European AI-focused startups raised $23 billion in the first half of 2026, up 130% year over year — 55% of all venture funding in the region, according to a joint Crunchbase and HumanX report presented at last month's HumanX conference in Amsterdam. Yet two executives who build AI for a living argue the capital figure answers the wrong question. Governments and large enterprises also have to buy what those startups build, and in Europe that purchasing culture is weak.

The report anchors a sharpening debate in sovereign AI: where should countries concentrate resources to capture the technology's economic value and keep control of their data? Fabrizio Del Maffeo, founder and CEO of chipmaker Axelera AI, and Mehdi Ghissassi, chief product and technology officer at Abu Dhabi-based AI71, offered a framework on stage at HumanX, built around Nvidia CEO Jensen Huang's "five-layer cake" of AI: energy, chips, infrastructure, models and applications.

Their shared argument: sovereignty does not require owning every layer.

The chip layer. Del Maffeo's five-year-old company develops inference chips and has sold two generations to roughly 600 customers. His thesis is that AI processing shifts from centralized data centers to devices — a transition that demands new silicon.

"Artificial intelligence will expand from cloud computing, from centralized data centers, to devices closer to us in the physical world," he said. "To enable this, you need specific chips which can run efficiently, at a lower cost, to connect these networks that today are running in the cloud. We are here to solve this problem."

Axelera AI's next step is products for decentralized cloud computing.

The application layer. Ghissassi, who previously led product development at Google DeepMind, dismissed the model layer as a strategic target. Given the capital required and how quickly models commoditize, only "two, three or four companies" can afford to compete there, he said. Applications are where sovereignty actually lives.

"You want to make sure that it [your data] stays with you, be it that you're a government or an enterprise," Ghissassi said. "If you're giving away your trade secrets and know-how, nobody stops whoever is being a provider to you today, from replacing you."

AI71's home market offers a working model of demand-side sovereignty, by his account. The UAE has the world's largest compute per capita, cheap and abundant energy, and a three-month mandate for every government agency to set up agentic processes for citizens. Over the next two years, the stated goal is for AI agents to handle half of every citizen's government interactions. Mandates, compute access — sovereign cloud, on-prem or global — and capital all accelerate adoption.

Europe's constraints are the mirror image. The region is a net importer of energy. It has semiconductor strengths but lacks the compute capacity for AI, neural networks and frontier labs, Ghissassi said.

Del Maffeo put the missing piece bluntly: Europe's large businesses rarely buy from startups, so the region never develops the procurement flywheel that powers U.S. growth. "What worries me is that we are a little bit lagging behind, and therefore we are missing this value creation, and this will weaken the economies of Europe," he said.

Europe does hold real assets — large pools of research talent and a population of 440 million who could benefit from AI. But assets without buyers do not compound.

"We should not be obsessed with controlling the entire stack," Del Maffeo said. In Europe, he argued, the task is creating value rather than paying for services. "Creating value means creating a wealthy economy."

The measured facts — $23 billion raised, 130% growth, 55% of regional venture funding, 600 chip customers — describe supply. What remains unmeasured, and asserted, is whether European procurement will follow. With Axelera AI expanding into decentralized cloud and the UAE targeting agent-handled government interactions within two years, both companies are betting that demand, not capital, is the variable that decides who captures AI's economic value.