As Europe races to strengthen its position in artificial intelligence (AI), much of the political debate has centred on semiconductor manufacturing. Policymakers have poured billions of Euros into efforts to expand local chip production capacity, reduce dependence on Asia and secure access to advanced computing.
But according to Mark Papermaster, chief technology officer (CTO) of AMD, Europe’s biggest opportunity in AI may lie elsewhere. He argues that the definition of AI leadership has shifted rapidly over the past year, moving beyond simple control of GPUs toward the broader challenge of deploying AI across enterprise systems. With the recent publication of the European Technological Sovereignty Package, and in particular the draft Chips Act 2.0 and the Cloud and AI Development Act, Europe is recognising this trend.
That shift has important implications for Europe. While the region lags the US and parts of Asia in hyperscale cloud infrastructure and leading-edge semiconductor manufacturing, it remains strong in industrial systems, automotive electronics, enterprise computing, research consortia and open technology ecosystems.
As AI becomes increasingly integrated into enterprise workflows, Papermaster believes interoperability and flexibility may matter more than vertically integrated technology stacks.
The next phase of AI
For the past several years, AI infrastructure strategy has largely been driven by the race to train increasingly large models. This created enormous demand for graphics processing units (GPUs), advanced fabrications and high-performance compute clusters. But the industry may now be entering a different phase.
“What we’re really seeing in the last several months is that AI applications are largely becoming agentic processes,” Papermaster says.
Rather than simply generating text or images, these systems orchestrate sequences of tasks across enterprise applications, databases and business processes. “You’re running AI with reasoning capability,” Papermaster says. “But what are you using it for? You’re using it to actually execute a whole vast process that used to be human driven.”
That evolution changes the nature of AI datacentre infrastructure itself. Instead of relying on isolated GPU clusters, enterprises increasingly require heterogeneous computing environments that need an increasing number of CPUs to be combined with GPUs for AI inference, databases and storage.
For CIOs, that means AI strategy is increasingly an integration challenge rather than simply a hardware procurement exercise. That includes decisions around data governance, interoperability, security and workload placement – areas where European deployments must generally comply with stricter regulation than those in the US.
Papermaster sees those principles not simply as technology preferences, but as strategic assets in an AI era increasingly shaped by concerns over sovereignty, resilience and dependence on a small number of providers. In his view, Europe’s long-standing support for open ecosystems may prove to be an advantage rather than a constraint.
“If Europe stays committed to diversity of supply by avoiding supplier lock-in through the use of open ecosystems and open standards, it will be poised very well,” Papermaster says.
His comments align with broader European efforts around digital sovereignty and reducing dependence on single technology providers. Rather than building vertically integrated AI stacks controlled by a handful of companies, Europe has often promoted federated and interoperable infrastructure models – an approach that fits naturally with the direction AI infrastructure is taking.
Advanced packaging becomes strategic
Papermaster does not dismiss semiconductor manufacturing as unimportant. But he argues that the current geopolitical debate often overemphasises fabs while underestimating the complexity of the wider AI supply chain. “No one has the complete supply chain within their borders,” he says.
Building a fully self-sufficient semiconductor ecosystem would require not only leading-edge fabs, but also packaging, assembly, testing, chemicals, software frameworks, storage systems and AI models. Instead of pursuing complete self-sufficiency, Papermaster advocates resilience through diversification.
“What people should be doing now is stepping back and understanding which components of the supply chain are important enough to have direct control over,” he says.
This includes reducing dependence on single suppliers and ensuring geographic diversity across the supply chain. These ideas increasingly resonate in Europe, where concerns over geopolitical risk, trade tensions and supply disruptions have intensified since the pandemic and the global semiconductor shortages that followed.
“What people should be doing now is stepping back and understanding which components of the supply chain are important enough to have direct control over”
Mark Papermaster, AMD
One area Papermaster believes deserves far more attention is advanced packaging. As AI systems become more heterogeneous, modern processors increasingly combine multiple specialised components rather than relying on a single monolithic chip.
AMD has been a major proponent of chiplet architectures, which divide processors into smaller functional blocks that can be combined in different configurations. That approach allows some parts of a system to use leading-edge process nodes while others rely on older, more mature manufacturing technologies. “It allows tremendous flexibility of choice,” Papermaster says.
But these architectures also make packaging technologies increasingly important. “We used to think of the supply chain for semiconductors and therefore AI only as having that fab to create those leading-edge chips,” Papermaster says. “But because of the trend of heterogeneity, needing these multiple different compute elements, it’s driving advanced packaging as an equally strategic investment.”
This may create another opportunity for Europe.
Security and sovereign AI
Papermaster argues that sovereignty increasingly depends on protecting models and data, rather than physically controlling every manufacturing layer.
AMD has been investing in confidential AI technologies that allow models and workloads to remain encrypted while running inside shared computing environments, he says. “You as the user control the cryptographic keys.”
That capability may become increasingly important for European enterprises operating under strict regulatory requirements around data protection, sector-specific compliance and AI governance. It also reflects a broader shift in how organisations are thinking about sovereign AI.
Rather than requiring every component to be physically local, some enterprises and governments are increasingly focused on ensuring secure access, protected data environments and operational control over models and workflows.
For European policymakers, Papermaster’s message is ultimately about broadening the definition of AI leadership. “What I’d recommend to governments is to look beyond the fab,” he says.
This means thinking not only about semiconductor manufacturing, but also about packaging, integration, open software ecosystems, security, enterprise deployment and resilient supply chains.
For CIOs, the same logic applies. The challenge is no longer simply gaining access to GPUs. It’s integrating AI into complex operational environments while maintaining flexibility, security and control. Both the draft Chips Act 2.0 as well as the proposal for a Cloud and AI Development Act are taking steps in that direction.
As AI moves deeper into enterprise infrastructure, Europe’s long-standing strengths in systems engineering, industrial computing and open ecosystems may prove more valuable than many expected.
