Once upon a time, people in the tech industry said, “Nobody gets fired for buying IBM.” You could experiment and bet on a smaller player. But why would you when you could buy what was widely accepted as the best product on the market? In 2025, that adage applies far more to Nvidia and the semiconductor chips that are set to power AI’s entry into every corner of our lives. According to Steven Schuchart, GlobalData’s enterprise networking principal analyst, “nobody can touch them right now.”
This attitude was clear throughout Nvidia CEO Jensen Huang’s European tour a few weeks ago, during which he met with government and industry figures to pitch Nvidia as Europe’s prime partner in achieving AI sovereignty. The trip was accompanied by major announcements, such as Nvidia’s plans to build an AI cloud platform with Deutsche Telekom, their new partnership with French company Mistral AI involving a data centre that will use 18,000 Blackwell chips and a skills training program with the UK.
Go deeper with GlobalData
Given the current state of the EU’s relationship with US tech firms, wrought with accusations of antitrust practices and eye-watering fines, the hospitality with which European leaders are embracing Huang paints a stark contrast.
“I think there is a difference of maturity [of these companies] as well as their impact on the economic and security future of the EU […] We can all live without Meta. All of us, we can’t live without Nvidia right now,” Schuchart outlines.
Following Huang’s trip, it was clear that his pitch of delivering sovereign AI to Europe resonated with the continent’s leaders. But what would achieving sovereign AI actually look like? And can a US company really deliver European sovereignty?
Wants and needs converge
The idea that Europe’s critical digital infrastructure should not be dependent on US companies has existed for a long time, but these worries have taken on a renewed sense of urgency since US President Donald Trump’s arrival in January. This has particularly been the case for software providers, given that AWS, Microsoft and Google supply around 70% of the European market. Some of these fears have already been substantiated. For example, in May, AP reported that Microsoft suspended the International Criminal Court’s chief prosecutor’s email after Trump issued an executive order against him for investigating Israel for war crimes.
US Tariffs are shifting - will you react or anticipate?
Don’t let policy changes catch you off guard. Stay proactive with real-time data and expert analysis.
By GlobalDataAt the same time, the AI industry’s hockey stick growth is ongoing, and European governments have mostly accepted that AI will form a central part of the continent’s digital infrastructure. And, while they may be wary of repeating past mistakes when it comes to regulating tech firms, the relative infancy of the AI industry means Europe does not have the luxury of choice.
“This is the wild west of AI, and right now you bet on the fastest horse, because there are hardly any other horses. There’s not a lot to do, and it is too important to let go,” Schuchart tells Investment Monitor of the position European leaders are in when choosing what companies to make deals with.
Nvidia is capitalising on that sentiment. But, for all its successes, it is expanding at a time of deep geopolitical realignment and heightened protectionism (particularly in high-tech sectors such as semiconductors). Huang has been a strong critic of US export bans in the semiconductor sector to China, which are mostly bipartisan. He has called these policies a “failure,” adding that Nvidia walked away from $15bn of sales in China due to the export curbs.
However, many have already pointed out the obvious irony of a US company providing European sovereignty, suggesting that the ‘sovereign’ label is a light wrapper for the same tech dependency habit that leaders seek to alleviate.
So, does having local semiconductor infrastructure actually make a difference? Even if it is provided by a foreign company?
Defining AI sovereignty
A common pressure point between tech companies and regulators boils down to supply chains. Where is the data they collect stored? Through what means was it obtained? What jurisdiction do governments have over their citizens’ data if it is stored in another country, as is often the case? Given AI’s long and complicated supply chain, these questions have become even more pertinent.
The same way that regulators might want to ensure that wood does not originate from illicit logging practices, there is increasingly a question about the origins of the data used to build AI models and how this information is deployed. In the UK, there is an ongoing consultation over whether AI developers should be allowed to use copyrighted material, for example.
“If you have a rule that says AI must have been trained in a place that respects copyright laws, then you have a choice. You, as a sovereign entity, can say, well, there are AI systems that comply with that, and therefore they’re the ones that will be used. That’s freedom to act,” Simon Thompson, head of data and AI at GFT technologies, tells Investment Monitor.
Governments have also accepted Huang’s view that AI will become a utility. “It will be considered infrastructure just as electricity, infrastructure, information connectivity,” Huang said when speaking with UK PM Keir Starmer at London Tech Week. It is worth noting that while governments and many businesses have embraced this, there is also broad scepticism about AI’s promises. If, for example, mass job displacement happens faster than people can upskill, there could be considerable backlash.
Whatever the case, if Huang’s vision is realised, then having the local infrastructure to run AI systems also becomes an issue of national security.
