Even in the most advanced AI markets, one thing is starting to become clear: adopting AI is relatively easy, but scaling is harder. As organizations move toward agentic systems, the real change is no longer about experimenting, but about making AI an essential part of how the business works. The companies that will succeed will be those that integrate AI into their strategy, infrastructure and decision-making, transforming it from a simple tool into a fundamental driver of growth.
AMD estimates reinforce this change. By 2026, more than 60% of global AI computing capacity will be allocated to inference rather than training. However, despite this strong progress, business infrastructure is struggling to keep pace.
The challenge of scaling AI: the keys that, according to IDC, are defining enterprise AI
A recent IDC whitepaper, commissioned by AMD, shows that adoption is widespread, but scaling is still in development. A striking 83% of companies are currently running fewer than ten AI use cases simultaneously, indicating that many are still in the early phases of their transformation. Just over two in ten (21.7%) conduct full ROI analyses, and 22.2% verify that their investment in AI is aligned with strategic objectives. Strengthening these foundations can unlock faster progress and clearer value narratives.
Added to this gap is a general lack of clarity about the architecture to follow. More than 75% of organizations are not clear about the use cases for agentive AI, which represents an opportunity to establish standards, define control mechanisms and align the different parties involved. Security and governance are priority areas: Companies say they are using access controls, anonymization techniques and policies to manage the entire AI lifecycle. In addition, many carry out adversarial tests and prompt manipulation exercises to reinforce the security and robustness of the models.
Cost management is also maturing: About two-thirds of organizations begin generative AI initiatives with comprehensive cost assessments spanning infrastructure, licensing, professional services, and scalability. Together, these trends point to an ecosystem that is learning quickly and developing the right capabilities to scale AI sustainably.
Redefining computing: why CPU and system architecture are key
As the demands of AI workloads skyrocket, enterprise IT leaders are encountering three critical constraints: cooling and power density constraints, rising cost per token, and governance complexity.
Scaling agentic AI requires understanding that AI infrastructure is fundamentally a challenge that must be addressed at the scale of the entire data center, and that there is no single computing solution that fits all AI loads. While GPUs remain essential for compute-intensive tasks, they are only part of the solution. Agentic workflows rely heavily on agent sandboxes, where logic is executed, vector databases are queried, orchestration frameworks operate, and business applications are interacted with. This change has put high-density server CPUs back at the center of the data center, necessitating servers optimized to deliver maximum core density, low latency, and high memory bandwidth, capable of continuously powering inference loads. The future of enterprise AI will be driven by high-performance CPUs and GPUs working together as part of an integrated system, rather than treating them as independent silos.
At the same time, hardware standardization and efficiency are coming to the fore. The IDC study confirms that the majority (77%) of enterprises prioritize an aligned hardware and software architecture to simplify operations, reduce complexity, and enable both workload portability and performance predictability.
Europe’s special conditions: regulation, sovereignty and sustainability
In markets across Europe, these infrastructure challenges are further amplified by strict regulatory frameworks related to data sovereignty, traceability and environmental sustainability. Far from slowing down progress, these requirements drive organizations to adopt stronger governance models and standardized and replicable processes from earlier stages, promoting greater coherence and reinforcing the trust of customers and other stakeholders.
To address both regulatory demands and financial constraints, business leaders are increasingly turning to open platforms. An open ecosystem that spans architectures, networks and software helps avoid lock-in to a single vendor and gives organizations the flexibility to adapt their infrastructure to specific region and regulatory requirements. Open ecosystems allow the entire industry – from hardware manufacturers to framework developers – to innovate together, and this is how Europe will drive innovation and long-term competitive advantage.
What businesses can do today: Build efficient, scalable and regulatory-compliant AI
AI does not scale solely through experimentation, but requires focus and clear priorities. Organizations that manage to move beyond pilot projects do three things differently:
Open ecosystems allow the entire industry—from hardware manufacturers to framework developers—to innovate together
They link AI to business objectives, associating each initiative with measurable results and applying rigorous return on investment (ROI) criteria.
They treat infrastructure as a strategic decision, building consistent, high-performance environments in the cloud, data center and edge, capable of supporting the new generation of agentic AI systems.
They integrate governance from day one, from data controls to model testing, making it an enabler of trust, speed, and reliable, production-ready AI.
José Manuel Gómez, Business Development Executive, Enterprise Sales, AMD
