Mistral AI rejects acquisition offers to preserve independence
Mistral AI rejected acquisition offers to preserve its independence during testimony before French parliamentary representatives. Executive Arthur Mensch stated that the company's models can identify all vulnerabilities detected by Mythos. The company is directing 1 billion euros to R&D this year and allocating 10 percent of salary expenditure to tokens. It operates collocated GPU clusters at 40 MW in France and 25 MW in Sweden with 80 MW scheduled for addition next year toward a 1 GW target by 2029.
@Ar_Douillard @eliebakouch @fouriergalois Approp time to say “skill issue” and drop the 8 way diloco variant over 10 MGW clusters
Arthur Douillard@Ar_Douillard@eliebakouch @fouriergalois > mention that they need the gpu to be collocated to train model 😶
ngmi
elie@eliebakouchArthur Mensch answers to the french representatives: "our (mistral) models are capable of finding all the vulnerabilities found by mythos" "There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed" some numbers: > 1B R&D spend at Mistral this year > at Mistral 10% of salary mass is spent on tokens > estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$ > 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. > revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. > a bit less than 30% of Mistral capital is held by US VCs > Mistral's goal is 1GW in 2029 > they train/will train bigger models internally and distill them to serve to customers > Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year some of their current clusters: > 40MW in France > 25MW in Sweden > 80MW in France (next year) > they train models on "10s of MW", mention that they need the gpu to be collocated to train model > insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint
#767elie@ELIEBAKOUCHArthur Mensch answers to the french representatives:
"our (mistral) models are capable of finding all the vulnerabilities found by mythos"
"There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed"
some numbers: > 1B R&D spend at Mistral this year > at Mistral 10% of salary mass is spent on tokens > estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$ > 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. > revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. > a bit less than 30% of Mistral capital is held by US VCs > Mistral's goal is 1GW in 2029 > they train/will train bigger models internally and distill them to serve to customers > Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year
some of their current clusters: > 40MW in France > 25MW in Sweden > 80MW in France (next year) > they train models on "10s of MW", mention that they need the gpu to be collocated to train model > insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint
#767elie@ELIEBAKOUCHfull transcript if you want https://gist.github.com/eliebak/5945ce3c84d77b0f30ea1190cf34ad5b
elie@eliebakouchArthur Mensch answers to the french representatives: "our (mistral) models are capable of finding all the vulnerabilities found by mythos" "There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed" some numbers: > 1B R&D spend at Mistral this year > at Mistral 10% of salary mass is spent on tokens > estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$ > 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. > revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. > a bit less than 30% of Mistral capital is held by US VCs > Mistral's goal is 1GW in 2029 > they train/will train bigger models internally and distill them to serve to customers > Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year some of their current clusters: > 40MW in France > 25MW in Sweden > 80MW in France (next year) > they train models on "10s of MW", mention that they need the gpu to be collocated to train model > insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint
#767elie@ELIEBAKOUCH@Ar_Douillard @fouriergalois ahah i thought about you when writing this part
Arthur Douillard@Ar_Douillard@eliebakouch @fouriergalois > mention that they need the gpu to be collocated to train model 😶
#841kalomaze@KALOMAZE
elie@eliebakouchArthur Mensch answers to the french representatives: "our (mistral) models are capable of finding all the vulnerabilities found by mythos" "There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed" some numbers: > 1B R&D spend at Mistral this year > at Mistral 10% of salary mass is spent on tokens > estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$ > 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. > revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. > a bit less than 30% of Mistral capital is held by US VCs > Mistral's goal is 1GW in 2029 > they train/will train bigger models internally and distill them to serve to customers > Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year some of their current clusters: > 40MW in France > 25MW in Sweden > 80MW in France (next year) > they train models on "10s of MW", mention that they need the gpu to be collocated to train model > insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint
@eliebakouch @fouriergalois > mention that they need the gpu to be collocated to train model
😶