“Proximity matters. Running AI workloads across borders adds latency, risk, and regulatory complexity. Sovereignty is essential, not optional. That’s why we’re seeing this expansion [by Nvidia] in Europe happening now. The opportunity is here, and the infrastructure must follow,” Kevin Dallas, CEO of open source data base software service company EDB, tells Investment Monitor via email.
If Europe is obtaining digital services through infrastructure in the US, “your access to service comes through fiber optic cables across the Atlantic, and you’re dependent on the availability of those cables which are owned by other US companies who are beyond your control, or they could be subject to action by the Russians,” Thompson explains.
He also notes that having domestic digital infrastructure means that, in an emergency, the government could take over essential services. If, for whatever reason, there was a “suggestion that these data centres would be just shut down or be taken away, we could pass laws to say, no that’s not going to happen,” Thompson notes, comparing it with what the UK government’s takeover of British Steel.
If AI becomes fully integrated in the UK’s NHS, as is the government’s plan, it will become even more critical to have local infrastructure to support it. Therefore, even if Nvidia is a US company, having more of its infrastructure located in Europe could dilute some security risks and facilitate regulatory jurisdiction.
Other US companies are also trying to address data sovereignty concerns by setting up different legal entities.
“Amazon and Google are planning on setting up these separate legal instances so they can claim sovereignty, so that they are exclusively under the control of the EU and the host country they’re in. As opposed to, being subject to the US government or US headquarters. The idea is to reassure European governments. I still think that a homegrown solution will be more popular,” Schuchart says.
However, given the US political landscape and the clear alignment of tech leaders with Trump, it is worth asking if a company that is at the whim of an unpredictable president can truly provide sovereignty guarantees to Europe.
Are fast and expensive chips always better?
Nvidia’s chips are not only the most advanced, but they are also some of the most energy-intensive and expensive. The 18,000 Blackwell chips that will be used by Mistral AI are estimated to cost $30,000 to $40,000 per unit. However, earlier this year, the appearance of the Chinese company DeepSeek and their chatbot challenged the assumptions underpinning these investments.
DeepSeek claimed to have made a product extremely similar to OpenAI’s ChatGPT for a tiny fraction of the cost. Nvidia lost 17% of its market value, representing the biggest market cap loss for a single stock ever ($589bn). However, this scare was relatively short-lived, and Nvidia stock has since recovered.
“Almost every disruption in the business, with the exception of genuinely new technologies, is more evolutionary than revolutionary, and it’s always about knocking down barriers. The energy costs for AI data centres are outrageous. They’re talking about putting so much computing power in a single rack that we’re not even sure how they’re going to cool it,” Schuchart explains.
“If you wanted to make a conservative bet about what will happen, it’s not going to be a breakthrough that breaks Nvidia. What there is going to be is a general ramp down of power costs as we move forward because there’ll be improvements in every area, and that will shrink the power envelope long term,” he adds.
This progress could come in the form of software optimisation, which is part of how DeepSeek managed to build its model. Anton Lokhmotov, founder and CEO of optimised AI solutions company KRAI, explains that, when trying to optimise how machines work, “the software part of the story is often forgotten” as companies would rather buy more hardware. “The Chinese, they did not forget.”
“It is easier for [companies] to buy that hardware than to invest in optimisation. To me, it is very painful to realise how inefficient it is […] It is kind of this principle, don’t waste time optimising. Things will be fast if you don’t optimise it, just move faster, and buy more, spend more,” Lokhmotov explains.
Given that the industry is still in an early growth stage, and that vast investment plans have been made around certain market conditions and price projections, optimisation is not a priority. This is despite the reality that the AI sector’s rapid growth is coming at massive environmental costs, such as through the continued use of non-renewable energy and exacerbating water scarcity. These are factors that contribute to a climate crisis that is already creating a more unstable global landscape.
“Right now, enterprises are chasing outcomes. The perceived upside is enormous. Maybe DeepSeek raised the right questions, but the market isn’t optimising for efficiency yet,” says Kevin Dallas, who is also Microsoft’s former corporate vice president for cloud & AI big data.
If we’ve learned anything from the tech giants of the past and their leaders, it’s that for better or for worse, they tend to transform. In less than two decades, wonderkid Mark Zuckerberg became the punchline of a joke about lizard people. Tesla is now one of many providers of electric vehicles. This is not to say that Jensen Huang and Nvidia face the same fate, but rather, it should serve as a reminder to European leaders that no one company, or person, has all the best answers amidst this time of AI revolution. Even those with cool leather jackets.
