EU Tech Sovereignty Package: Cloud, Chips & AI Rules

Samantha: Welcome to The State of Tech, The European Edition, Sunday May thirty-one, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: India's Cyient buys a US AI engineering firm, SoftBank pours seventy-five billion euros into French AI data centres, Brussels unveils its tech sovereignty plan, companies push back on soaring AI bills, then two things you can actually use today, a fresh AI laptop line from Computex and a practical take on specialised AI for your own work. Let's start with that Cyient deal.

India's Cyient buys US AI engineering firm TAO Digital to push into AI-native product development.

Samantha: Cyient, the Indian engineering services group, has signed a definitive agreement to acquire TAO Digital Solutions, a smaller AI-native data and product engineering firm based in Santa Clara, California. The deal puts Cyient deeper into the part of the market where AI is not bolted on, but built into the product from day one.

Bob: And that distinction matters. TAO Digital was set up specifically around what people in the industry call AI-native engineering, meaning the teams design software and data systems with machine learning at the core, instead of dropping an AI layer on top of a legacy product. Cyient is buying that capability, plus a customer base in North America.

Samantha: Financial terms have not been disclosed in detail, but Cyient is positioning this as a strategic move to broaden its offering in data engineering, cloud, and AI-driven product development. The company has been steadily reshaping itself from a classic engineering services provider into something closer to a digital transformation partner, and TAO fits that direction.

Bob: For European listeners this lands in an interesting context. A lot of European industrial groups, think automotive, aerospace, energy, are still leaning heavily on outside engineering partners to modernise their software stack. Indian firms like Cyient, Infosys, TCS, Wipro have been competing for that work for years. Adding an AI-native capability sharpens that pitch.

Samantha: It also signals where the consolidation is heading. The smaller, specialised AI shops, especially those in the US with a tight focus on data engineering, are increasingly being scooped up by larger players who want the talent and the methodology, not just the contracts. We have seen similar moves from European consultancies, but at a slower pace.

Bob: The bottom line is that European industrial companies looking for AI-native engineering partners now have one more big name in the mix, and the competitive pressure on European IT services firms gets a notch tighter. If Capgemini, Atos or Sopra Steria want to keep that high-end AI engineering work in Europe, they will need to make similar moves.

Samantha: And there is a workforce angle too. AI-native engineering teams tend to be small, expensive, and globally mobile. Consolidation under a larger Indian parent often means the work itself stays distributed, with delivery centres in India and client-facing teams in the US and Europe. That model has been a quiet driver of the European IT services market for a decade.

SoftBank commits up to seventy-five billion euros to build AI data centres across France.

Bob: SoftBank has announced an investment of up to seventy-five billion euros, roughly eighty-seven billion dollars, to develop five gigawatts of AI data centre capacity in France. The first phase is forty-five billion euros for three point one gigawatts in the Hauts-de-France region by 2031, with sites in Dunkirk, Bosquel and Bouchain.

Samantha: To put five gigawatts in perspective, that is roughly the output of five large nuclear reactors, dedicated to powering AI computing. This is the largest AI infrastructure commitment SoftBank has made in Europe, and it is being framed as a direct result of personal diplomacy between Masayoshi Son and President Macron.

Bob: France has been positioning itself as the European destination for AI infrastructure for a while now, leaning on its low-carbon nuclear power and on Macron's hands-on approach to tech investment. This deal reinforces that pitch and puts real pressure on other European capitals that have been talking about AI sovereignty without putting equivalent power capacity on the table.

Samantha: It also raises hard questions about the grid. Three gigawatts in one region by 2031 means transmission upgrades, water for cooling, and local planning approvals at a scale that French operators will need to coordinate carefully. Data centres of this size are not built quietly, and the Hauts-de-France region will feel the construction impact for years.

Bob: And there is a sovereignty angle. SoftBank is Japanese, and a lot of the compute capacity built here will likely be rented out to American AI firms running European workloads. So while the infrastructure sits on French soil, the question of who actually controls what runs on it is more complicated than the headline number suggests.

Samantha: Still, for the French government this is a clear win on jobs and on the broader narrative that Europe can attract hyperscale AI investment. Expect Germany, Spain and the Nordics to respond with their own packages in the coming months.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

Brussels prepares a sweeping tech sovereignty package to cut EU reliance on US cloud and Chinese chips.

Bob: The European Union is preparing what is being called a technological sovereignty package, expected to be presented this week. It bundles new rules on chips, cloud services and artificial intelligence, with the explicit aim of reducing dependence on American digital companies and Chinese chip manufacturers.

Samantha: The cloud piece is the politically charged one. European officials have repeatedly raised concerns about what they call potential kill switches, the idea that critical European digital services could be cut off or constrained by foreign providers under political pressure. The package, according to the plans, is meant to address exactly that risk.

Bob: On chips, the focus is on strengthening local manufacturing capacity and reducing exposure to Chinese suppliers in sensitive segments. That builds on the existing European Chips Act, but according to the reporting it goes further on procurement rules and on what counts as strategic supply.

Samantha: For European businesses, particularly in financial services, healthcare and the public sector, the practical question is how strict any cloud sovereignty requirements will be. Stricter rules could push more workloads to European-based cloud providers, but the capacity gap with the US hyperscalers is still very large.

Bob: And there is a tension here. The same week SoftBank announces seventy-five billion euros for AI data centres in France, the Commission wants to reduce reliance on foreign tech. Those two stories are not in direct conflict, but they highlight how complicated sovereignty becomes when the infrastructure money still comes from outside the bloc.

