Deconstructing The DeepSeek Story
Using the classic 3-act story model
Hi friend,
How are you? I hope you had a productive and creative week.
Is your head still spinning from the recent whirlwind drama around DeepSeek?
Well, no worries, I’m here for you to deconstruct the whole story to its basic parts and focus only on the key aspects of what happened.
Baseline Information
Let’s start with labeling the key characters.
A foundation AI model - like OpenAI’s ChatGPT or Anthropic’s Claude chatbot - powers many applications like search engines and image generators.
To create such a chatbot, you need three ingredients:
1. Compute – Powerful computers, often using expensive NVIDIA chips, to run AI models.
2. Data – The information AI learns from.
3. Training & Algorithms – The rules and methods AI follows to learn and make decisions.
Now, let’s break down the story into its 3 acts:
Act 1: Setting
The Dominant Narrative
Since OpenAI released ChatGPT in September 2022, the common narrative has been that these foundation models are expensive because they need a lot of computing power for training.
In support of this narrative, PitchBook reports a funding frenzy pouring more than $155 billion into AI startups between 2023 and 2024.
AI startups - OpenAI and Anthropic - are raising $24 billion and $16 billion to build AI that aims to be as intelligent as humans.
To enable this massive compute power startups relied on Nvidia’s expensive chips that catapulted the company to fame and untold fortune.
Act 2: Conflict
But…
Early last week, a Chinese startup called DeepSeek rocked the global markets by introducing a powerful AI model with far less money and less compute power to train than the common narrative earlier suggested.
How less?
$6 million compared to $100 million to $1 billion other AI startups spend on training their models.
Huge difference, right?
Over a few days, the DeepSeek app surpassed OpenAI’s ChatGPT as the top downloaded app in the Apple App Store in the U.S.
What's more, the app’s ascend in popularity also wiped $1 trillion from the leading U.S. tech index, including $600 billion from Nvidia - a leading maker of those expensive computer chips that power AI models.
Act 3: Resolution
The ricochets from what the media calls a ‘Sputnik moment’ are still percolating, yet a few lessons start to surface.
DeepSeek used an open-source model, partly from Meta’s open-source Llama model, which vindicates Meta’s narrative for using open-source vs. proprietary models like OpenAI’s.
Investors are now scratching their heads why invest so much money in AI startups if you can get the same quality at a minuscule fraction of the cost?
Some investors now say the current AI foundation model market is unstable and too risky.
DeepSeek’s rise to the top comes in the face of the U.S. government's attempts to keep advanced chips out of the hands of Chinese companies to avoid risks they’ll use them for military purposes.
Some speculate that’s exactly what forced DeepSeek to innovate by finding a workaround.
Another emerging moral that started to gel from DeepSeek’s overnight success - is that, hey, this is proof that the game is now open to healthier competition even from smaller AI startups.
As evidence of this, the Hugging Face website where people post their open-source projects, now has more than 600 versions of the DeepSeek model!
At the heels of the TikTok debacle, we also witness another emerging truth: consumers prefer quality over the origin country of the product, considering DeepSeek’s overnight popularity.
And that, yes we’re still in the early days of AI evolution with more surprising plot twists ahead.
There is still a lot of fog surrounding the validity of DeepSeek’s claims, yet no matter how you explain it, in just a few days, this startup generated seismic changes:
Stock market havoc
Toppling down ChatGPT as the most downloaded app
Rocking 3 narratives:
It’s expensive to train foundation models
Proprietary models are better than open-source
Consumers care about products’ country of origin
Stay tuned to the next episode as the AI landscape continues to evolve.
It should be exciting :)
See you next time!
Best,
- Shlomi
Shlomi Ron
Chief Storytelling Officer
shlomi@visualstorytell.com
Visual Storytelling Institute
story > visual > emotion > experience
Join my Fall Visual Storytelling Masterclass! Spots are limited.
Paid subscribers get 40% off!
Enjoyed this story? A paid subscription gets you:
Top 10 Business Storytelling Formulas ebook to revamp your marketing.
Exclusive stories and bi-weekly news commentary on Sundays with my analysis around visual storytelling news and trends.
Full access to my newsletter archive.
Exclusive discounts on training programs.