Skip to content

AI Governance

The European Union just opened bidding to build seven AI 'gigafactories' backed by roughly 10 billion euros in public money. Here is what business owners should know before waiting on government infrastructure to solve a vendor problem.

The European Commission opened bidding on July 30, 2026 for companies to build and operate up to seven AI 'gigafactories,' large-scale data centers packed with at least 75,000 to 100,000 specialized AI chips each. The program is backed by roughly 10 billion euros in public funding from the EU and member states, with the Commission hoping to draw in another 20 billion euros or so from private investors. The stated goal is straightforward: reduce Europe's reliance on American cloud and chip providers for the computing power behind advanced AI models. The direct answer for a business owner: this is a government funding call, not a product launch. Nothing about it changes what cloud AI tools cost or where they run today. Construction is not expected to start before 2027, the facilities are targeted for mid-2028, and a large share of the promised public money is still tied to an EU budget negotiation that has not been settled. The news is worth understanding anyway, because it is a clear public signal of where AI infrastructure spending is heading globally, and because the details of the plan carry a lesson about vendor dependency that applies well beyond Europe.

By Fabio Rabelo · Founder, ATLACIS ·

What happened

On July 30, 2026, the European Commission formally opened a call for tenders inviting companies and investors to bid on building up to seven AI "gigafactories" across the EU. Each facility is meant to house tens of thousands of the most advanced AI chips available: four smaller sites with at least 75,000 chips each, eligible for up to 500 million euros in public funding, and three larger sites with at least 100,000 chips each, eligible for up to 1 billion euros. The public funding pool totals roughly 10 billion euros between the EU and member states, and the Commission expects that to draw in another 20 billion euros or so in private investment. The plan was first proposed by Commission president Ursula von der Leyen at the Paris AI Action Summit in February 2025, explicitly modeled on the success of CERN, the European particle physics lab. Interest has been strong enough that the Commission expanded the program from an original four or five sites to seven, after 76 potential consortia expressed preliminary interest. Construction is expected to begin in early 2027 at the soonest, with facilities targeted to come online around mid-2028. The Commission has already signed letters of intent with Nvidia, AMD, and Qualcomm so that winning bidders can access the chips they need to build these sites.

Why it matters for business owners

Most small and medium business owners will never bid on a gigafactory contract and will not touch this program directly. What it signals matters more than the program itself: governments, not just private companies, are now committing real public money to build the physical infrastructure that AI runs on, because they see dependency on a handful of foreign cloud and chip providers as a genuine strategic risk, not just a cost line. A Commission report cited in the coverage found that the EU's top five cloud providers today are all American companies, and that electricity for data centers in Europe can cost two to three times what it costs in the US or China. That is the same dependency and cost math that shows up at a much smaller scale in an individual business's AI vendor decisions: which cloud a business's AI tools run on, where its data physically sits, and what happens if access, pricing, or terms change with a provider the business does not control.

What owners should not misunderstand

This is not a new AI product, a new cloud region open for business, or a reason to change any AI vendor decision today. The earliest these facilities could be operational is mid-2028, construction has not started, and only a small slice of the promised public funding is actually confirmed right now. The rest depends on the EU's next multi-year budget, which is still being negotiated among member states. Coverage of the announcement also notes the Commission has already scaled back its ambitions once, from an earlier, larger funding figure down to the 10 billion euro public commitment opened for bidding this week. It is also not a story about Europe achieving independence from US AI infrastructure. The Commission's own tender process leans on letters of intent with Nvidia, AMD, and Qualcomm, the same small set of US chipmakers that supply AI infrastructure everywhere else in the world. Building the buildings in Europe does not remove the chips inside them from a handful of US suppliers. That distinction, physical location versus actual supplier dependency, matters more than the sovereignty framing suggests.

The operational lesson

The useful lesson has nothing to do with predicting whether this specific EU program succeeds. It is that "reducing dependency" and "eliminating dependency" are two different goals, and even a government with 10 billion euros and a multi-year runway is only pursuing the first one. Europe wants more of its AI compute to run on European soil, under European rules, which is a real and reasonable goal. It is not trying to, and cannot, remove the underlying dependency on the small number of companies that make the chips everything runs on. The same distinction applies to a business evaluating its own AI vendor setup. Moving a workload to a different cloud region, a different reseller, or even a private, on-premise deployment can reduce certain risks, such as where data physically sits or which company's usage terms apply, without eliminating dependency on the underlying model providers or chip makers further up the chain. Knowing which layer of dependency an option actually changes, and which layer it leaves untouched, is what separates a real risk reduction from a change that only looks like one.

