Deck

Labs & Companies

Cards for the companies and labs that build AI.

NVIDIA Legendary

Company · Base Set

LegendaryCompanyBase Set

Power 10 · Danger 6 · Speed 10 · Difficulty 7

Year

2007

What they did

American semiconductor company (founded 1993) that makes GPUs. Its CUDA software platform (2007) turned GPUs into the dominant hardware for AI training and inference.

Open or closed

Closed

Description

NVIDIA is a company that makes special computer chips called GPUs. In 2007, they created software that let these chips power AI training.

Analogy

It is like turning a single fast runner into a team of thousands who can all run at once.

Example

Their 2007 software platform turned GPUs into the main hardware for AI.

Flavour

The green scales shimmer with the hum of a thousand circuits working in unison.

Signature

While in play, your Foundation row's total Compute counts +2.

Learn this in chapter 9 →

OpenAI Legendary

Company · Base Set

LegendaryCompanyBase Set

Power 10 · Danger 5 · Speed 9 · Difficulty 4

Year

2022

What they did

US AI research company founded in 2015, known for the GPT family of large language models; ChatGPT (2022) brought LLMs to a mass audience. It also released Whisper, an open-source speech-recognition model (2022).

Open or closed

Both

Description

OpenAI is a US AI research company founded in 2015. It built the GPT family of large language models. In 2022, it released ChatGPT and the open-source speech model Whisper.

Analogy

It is like a giant library that learned to talk to everyone at once.

Example

In 2022, ChatGPT brought large language models to a mass audience.

Flavour

The gate stands closed, yet the light rings whisper to all who pass.

Signature

Cloud Lab only for its GPT models. Once per game, your Run counts +1 Power (mass audience) but costs +3 Credits.

Learn this in chapter 2 →

PyTorch Epic

Company · Base Set

EpicCompanyBase Set

Power 9 · Danger 3 · Speed 8 · Difficulty 7

Year

2016

What they did

Open-source (BSD-licensed) deep-learning library for Python, first released September 2016 and developed out of Facebook AI Research. Its dynamic computation graph made it the dominant framework for AI research.

Open or closed

Open Weights

Description

PyTorch is a free tool for building smart programs in Python. It was created by Facebook AI Research in 2016. It helps computers learn by using a flexible map of tasks called a dynamic computation graph.

Analogy

It is like a smart notebook that lets you change the plan while you are still writing it.

Example

It was first released in September 2016 and is free to use because it is open-source.

Flavour

The laboratory burns with a living fire that reshapes itself with every new idea.

Signature

While in play, your Python cards cost -1 Talent (research runs on it).

Learn this in chapter 9 →

Hugging Face Epic

Company · Base Set

EpicCompanyBase Set

Power 8 · Danger 2 · Speed 8 · Difficulty 4

Year

2018

What they did

US machine-learning company (founded 2016) that built the Transformers library (2018) and the model-sharing hub that became the de facto home of open AI.

Open or closed

Open Weights

Description

Hugging Face is a US company that started in 2016. They built the Transformers library in 2018. They also made the main place where people share open AI models.

Analogy

It is like a giant library where anyone can borrow books for free.

Example

In 2018, they created the Transformers library.

Flavour

The great archive stands open, its shelves glowing with the warmth of shared knowledge.

Signature

Once per turn, pip install may pull any Open-Weights card from the supply pile for free. (The hub.)

Learn this in chapter 11 →

Meta AI Rare

Company · Base Set

RareCompanyBase Set

Power 8 · Danger 3 · Speed 8 · Difficulty 5

Year

2023

What they did

The AI division of Meta (formerly Facebook). Its research lab (FAIR, 2013) developed PyTorch and, in 2023, began releasing the open-weight Llama family of models.

Open or closed

Both

Description

Meta AI is the research team inside the company formerly known as Facebook. They built PyTorch, a tool for making AI, and released the Llama models in 2023.

Analogy

It is like a public library that gives out free copies of complex instruction manuals so anyone can read them at home.

Example

In 2023, they started sharing the Llama family of models so people could use them locally.

Flavour

The Great House opens its vaults, letting the common folk carry the heavy scrolls of wisdom into their own homes.

Signature

While in play, your Open-Weights model cards' VRAM cost -1.

