Twitter user Morgan Brown
VP of Product Development at Instagram, Dropbox, Shopify
DeepSeek's innovations in AI are astounding (and possibly threatening Nvidia's $2T market cap).
1/ First, a bit of context: Training top AI models is EXTREMELY expensive right now.
OpenAI, Anthropic, and others spend over $100 million just on computing resources. They need huge data centers with thousands of $40K GPUs. It's like needing a whole power plant to run a factory.
2/ Then DeepSeek came along and said: "What if we could do this for $5M?"
And they didn't just talk — they actually did it. Their models match or outperform GPT-4 and Claude in many tasks.
The AI world is shedding tears into tiramisu.
3/ How? They rethought everything from scratch. Traditional AI is like writing every number with 32 decimal places. DeepSeek said: "What if we only use 8? That's enough!"
Boom — 75% less memory.
4/ Then their "multi-token" system.
Regular AI reads like a first-grader: "Cat... sits... on...". DeepSeek reads whole sentences at once. Twice as fast, with 90% accuracy. When you're processing billions of words, this is IMPORTANT.
5/ But the coolest part of their approach:
They created an "expert system".
Instead of one massive AI trying to know everything (like one person being a doctor, lawyer, and engineer), they have specialized experts who activate only when needed.
6/ Traditional models?
All 1.8 trillion parameters are active ALL THE TIME.
DeepSeek? 671 billion parameters in total, but only 37 billion are active at the same time.
It's like having a huge team but only calling in the experts who are needed for each task.
7/ The results are staggering:
Training cost: $100M → $5MGPUs needed: 100,000 → 2,000API cost: 95% cheaperCan run on gaming GPUs instead of server hardware.
8/ "But wait," you might say, "there must be a catch!" Here's the interesting part — it's all open-source.
Anyone can verify their work. The code is public. Technical documents explain everything. It's not magic, just incredibly smart engineering.
9/ Why is this important? Because it breaks the model of "only big tech companies can play in AI".
Now you don’t need a billion-dollar data center. A few good GPUs might be enough.
10/ This is scary for Nvidia. Their entire business model is built on selling super-expensive GPUs with 90% margins. If everyone can do AI on regular gaming GPUs...
11/ DeepSeek did this with a team of fewer than 200 people. Meanwhile, Meta has teams where just the employee compensation exceeds DeepSeek's entire training budget... and their models aren't as good.
12/ This is a classic disruption story: Market leaders optimize existing processes, while disruptors rethink the fundamental approach.
DeepSeek asked: "What if we just make it smarter instead of using more hardware?"
13/ The consequences:
AI development becomes more accessibleCompetition increases dramaticallyHardware from big tech companies seems unnecessaryEquipment requirements (and costs) drop sharply
14/ Of course, giants like OpenAI and Anthropic won’t sit still.
They’re probably already implementing these innovations. But the genie is out of the bottle — there’s no going back to the "just throw more GPUs at it" approach.
15/ This feels like one of those moments we'll look back on as a turning point. Like when PCs made mainframes less relevant, or when cloud computing changed everything.
AI will soon be much more accessible and much cheaper. The question is not whether this will be a breakthrough for current players, but how quickly it will happen.