What a serious business should do next

Do not wait on this program, or any government infrastructure project, to solve a data residency, cost, or vendor concern that exists today. If a business already has a real reason to care where its AI workloads run, a client contract that requires data to stay in a specific country, a regulatory requirement, or simply a wish to reduce reliance on one hyperscale provider, that decision should be made against what is actually available now: specific cloud regions, specific private or on-premise deployment options, and specific contract terms, not a facility that will not exist before 2028 at the earliest. For a business with EU data residency requirements or serving EU clients, it is worth watching this program over the next two to three years as an emerging option, without changing anything today. For every business, the more durable move is to map, in plain terms, which parts of an AI setup depend on which vendor and at which layer (the model, the cloud host, the underlying chips), and to know what would actually happen, and what would not change, if any one of those layers had a problem or changed terms.

The Atlacis view

Atlacis is not in the business of predicting EU budget negotiations or handicapping a seven-site infrastructure tender. What is directly useful to a business owner is the pattern underneath the announcement: infrastructure dependency is being taken seriously at the highest levels, and even a government-scale attempt to reduce it stops well short of eliminating it. Atlacis helps owners apply that same honest layering to their own AI setup: mapping where data actually lives, which vendor controls which layer, and where cloud, private, or on-premise deployment genuinely changes the risk instead of just relocating it, so decisions get made on what is real today rather than on infrastructure that is still years from existing.

The short version

  • On July 30, 2026, the European Commission opened bidding for companies to build up to seven publicly subsidized AI 'gigafactories,' backed by roughly 10 billion euros in public funding and an expected 20 billion euros or so in private investment.
  • The goal is to reduce Europe's dependence on American cloud and chip providers, after a Commission report found the EU's top five cloud providers are all US companies.
  • The facilities are not operational today. Construction is not expected to start before 2027, sites are targeted for mid-2028, and most of the funding still depends on an unresolved EU budget negotiation.
  • The program still depends on letters of intent with Nvidia, AMD, and Qualcomm for chips, showing that relocating infrastructure reduces certain risks without removing dependency on a small set of hardware suppliers.
  • The lesson for any business: know which layer of an AI vendor setup (model, cloud host, underlying chips) an option actually changes, and which layer it leaves untouched, before treating a change as a real reduction in dependency.
Tags:AI infrastructureAI governancedata residencyvendor dependencycloud vs on-premisebusiness AIAI decision-makingAI costprivate AIAI buying decisions
FAQ

Common questions

Does the EU's AI gigafactory plan give businesses a new cloud AI option today?
No. The Commission opened bidding on July 30, 2026 for companies to build these facilities. Construction is not expected to start before 2027, and the sites are targeted to come online around mid-2028. There is nothing for a business to use yet.
Will this make Europe independent of US AI infrastructure providers?
Not entirely. The Commission's own tender process relies on letters of intent with Nvidia, AMD, and Qualcomm, all US chipmakers, to supply the hardware inside these facilities. The program aims to reduce dependency on foreign cloud and chip providers, not eliminate it.
Should a business change its AI vendor or cloud setup because of this announcement?
Not based on this announcement alone. If a business already has a real data residency, cost, or vendor-dependency concern, that decision should be made against options available today. This program is worth watching over the next two to three years, not acting on now.
Keep reading

More from the blog

Nvidia may guarantee $250 billion of OpenAI's debt so OpenAI can build a data center neither company could finance alone. Here is what business owners should know about AI vendors whose finances are tangled together.

The Wall Street Journal reported on July 26, 2026 that Nvidia is in talks to guarantee roughly $250 billion in financing for OpenAI, helping it lease a 10-gigawatt data center in Ohio that OpenAI could not easily borrow for on its own credit. Reuters, CNBC, and other outlets independently confirmed the reported figures, though neither company has confirmed the talks. The deal is not an investment and not a compute lease. It is a chip supplier co-signing its biggest customer's debt, a deeper and less visible form of financial entanglement than either of the two vendor-financing stories covered here in the last two weeks.

Cloud AI vs on-premise AI: how to choose without overbuilding

Almost every business should start AI in the cloud. The question is what would make on-premise worth it for you, and whether the middle path covers your real concern at a fraction of the cost.

Meta may lease a competitor $10 billion in computing power. Here is what business owners should know about AI capacity, not just AI price.

Meta Platforms is in early talks to lease Anthropic up to $10 billion in AI computing power over two years, even though Anthropic competes with Meta on AI models. Anthropic is already spending $50 billion on its own data centers and already leases capacity from SpaceX and TeraWulf. The lesson is not about either company's finances. It is that AI compute capacity is scarce enough that even the best-funded labs in the world cannot buy their way past it, and that changes what a business should ask any AI vendor before depending on it.

Make better AI decisions, starting with one call.

Book a free AI Fit Call. We will tell you what to use, what to avoid, and where to start. No jargon, no pressure.