery, fragmented systems, or outdated IT environments that don’t easily connect with physical AI. This makes integration complex, requiring custom connectors, retrofitting and modernization efforts to ensure interoperability and scalability.

Workforce readiness and skills:

Physical AI adoption requires employees to collaborate with machines and adapt to new workflows, often changing job roles and skill requirements. Success depends on training, change management and fostering trust.

What are people saying about physical AI?

The most significant challenge according to AI leaders, cited by 35% of respondents, is infrastructure integration. Workforce skills and readiness follow closely at 26%. LinkedIn respondents prioritize safety/security (30%) and workforce skill and readiness second (26%)—potentially because they are considering their own interactions with physical AI technologies and the direct impact on their roles and environments. Six percent of LinkedIn respondents identified additional barriers to adopting physical AI, such as aligning with organizational strategy, difficulties in demonstrating ROI, and the perception that organizations lacking expertise in agentic AI may not be “ready” to pursue physical AI use cases.

More details

Physical AI

* Percentages may not total to 100% due to rounding

Sovereign AI: Balancing control and innovation

Sovereign AI ensures that data, models and compute resources remain under controlled boundaries—whether at the national, regional or organizational level. For governments, this means keeping AI infrastructure within borders to comply with regulations and data localization laws. For enterprises, it can mean building organizationally owned, vendor-independent AI that reduces reliance on single providers and maintains control over proprietary data and models. Balancing these considerations with innovation requires navigating complex legal frameworks, managing cross-border operations, mitigating ecosystem dependencies and keeping pace with evolving compliance standards.

What are the common organizational challenges around sovereign AI?

Data residency:

For governments, sovereignty often requires that data is stored and processed within specific national or regional boundaries, complicating operations for organizations with global footprints. For enterprises, sovereignty may also mean retaining ownership of proprietary data pipelines rather than using vendor-hosted environments. Both can demand investment in local data centers, regional cloud providers or private infrastructure.

Model localization and adaptability:

At the national level, models must be adapted to local languages, cultural norms and regulatory expectations. At the organizational level, enterprises face the challenge of fine-tuning vendor-independent models that align with their own governance and risk frameworks. This often requires retraining on proprietary or domain-specific data and continuously updating models to reflect new standards and business priorities.

Infrastructure control:

Maintaining control over AI infrastructure is critical for sovereignty, whether ensuring compute resources remain under local jurisdiction or avoiding over-dependence on a single cloud or model provider. For many organizations accustomed to leveraging hyperscale platforms, building a flexible architecture that allows model switching and multi-cloud deployment is a significant shift.

Regulatory monitoring:

Ongoing monitoring is essential as AI-related laws and frameworks are evolving rapidly and may vary significantly across jurisdictions. For enterprises, sovereignty also means anticipating shifts in vendor terms and industry standards to avoid lock-in. Organizations need dedicated resources to monitor and interpret changes: legal, regulatory or vendor-related. This ongoing effort to track the future of AI as it unfolds adds operational overhead and heightens the risk of non-compliance if updates are overlooked.

What are people saying about sovereign AI?

More than 50% of AI leaders highlighted regulatory monitoring and infrastructure control as the most significant challenges in the realm of sovereign AI. Data residency ranked as the third most significant challenge among respondents from industry events and forums. LinkedIn respondents presented a slightly different perspective. Regulatory monitoring was again the leading challenge, cited by 40% as their organization’s primary concern. Data residency closely followed, with 37% noting it as a significant issue. Infrastructure control, in contrast to the AI leader group, was selected as the least challenging aspect by LinkedIn respondents behind model localization. AI leaders may have a more technical view of challenges, emphasizing infrastructure as a foundation for sovereign AI. LinkedIn respondents may interpret challenges through the lens of immediate operational or compliance pressures, making regulatory monitoring and data residency more critical.

More details

Sovereign AI

* Percentages may not total to 100% due to rounding

Looking ahead: AI trends in 2025 and AI predictions for 2026

In our previous post, we anticipated growing momentum in autonomous systems, physical intelligence and regulatory uncertainty. These trends are now materializing:

Agentic AI is evolving, but governance and business alignment are gating factors

As adoption increases, organizations will embed governance frameworks to manage risks while building new workforce capabilities to monitor, train and guide autonomous AI agents. The leaders will be those that can simultaneously scale adoption, ensure compliance and equip their people to collaborate effectively with AI.

Physical AI is gaining traction and moving to pilots for manufacturing, logistics and agriculture.

Physical AI will no longer be considered experimental. It will be an essential part of frontline operations in industries where safety, scale and human-AI collaboration can deliver measurable economic advantage.

Sovereign AI is still poised to be a top strategic priority.

For governments, it means keeping data, models and compute within borders. For enterprises, it means building vendor-independent AI that safeguards control over data and infrastructure. Competitive advantage will depend less on model performance and more on governance, secure infrastructure and resilience amid shifting regulations and ecosystems.

Is your organization ready for the future of AI?

As AI continues to reshape industries, the organizations that thrive will be those that approach adoption with both ambition and caution. Successfully adopting agentic, physical, and sovereign AI requires more than technological investment. It demands a holistic strategy that addresses integration, governance, compliance, and workforce readiness.

The first step is conducting a candid assessment of your organization’s specific challenges: Where do technical limitations, operational barriers, organizational hurdles, or regulatory and compliance uncertainties pose the greatest challenges for your AI initiatives?

Want more insights?

Deloitte’s AI trends research is just one example of how we remain ahead of the curve in the evolution of AI. To find out the latest research, market trends, and commentary on technologies that are shaping our future, subscribe to the Deloitte AI Institute newsletter and follow the AI Institute for future blog posts in this series.

AI trends 2025: Adoption barriers and updated predictions