Sounds like primetime to bring charges against the labs with multibillion dollar fines, to force them to create a domestic lab with SOTA model development for Europeans.
You know, anything but create fertile business conditions.
You know there was a lab that was cutting edge in Europe and it is still there - Deepmind - but the UK let it be sold to the US for a fraction of what it's worth today - $400m. We just allow US companies to asset strip the EU and then people complain about us never having big companies here...
Deepmind would likely be worth conservatively $200 BILLION today (probably more without the overheads of joining Google to be honest) or around 500 times the sale price in 2014, not a bad return.
How much of the current valuation is due to Google's resources and support, though?
"Sell company now and it's worth more a decade later" is the entire point; acquisitions are an investment and the investment is supposed to appreciate.
It’s impossible to know for sure but if they’d stayed private I think they’d be more not less successful, just because startups can move more nimbly. Although we’d need to define success, we’d be here forever…
It wouldn't have. They had over a billion in debt, which Google casually fixed. Google then very conveniently supplied them with GCP server power. And remember all of this was in like 2014, way before the AI hype. They wouldn't have had access to capital to survive long enough to make it to today.
This applies to pretty much everyone who isn't the US/China, no? I guess Russia also has a good amount of home grown web services (Yandex, Wildberries, etc), but not much in the AI space that I'm aware of.
The real surprising part is that Anthropic is also negligible (according to this article) in Germany. It's all OpenAI. But I didn't dare to change the title that much.
> The real surprising part is that Anthropic is also negligible (according to this article) in Germany. It's all OpenAI.
Besides a few software devs about zero people I talk to (I'm in the EU) have ever heard of Claude / Anthropic. It's all Gemini (the default in Google), ChatGPT (pretty much a verb for "I ask questions to an AI") and some have Grok subscriptions (!).
When I mention Anthropic and/or Claude people have zero clue what I'm talking about.
One piece of anecdata: after first getting into Claude Code and then discovered the Claude app I briefly got excited about using it in voice chat mode, demonstrated this to my Austrian wife who found it immediately cool then of course asked if she could chat with it in German. On finding the answer was no, she immediately signed up for ChatGPT so she could use the same feature in her native language. I’m sure Anthropic models are decent with German in text chat, but the lack of that one feature is what drove her straight to a competitor.
One of the reasons self-hosting is not sustainable in Germany is that electricity is prohibitively expensive for non-industrial purposes. If you run a service company and do not own a data center, you pay more for the electricity than the equivalent in token costs overseas or in China.
I think since electricity costs are the most important facet of the margins, and the cheapest electricity source is solar, this means the largest inference providers will run massive photovoltaic parks.
Not sure if anybody wants to combine the CAPEX risk of running a cloud business and the CAPEX risk of building out the massive capacities of solar, plus lease for the land, plus managing electricity backups for cloudy days.
If there was a good EU alternative it may have been used but Mistral was out of the game a long time ago now that does not mean they will close down or will not be profitable so long as they work on regulation and other things like ethics and how good they are for the environment etc some may choose them.
There are "sovereign" clouds like StackIT, OVH, OTC or Scaleway. What they all have in common is that they are lightyears behind hyperscalers in terms of performance, reliability and security. Needless to say, these alternatives can only serve open-source models because they are not capable of training a SOTA equivalent by themselves. The cherry on top is that you have to pay a sovereignty premium for inferior performance.
Local DeepSeek, qwen or kimi models are far better privacy-wise, faster, but much more expensive (to get a 2TB GPU RAM cluster). So everybody went with the Claude, OpenAI or CoPilot business contracts, so their secret business data leaves the EU, but will not be used to train the model. Unlike with the cloud Chinese services.
Sounds like primetime to bring charges against the labs with multibillion dollar fines, to force them to create a domestic lab with SOTA model development for Europeans.
You know, anything but create fertile business conditions.
You know there was a lab that was cutting edge in Europe and it is still there - Deepmind - but the UK let it be sold to the US for a fraction of what it's worth today - $400m. We just allow US companies to asset strip the EU and then people complain about us never having big companies here...
Deepmind would likely be worth conservatively $200 BILLION today (probably more without the overheads of joining Google to be honest) or around 500 times the sale price in 2014, not a bad return.