elie@eliebakouchArthur Mensch answers to the french representatives: "our (mistral) models are capable of finding all the vulnerabilities found by mythos" "There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed" some numbers: > 1B R&D spend at Mistral this year > at Mistral 10% of salary mass is spent on tokens > estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$ > 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. > revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. > a bit less than 30% of Mistral capital is held by US VCs > Mistral's goal is 1GW in 2029 > they train/will train bigger models internally and distill them to serve to customers > Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year some of their current clusters: > 40MW in France > 25MW in Sweden > 80MW in France (next year) > they train models on "10s of MW", mention that they need the gpu to be collocated to train model > insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint
@_arohan_ @eliebakouch @fouriergalois This is the way
rohan anil@_arohan_@Ar_Douillard @eliebakouch @fouriergalois Approp time to say “skill issue” and drop the 8 way diloco variant over 10 MGW clusters
#897Alexander Doria@DORIALEXANDERi'm not so sure: reality might hit violently after summer…
Lisan al Gaib@scaling01You will see a lot more european turbo cope in the next couple years
#984Lisan al Gaib@SCALING01holy cope
they are further behind than Meta, xAI and all of the chinese labs
they will probably only replicate Mythos in 1-2 years
elie@eliebakouchArthur Mensch answers to the french representatives: "our (mistral) models are capable of finding all the vulnerabilities found by mythos" "There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed" some numbers: > 1B R&D spend at Mistral this year > at Mistral 10% of salary mass is spent on tokens > estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$ > 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. > revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. > a bit less than 30% of Mistral capital is held by US VCs > Mistral's goal is 1GW in 2029 > they train/will train bigger models internally and distill them to serve to customers > Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year some of their current clusters: > 40MW in France > 25MW in Sweden > 80MW in France (next year) > they train models on "10s of MW", mention that they need the gpu to be collocated to train model > insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint
even saying they can replicate it is wrong
they will probably never be able to replicate it
but they can get similar benchmark performance in smaller models
Lisan al Gaib@scaling01holy cope they are further behind than Meta, xAI and all of the chinese labs they will probably only replicate Mythos in 1-2 years
#984Lisan al Gaib@SCALING01You will see a lot more european turbo cope in the next couple years
Lisan al Gaib@scaling01holy cope they are further behind than Meta, xAI and all of the chinese labs they will probably only replicate Mythos in 1-2 years
#1483bilal@BILALTWOVEC> our (mistral) models are capable of finding all the vulnerabilities found by mythos no they're not lmao. anyways people doubling down on points like these to try and reassure investors that their models are good enough to not be irrelevant are lying
elie@eliebakouchArthur Mensch answers to the french representatives: "our (mistral) models are capable of finding all the vulnerabilities found by mythos" "There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed" some numbers: > 1B R&D spend at Mistral this year > at Mistral 10% of salary mass is spent on tokens > estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$ > 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. > revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. > a bit less than 30% of Mistral capital is held by US VCs > Mistral's goal is 1GW in 2029 > they train/will train bigger models internally and distill them to serve to customers > Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year some of their current clusters: > 40MW in France > 25MW in Sweden > 80MW in France (next year) > they train models on "10s of MW", mention that they need the gpu to be collocated to train model > insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint
elie@eliebakouchArthur Mensch answers to the french representatives: "our (mistral) models are capable of finding all the vulnerabilities found by mythos" "There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed" some numbers: > 1B R&D spend at Mistral this year > at Mistral 10% of salary mass is spent on tokens > estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k$ > 1GW datacenter is $50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. > revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. > a bit less than 30% of Mistral capital is held by US VCs > Mistral's goal is 1GW in 2029 > they train/will train bigger models internally and distill them to serve to customers > Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year some of their current clusters: > 40MW in France > 25MW in Sweden > 80MW in France (next year) > they train models on "10s of MW", mention that they need the gpu to be collocated to train model > insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint