On July 9, 2026, MIT Technology Review reported that China is actively finding ways to bypass US export restrictions on Nvidia chips, particularly the H100 and B200 semiconductors essential for training advanced AI models. While the US celebrated a nuclear power milestone, the real geopolitical flashpoint is computing power control.
For Europe, the stakes are double. On one hand, our dependence on US and Asian chips is extending to the entire AI supply chain, where we are already lagging. On the other hand, US sanctions create regulatory gaps that China exploits with agility, while Brussels keeps drafting regulations that fail to move a single euro into European foundries. We see it every day: Italian SMEs wanting to compete with AI can't find affordable hardware, and European cloud hubs are still too small to offer computing capacity comparable to AWS or Azure.
Sponsored Protocol
But the deeper issue is cultural. While the Italian government talks about digital sovereignty, our PNRR funds for semiconductors go to pilot projects that will never reach production. A client recently asked us: “Can I train a custom model on a European cloud?” The answer: yes, but at 3–4 times the cost with half the performance. Translation: our AI startups start at a disadvantage.
Our position is clear: digital sovereignty is not defended by laws alone, but by real manufacturing capacity.
We at Meteora Web have worked for years with small and medium businesses trying to adopt AI tools. And we know that competitiveness rests on three pillars: affordable hardware, connectivity, and skills. Without the first, the other two aren't enough. Europe cannot keep copying US rulebooks while the Chinese play a different game. We need an industrial strategy that funds chip fabs in Europe – not just research centers. And we need it now: in two years the gap will be insurmountable.
Sponsored Protocol
What to do? If you are an Italian entrepreneur or developer, don't wait for your cloud provider to roll out next-gen hardware. Evaluate hybrid solutions: training on European cloud with disaster recovery on on-premise resources. And above all, demand transparency on compute costs. In our projects, we always recommend comparing the cost per training hour across providers, including the new RISC-V open-source chip. If there are no alternatives, we must build them. But politically, the message is one: unless Italy and Europe invest in sovereign fabs and data centers, we will remain paying tenants of American and Chinese Big Tech.