How much of the current valuation is due to Google's resources and support, though?
"Sell company now and it's worth more a decade later" is the entire point; acquisitions are an investment and the investment is supposed to appreciate.
It’s impossible to know for sure but if they’d stayed private I think they’d be more not less successful, just because startups can move more nimbly. Although we’d need to define success, we’d be here forever…
It wouldn't have. They had over a billion in debt, which Google casually fixed. Google then very conveniently supplied them with GCP server power. And remember all of this was in like 2014, way before the AI hype. They wouldn't have had access to capital to survive long enough to make it to today.
You don't say?! :joy:
This applies to pretty much everyone who isn't the US/China, no? I guess Russia also has a good amount of home grown web services (Yandex, Wildberries, etc), but not much in the AI space that I'm aware of.
The real surprising part is that Anthropic is also negligible (according to this article) in Germany. It's all OpenAI. But I didn't dare to change the title that much.
> The real surprising part is that Anthropic is also negligible (according to this article) in Germany. It's all OpenAI.
Besides a few software devs about zero people I talk to (I'm in the EU) have ever heard of Claude / Anthropic. It's all Gemini (the default in Google), ChatGPT (pretty much a verb for "I ask questions to an AI") and some have Grok subscriptions (!).
When I mention Anthropic and/or Claude people have zero clue what I'm talking about.
One piece of anecdata: after first getting into Claude Code and then discovered the Claude app I briefly got excited about using it in voice chat mode, demonstrated this to my Austrian wife who found it immediately cool then of course asked if she could chat with it in German. On finding the answer was no, she immediately signed up for ChatGPT so she could use the same feature in her native language. I’m sure Anthropic models are decent with German in text chat, but the lack of that one feature is what drove her straight to a competitor.
German has been supported in Claude voice mode since May.
Russian companies rely a lot on chatgpt
One of the reasons self-hosting is not sustainable in Germany is that electricity is prohibitively expensive for non-industrial purposes. If you run a service company and do not own a data center, you pay more for the electricity than the equivalent in token costs overseas or in China.
I think since electricity costs are the most important facet of the margins, and the cheapest electricity source is solar, this means the largest inference providers will run massive photovoltaic parks.
Not sure if anybody wants to combine the CAPEX risk of running a cloud business and the CAPEX risk of building out the massive capacities of solar, plus lease for the land, plus managing electricity backups for cloudy days.
Vertical integration, amirite?
You still can use EU-Hosted Opensources models like from Tensorix.
If there was a good EU alternative it may have been used but Mistral was out of the game a long time ago now that does not mean they will close down or will not be profitable so long as they work on regulation and other things like ethics and how good they are for the environment etc some may choose them.
You can use Asian Opensource models from EU-only hosters. Zero problems - they are around for a while now.
Title is edited as the original title buries the "German" part.
Actually the German title of the aricle reads: > China's open-weight models do not yet play a significant role
We have long accepted the fact that EU plays no role I guess
Isn't it applicable to all EU companies though?
It obviously is applicable to all companies outside China, but the article is about German companies.
AI is centralization of compute. And the new fortress USA makes that even worse.
Until we get local AI, which Anthropic and OpenAI are trying their hardest to skuttle, we are in big trouble.
I mean what other competitively similar AI models are there? DeepSeek is nowhere near. Are there others?
The short answer is No.
There are "sovereign" clouds like StackIT, OVH, OTC or Scaleway. What they all have in common is that they are lightyears behind hyperscalers in terms of performance, reliability and security. Needless to say, these alternatives can only serve open-source models because they are not capable of training a SOTA equivalent by themselves. The cherry on top is that you have to pay a sovereignty premium for inferior performance.
Local DeepSeek, qwen or kimi models are far better privacy-wise, faster, but much more expensive (to get a 2TB GPU RAM cluster). So everybody went with the Claude, OpenAI or CoPilot business contracts, so their secret business data leaves the EU, but will not be used to train the model. Unlike with the cloud Chinese services.
I use DeepSeek and GLM for coding. They are on-par or better than Opus/Astra.
You can use Deepseek from EU-only hosters. DS is openweight.