Mistral AI pitch deck (2023)

What's in the Mistral AI pitch deck

All 7 slides, in order.

1. 

   Slide 1 · Generative AI and the oligopoly

   Generative AI will lift productivity across every sector, a market put at $10B in 2022 and $110B by 2030. A few actors, led by OpenAI, are taking the market, and the memo argues most of the value will sit in the models themselves.

2. 

   Slide 2 · Why closed models fall short

   All the major players are American, which the memo calls a geopolitical issue for Europe. Closed models force businesses to send data to a black box, are hard to connect to other systems and hide their training data.

3. 

   Slide 3 · Technological counter-positioning

   Mistral's differentiators. Release strong open-source models, give customers access to model weights, train on licensed high-quality data, and make models small enough to run on private clouds or devices. 1% of funding goes to an open-source foundation.

4. 

   Slide 4 · Business development and team

   Sell to European industry through licensing, custom models and integrator partners. The team is Arthur Mensch (ex-DeepMind), Guillaume Lample and Timothée Lacroix (ex-Meta, Llama), plus Alan's founders and former French minister Cédric O.

   How to write this slide
5. 

   Slide 5 · Infrastructure, clients and roadmap

   Plans to rent 1,536 H100 GPUs from September, a claim of 10 to 100 times better training efficiency than public methods, a 6-person product team by the end of year one, and two model generations in the first year.

6. 

   Slide 6 · First-year milestones

   By the end of 2023, models that beat ChatGPT 3.5 by a large margin, partly open source. By Q2 2024, a model that runs on a 16GB laptop, models with millions of words of extra context, and one or two major industrial clients.

7. 

   Slide 7 · Next stages

   Safety through staged releases and red-team access, and a planned Series A of 200M in Q3 2024 to train models beyond GPT-4.

What the Mistral AI deck does well

• ✓Page 3 turns the competition's choices into Mistral's plan, open models, open weights and private deployment, so the strategy reads as a direct answer to OpenAI.
• ✓Page 4 puts the team at the center, naming the papers and models each founder led, like Chinchilla and Llama, which is the real proof for a company with no product yet.
• ✓Pages 5 to 7 commit to dated milestones, from 1,536 H100s in September to a laptop-sized model by Q2 2024 and a Series A in Q3 2024, so investors can check progress later.

What founders can learn from it

• →If you have no product, sell the argument and the team. Mistral spends its first three pages on why the market will split its way and the fourth on who can build it.
• →Use specific numbers in your plan, like a GPU count or a model size, rather than vague goals. They show you know what the work takes.
• →A written memo can work better than slides when the case is an argument. It lets investors read the full reasoning without you in the room.

Questions about the Mistral AI pitch deck

Is there a Mistral AI pitch deck?

Mistral raised its seed round with a seven-page written memo, not a slide deck. Sifted published the memo in June 2023.

How much did Mistral raise with this memo?

Mistral AI raised a €105M seed round in June 2023, led by Lightspeed Venture Partners.

Who founded Mistral AI?

Arthur Mensch, formerly of DeepMind, and Guillaume Lample and Timothée Lacroix, formerly of Meta, where they worked on Llama.

What did the Mistral memo promise?

Open-source models better than ChatGPT 3.5 by the end of 2023, a model small enough to run on a laptop by mid-2024, and a Series A of around 200M in Q3 2024.

More pitch decks

• AirbnbAirbnb pitch deck2008 · Seed · $600K
• UberUber pitch deck2008 · Seed
• CoinbaseCoinbase pitch deck2012 · Seed · $600K
• NotionNotion pitch deck2013 · Seed