Europe is currently gambling away its AI future

A bombshell on Wednesday in AI Europe. Silo AI from Finland announced that it was selling to the US chip manufacturer AMD for $665 million. The team around founder and CEO Peter Sarlin describes itself as the largest private AI laboratory in Europe, has 300 AI researchers and developers on the team and has also launched open source models for Finnish, Danish, Icelandic, Norwegian and Swedish with “Viking” and “Poro”. All of this will in future belong to Advanced Micro Devices (AMD) from Santa Clara in California.

Newsrooms.AI launches software packages for corporates, media & startups

From AMD’s point of view, the deal is understandable – and comparatively cheap. The Americans pay around $2.2 million per employee with AI expertise. In Silicon Valley, annual salaries for AI developers often amount to high six- or seven-figure sums, so the acquisition of the team from Finland alone would be a good deal. Compared to the best-known and most valuable European startup, Mistral AI from Paris, Silo AI is a bargain. Mistral AI was last valued at €5.8 billion. With, as reported, around 55 employees, this results in a value per employee of €105 million.

With a market value of $300 billion, AMD is a giant from a European perspective, but a dwarf from a US perspective. Nvidia, AMD’s huge competitor in the area of ​​GPUs, which are so essential for AI training, is more than ten times larger than AMD with a market value of $3.3 trillion. And not only does it have the coveted H100 chips (and B100 chips in the future), it also ties its customers (Microsoft, Apple, Google Deepmind, Meta, Amazon, Oracle, OpenAI, Dell, etc.) with the CUDA programming language, an industry standard. This means that if you want to use CUDA, you have to use Nvidia chips. This software-hardware combination has made Nvidia an indispensable building block of Generative AI.

AI investments and the 500 (!) billion dollar gap

Silo AI becomes an extended AI arm of AMD

With Silo AI, AMD wants to better position its own GPUs, which compete with those from Nividia, in the enterprise market, and the Finnish company’s SiloGen platform and “Silo Operating System” will play a central role. The 300 AI specialists will eventually become part of the AMD Artificial Intelligence Group if the deal goes through in the second half of 2024.

The deal, which is certainly being celebrated, also means that one of the three or four most important AI startups in Europe will soon belong to a US chip company. In the future, Silo AI will also be asking the question: what about the open source strategy, what about the focus on European languages, what about the close collaboration on LUMI, Europe’s fastest supercomputer (which, by the way, runs on 12,000 AMD GPUs)? The press release speaks of “global customers” – and they can also be found in the USA and Asia. Europe is therefore no longer a focus for Silo AI.

Amazon snaps CEO and all co-founders of AI startup Adept AI

Mistral AI is mainly funded by US investors

What about the other AI startups? Mistral AI should really have been part of the US VC and enterprise world for a long time. The 5.8 billion euro valuation and the more than one billion euros that have been invested do not come largely from Europe, but from:

• Microsoft (US)
• Databricks (US)
• NVIDIA (USA)
• Salesforce (US)
• DST Global (UK)
• General Catalyst (USA)
• Lightspeed Venture Partners (US)
• Andreessen Horowitz (USA)
• Motier Ventures (FRA)
• The Family (GER)
• Headline (US)
• Exor Ventures (USA)
• Sofina (BEL)
• firstminutecapital (UK)
• Bpifrance (FRA)

The bottom line is that without US investors, Mistral AI would not be where it is today. The know-how also comes from US companies: The founders Arthur Mensch (CEO), Guillaume Lample (Chief Scientist) and Timothée Lacroix (CTO) are former employees at Google and Meta, where they worked on AI models (Transformer, Llama).

Digital Fortress Europe: Fortress of Regulations

Aleph Alpha falls far behind the US market leaders

That leaves number three in the AI ​​race: Aleph Alpha from Heidelberg in Germany. While Mistral AI published a total of 7 different AI models last year and raised more than a billion euros, the press is currently concerned with the question of whether and how much money Aleph Alpha really received and how good or bad the German company’s AI models really are compared to OpenAI, Anthropic, Google or Mistral AI (Hin: Unfortunately rather bad).

At the end of 2023, Aleph Alpha was still praising the fact that the company founded by Jonas Andrulis had at least managed to get financing mainly from European investors – i.e. Innovation Park Artificial Intelligence in Heilbronn (Ipai), Bosch Ventures, the Schwarz Group of the Lidl founder, the Berlin-based Christ&Company Consulting, Hewlett Packard Enterprise, SAP and Burda Principal Investments. In this way, a European AI alternative could really be built, while Mistral AI would become more of an extension of Silicon Valley.

But now people are worried about Aleph Alpha and are eagerly awaiting the imminent release of new AI models that can compete with the Americans. Time is of the essence in the AI ​​race: small and large companies in many places are currently making decisions about which AI technologies they want to build on in the future – and if Aleph Alpha is not an option, then GPT-4, Claude 3, Gemini 1.5 or Mistral Large are preferred. Instead, Aleph Alpha is in danger of becoming a contract developer for a few partners.

01.AI: Chinese AI startup shakes up the world of LLMs

Do the people of Linz have to fix it?

Must Europe’s AI hopes now rest on Linz? That’s where AI researcher Sepp Hochreiter and tech entrepreneur Albert Ortig launched NXAI. And they’re chipping away at the foundations on which OpenAI and others are built – the transformers. xLTSM is intended to be a better alternative. That needs to be proven, but more and more people are looking to Sepp in Linz to see what’s happening there. NXAI needs money, in any case, a major financing round is to be held this year and a competitive AI model is to be launched on the market. And then we’ll see: Can Europe finance a GPT challenger itself, or will the Americans end up raising the capital again – and buying influence?