Learn this in chapter 2 →

Google DeepMind Rare

Company · Base Set

RareCompanyBase Set

Power 9 · Danger 4 · Speed 8 · Difficulty 7

Year

2016

What they did

UK AI research company founded in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman, acquired by Google in 2014. AlphaGo beat the Go world champion in 2016; it now builds Google's Gemini models.

Open or closed

Closed

Description

Google DeepMind is a UK AI research company founded in 2010. It was acquired by Google in 2014. It now builds Google's Gemini models.

Analogy

It is like a team of expert chefs who invent new recipes and then open a famous restaurant.

Example

In 2016, its program AlphaGo beat the world champion at the game of Go.

Flavour

Within the spiral walls, the dragon weaves patterns that even the masters cannot see.

Signature

Cloud Lab only. Once per game, one Run check's arithmetic counts +2 (superhuman play).

Learn this in chapter 2 →

llama.cpp Rare

Company · Base Set

RareCompanyBase Set

Power 8 · Danger 2 · Speed 7 · Difficulty 5

Year

2023

What they did

Open-source C/C++ project (started 2023) for running large language models locally, often without a GPU. It defines the GGUF file format used by most local LLM tools. A community project, not a company.

Open or closed

Open Weights

Description

llama.cpp is a free, open-source project that started in 2023. It lets people run big AI models on their own computers, often without special graphics cards. It also created the GGUF file format used by most local AI tools.

Analogy

It is like a universal adapter that lets powerful AI software run on simple, everyday hardware.

Example

This project defines the GGUF file format used by most local LLM tools.

Flavour

In the quiet workshop of the metal llama, heavy minds are coaxed to breathe on humble engines.

Signature

While in play, your models may Run on a CPU-only Foundation: their Compute cost -1.

Learn this in chapter 9 →

Anthropic Rare

Company · Base Set

RareCompanyBase Set

Power 7 · Danger 3 · Speed 7 · Difficulty 5

Year

2021

What they did

US AI safety and research company founded in 2021 by former OpenAI members including Daniela and Dario Amodei. It builds the Claude family of models and focuses on interpretability and safety.

Open or closed

Closed

Description

Anthropic is a US company that started in 2021. It was founded by people who used to work at OpenAI. They build the Claude family of AI models and focus on keeping them safe.

Analogy

It is like a careful engineer who builds a car and checks the brakes twice before letting anyone drive it.

Example

The company was founded in 2021 by former OpenAI members including Daniela and Dario Amodei.

Flavour

The keepers of the quiet eye watch the stars to ensure no shadow falls on the mind.

Signature

Cloud Lab only. Once per game, ignore one failed Run (safety net).

Learn this in chapter 2 →

Alibaba (Qwen) Rare

Company · Base Set

RareCompanyBase Set

Power 7 · Danger 2 · Speed 7 · Difficulty 4

Year

2023

What they did

Chinese technology company behind the Qwen (Tongyi Qianwen) large language model family, launched April 2023. It open-sourced the Qwen models, making them among the most widely used open-weight LLMs.

Open or closed

Open Weights

Description

Alibaba is a Chinese tech company that built the Qwen model family. They launched it in April 2023 and shared the code with everyone.

Analogy

It is like a library that gives away its rarest books for free so anyone can read them.

Example

The Qwen models were released in April 2023 and are among the most widely used open-weight LLMs.

Flavour

The golden flame rises from the circuit sea, sharing its light with all who seek knowledge.

Signature

While in play, your model counts +1 Power (the most-downloaded open family).

Learn this in chapter 2 →

Mistral AI Rare

Company · Base Set

RareCompanyBase Set

Power 7 · Danger 2 · Speed 8 · Difficulty 4

Year

2023

What they did

French AI company founded in 2023 in Paris by ex-DeepMind researchers including Arthur Mensch. Known for the open-weight Mistral 7B model, released September 2023.

Open or closed

Open Weights

Description

Mistral AI is a French company started in Paris in 2023. It was built by former DeepMind researchers, including Arthur Mensch. They created the open-weight Mistral 7B model.

Analogy

It is like a small, fast workshop that builds powerful tools and shares the blueprints with everyone.

Example

In September 2023, they released the Mistral 7B model.

Flavour

The wind-swept halls of this laboratory hum with the quiet speed of seven steps.

Signature

While in play, your model's Speed counts +1 (small, sharp, fast).