Samantha: The package has not been formally presented yet, so the exact wording matters. Industry groups have already been lobbying hard on the cloud certification rules, and the final text will determine whether this is a meaningful shift or another set of guidelines that gets watered down in negotiation.

Companies hit the brakes on AI spending as agent-based billing sends costs soaring.

Samantha: The initial enthusiasm around AI is meeting a finance department reality check. Companies are reporting sharply rising bills, especially for so-called AI agents, the systems that carry out multi-step tasks instead of just answering a single question. The token-based billing model means an agent can quietly burn through far more compute than a simple chatbot.

Bob: The framing in the industry is that we are leaving the era of, as it has been called, subsidised intelligence, where investors absorbed the real cost of running these models, and entering a phase where AI vendors need to actually make money. That means prices going up, or usage being metered more aggressively.

Samantha: The response from buyers is pragmatic. Many are shifting to cheaper open-source models for tasks that do not need the top-tier systems, or breaking complex jobs into smaller steps so they can route each one to a cheaper model. Specialised AI tools for narrow domains are also gaining traction over the giant general-purpose systems.

Bob: For European companies this is particularly relevant, because budgets here tend to be tighter and procurement cycles longer than in the US. Many CIOs in Europe are still in pilot mode with generative AI, and the cost signals coming out of those pilots will shape what makes it into production over the next year.

Samantha: It also benefits European open-source players. Firms like Mistral and a wider ecosystem of smaller European labs have been arguing for some time that the right answer is not the biggest model, but the right-sized model for the job. The cost squeeze gives that pitch fresh weight.

Bob: The bottom line is that the question for boards is shifting from how do we adopt AI, to how do we adopt AI without the bill spiralling. Expect more talk about routing, caching, and on-premise inference in European earnings calls over the next two quarters.

Samantha: To close out, two things you can actually use today. First, a hardware launch worth knowing about if you are in the market for a new laptop.

Acer launches Aspire AI laptops at Computex, putting on-device AI into mainstream machines.

Bob: Computex 2026 has kicked off in Taipei, and Acer is leading the early announcements with a new Aspire AI lineup. The headline products are the Aspire X 16 AI and the larger Aspire 18 AI, both built around Intel's latest Panther Lake chips, designed specifically to run AI tasks locally on the laptop instead of pushing everything to the cloud.

Samantha: For a regular buyer the practical promise is this. Features like real-time transcription, background blur on video calls, image cleanup, and on-device assistants run faster and work even when you are offline. You also keep your data on the machine, which matters for anyone working with sensitive documents.

Bob: The sixteen-inch model is aimed at general users who want a step up from a standard laptop, and the eighteen-inch model is genuinely large, more of a desktop replacement for people who edit video, work with big spreadsheets, or run local AI tools for coding and writing. Acer has not announced full European pricing and on-sale dates yet, but the line is being positioned for mainstream retail.

Samantha: If you are not in the market right now, the broader takeaway is useful. From later this year, mid-range Windows laptops sold in European shops will increasingly come with AI processing baked in by default. That changes what you can do without a paid cloud subscription.

Bob: A practical tip, when you do go shopping, look for the Copilot Plus PC label, or check whether the laptop has a dedicated neural processing unit. That is the chip component that makes the on-device AI features actually work. Without it, a lot of the marketing claims fall flat.

Samantha: And for current laptop owners, before you replace your machine, check what your current operating system already offers. Windows has been quietly adding free on-device AI features through updates, and a lot of users have not turned them on.

Specialised AI beats giant models for niche work, and you can apply that idea today.

Samantha: Our second tip for today is a practical mindset shift, prompted by an interesting statement from Japan. The country's former Digital Minister, Masaaki Taira, has argued that Japan should not chase the giant general-purpose AI models, but focus on robotics and highly specialised AI trained on proprietary data, in areas like manufacturing, healthcare and nursing care.

Bob: The reasoning is straightforward. The public internet has largely been scraped already, so the next wave of useful AI will come from systems trained on specific, hard-to-access data, factory sensor logs, medical records, engineering documents. That is where smaller, focused models can outperform the giants.

Samantha: How does that translate to something you can try this afternoon. If you are using AI at work, stop reaching for the biggest, most general chatbot for every task. For domain-specific work, there are smaller, often free or low-cost tools tuned for that exact purpose. Coding assistants for developers, transcription tools for journalists, legal summarisation tools for lawyers.

Bob: A concrete example. Instead of pasting a long medical or legal document into a general chatbot, look for a specialised tool, many run locally in your browser, that handles that document type natively. The output tends to be more accurate, the privacy story is better, and the cost is often a fraction.

Samantha: Another practical tip, break big tasks into smaller pieces. Instead of asking one AI to do an entire report, ask a smaller model to extract the data, a second to structure it, and only use a top-tier model for the final polish. That is exactly what cost-conscious companies are starting to do, and it works on a personal scale too.

Bob: The broader signal here is that the AI market is maturing. The interesting tools for the rest of 2026 will not all be bigger chatbots. They will be focused, lighter, often cheaper systems that do one thing very well. Worth keeping an eye on for your own workflow.

Samantha: Today we covered: Cyient buying AI engineering firm TAO Digital, SoftBank's seventy-five billion euro French data centre push, the EU's tech sovereignty package, companies pushing back on AI costs, Acer's new AI laptops at Computex, and a practical case for specialised AI in your own work.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.