These datacenters planned by SAP, Telekom etc. appear to focus on AI datacenters, not general digital sovereignty. I think LLM training is much, much less important than getting governments, companies and individuals away from american tech giants.
Also, who will get most of the money? NVIDIA, an american company. Buying 500k american GPUs for billions isn’t the grandiose step towards sovereignty they make it out to be.
Why not invest the money in EU chip design? Why not EU hardware design and manufacturing? Why not industrial automation and robotics? Installing renewables, improving infrastructure? There are other challenges which are more likely to be useful than power-hungry slop machines. I have yet to see LLMs being used for non-trivial tasks reliably. The most frequent use for LLMs I have seen is to relieve people from thinking critically or putting in the effort to learn skills properly. Which are both extremely bad and dangerous things that will actually reduce productivity and quality in the long term.
I think maybe both objectives can be saught after simultaneously. Lidl’s parent company is starting to sell business cloud services in Europe. The private market, if protected, can fill the gap for sovereign European cloud services. Not sure about the industrial part of the equation though.
Yeah, I’m not against AI research fundamentally, I just think they are (i) investing in the wrong kind of AI research and (ii) investing more in AI than desirable given the numerous other challenges which are under-funded in comparison. I’m going to argue this in a bit more detail:
(ii)
According to this article referenced by OP’s article the EU is going to spend about 35% of 200 billion, so 70 billion, for AI. Imagine they’d invest just 1 percent of that into FOSS. The German sovereign tech fund, which is hugely successful, has a budget of around 20 million euro per year. That’s literally nothing for a government. With 1% of the 70 billion, the EU could invest 700 million euro in digital sovereignty, which is more than SUSE’s 2022 revenue and more than double the Linux Foundation’s 2024 budget (around 290 million USD). Additionally, leading robotics company KUKA (meanwhile bought by a chinese firm) had just 4.4 billion USD revenue in 2022. All of the above ventures would provide better returns in terms of (a) improving labor productivity and (b) sovereignty versus USA’s big tech and Chinese manufacturing. These 500k NVIDIA GPUs will become outdated within a few years, meanwhile it’s still unclear if and how they will facilitate any benefit at all to EU citizens.
(i)
Now, IMO there are also decent ways to invest in AI: Invest in academic research and invest in under-developed areas of machine learning. Not every machine learning problem is an LLM problem. Today’s transformer-based LLMs are only a small subset of the thinkable architectures of neural networks, let alone machine learning in general. Robotics (e.g. controlling robot arms in a more dynamic manner), medicine (e.g. protein folding), infrastructure (e.g. predicting failures, optimized planning) or economics (e.g. coordinating production processes to optimize cost/emissions/latency/fault tolerance) are all very concrete, high-value targets for research funding. Regulations like the Corporate Sustainability Due Diligence Directive are currently very expensive for small to medium size companies due to the bureaucratic overhead, but a EU-driven standard for digital and automated tracking of supply chain metrics could reduce the running cost to near-zero if implemented well. Companies like SAP (34 billion USD revenue in 2024), which is a leading company in enterprise resource planning, have the potential to co-develop and implement such standards to provide affordable yet precise enforcement of regulations. Also, if the EU defines such a standard, it will likely also be adopted by global suppliers because they need it to sell to EU customers. This would of in turn give EU software and cloud companies a huge advantage globally.
To summarize, both better AI and non-AI investments compared to the EU’s planned AI/LLM training datacenters.
Sorry for the long post, but I feel it is neccessary to properly make my point. Am I making any sense? 🙃
Well I’m not very proficient in AI so I’ll have to trust you on this one.
The governments in Germany and France want to speed up the construction of clouds in the EU. A European Darpa for war research is being set up.
[…] A new Franco-German agency for innovations is planned for military research - based on the model of the US Defense Advanced Research Projects Agency (Darpa).
In the trade conflict with the Trump administration, Germany and France are focusing on unity - an agreement with the USA should remain the declared goal. But the paper goes further: for the first time, the possibility of using “monetary and austerity policy levers” against Trump is brought into play.Aaw great, now Europe wants to have its own implementation of this scourge. Like American Big Data wasn’t nefarious enough.
I prefer one with European rules concerning data over the American ones.
More choice = better. On the other hand, Germany and France are both getting uncomfortably close to having neo-Nazi governments, with AfD and RN respectively being both very pro-Russia, anti-EU and aligned with Trump policies (even though Le Pen can’t say it too loudly because of her party’s historical anti-Americanism).
I hope these efforts to create a EU infrastructure succeed, but on the other hand big investments on AI:
-
take time that I’m not sure we have
-
sound exactly like the technocratic solutions “far from the real problems of the people” that these parties love to exploit to gain votes and get to power
Can you elaborate on the connection between “national” cloud initiatives and Nazi/Putin fans? I don’t get it.
-
Try cloud seeding…