Learn this in chapter 2 →

DeepSeek Rare

Company · Base Set

RareCompanyBase Set

Power 8 · Danger 3 · Speed 8 · Difficulty 6

Year

2024

What they did

Chinese AI research company that develops and open-sources frontier large language models. Its V3 (December 2024) and R1 (January 2025) releases were noted for strong performance at low training cost.

Open or closed

Open Weights

Description

DeepSeek is a Chinese AI research company. They build powerful large language models. They share these models with the public for free.

Analogy

It is like a master chef who cooks a five-star meal using very little money and shares the recipe with everyone.

Example

Their V3 and R1 models showed strong performance while costing less to train.

Flavour

The lanterns of the Deep Well burn bright, yet the oil is never wasted.

Signature

Once per game, your Cloud costs are reduced by 2 Credits for the rest of the game.

Learn this in chapter 2 →

EleutherAI Rare

Company · Base Set

RareCompanyBase Set

Power 6 · Danger 1 · Speed 6 · Difficulty 4

Year

2020

What they did

Non-profit AI research collective formed in 2020 (started as a Discord server) to build open-source equivalents of GPT-3. It created The Pile, an ~800GB open training dataset, and open models like GPT-J, GPT-NeoX and Pythia.

Open or closed

Open Weights

Description

EleutherAI is a non-profit group that started in 2020 to build open AI models. They created The Pile, a large open dataset, and models like GPT-J and Pythia.

Analogy

It is like a community kitchen where everyone shares the recipes and ingredients instead of keeping them secret.

Example

They built The Pile, an open training dataset that is about 800GB in size.

Flavour

The keepers of the open flame share their light with all who seek knowledge.

Signature

Once per game, gain 3 Data (the open dataset).

Learn this in chapter 2 →

Allen Institute for AI Rare

Company · Base Set

RareCompanyBase Set

Power 6 · Danger 1 · Speed 5 · Difficulty 4

Year

2024

What they did

Non-profit AI research institute founded in 2014 by Microsoft co-founder Paul Allen, based in Seattle. It released OLMo (February 2024), a fully open LLM — weights, data and code.

Open or closed

Open Weights

Description

The Allen Institute for AI is a non-profit research group started in 2014 by Paul Allen. It is based in Seattle and focuses on open AI research. In 2024, it released OLMo, a large language model that shares its weights, data, and code.

Analogy

It is like a public library where anyone can borrow the books, see how they were printed, and even check the recipe for the paper.

Example

In February 2024, the institute released OLMo, a fully open large language model.

Flavour

The gates of this house stand wide open, inviting all to study the secrets within.

Signature

Once per game, reveal the top card of the supply and play it for free (fully open).

Learn this in chapter 2 →

Ollama Rare

Company · Base Set

RareCompanyBase Set

Power 6 · Danger 1 · Speed 8 · Difficulty 2

Year

2023

What they did

Open-source (MIT-licensed) software platform released July 2023 for downloading and running large language models on a local machine. It uses the llama.cpp backend and became one of the easiest ways to run local LLMs.

Open or closed

Open Weights

Description

Ollama is a free software platform released in July 2023. It lets you download and run large language models on your own computer. It is one of the easiest ways to use local AI.

Analogy

It is like a simple app store that lets you download and run big brain programs right on your own device.

Example

The platform was released in July 2023 and uses the llama.cpp backend to run models locally.

Flavour

The Gatekeeper offers a free pass to the first traveler who enters the laboratory.

Signature

Your first model each game is played at half cost (one command).

Learn this in chapter 9 →

ComfyUI Rare

Company · Base Set

RareCompanyBase Set

Power 6 · Danger 2 · Speed 6 · Difficulty 5

Year

2023

What they did

Open-source (GPLv3) node-based program released January 2023, created by 'comfyanonymous'. It lets users build image-generation workflows with diffusion models, each tool as a node. A community project, not a company.

Open or closed

Open Weights

Description

ComfyUI is a free, open-source tool released in January 2023. It lets you build image creation steps by connecting tools like puzzle pieces.

Analogy

It is like a digital circuit board where you plug in different lights to make a picture.

Example

This project started as a community effort in 2023, not as a business.

Flavour

The hidden lab hums with the glow of a thousand tiny stars, each one a step in the magic.

Signature

Once per game, play two cards this turn at -1 cost each (wire the nodes).

Learn this in chapter 9 →

← All decks