China’s $5.6M AI Just Wiped $1 Trillion—How?! | DeepSeek vs OpenAI

DeepSeek’s Deep Reasoning R1 sent shockwaves through the AI world, triggering a $1 trillion market wipeout. NVIDIA, Broadcom, and Oracle all saw massive losses overnight. But why?

DeepSeek claims it built a model rivaling GPT-4o & o1 for just $5.6 million—a fraction of OpenAI’s reported $100M+ training costs. If true, this shatters Silicon Valley’s AI dominance. But is there a catch? Some experts suggest DeepSeek secretly used 50,000 NVIDIA GPUs, bypassing U.S. trade bans.

In this deep dive, we unpack:
✅ How DeepSeek's AI threatens the GPU industry
✅ The Mixture of Experts breakthrough slashing compute costs
✅ The security risks of DeepSeek’s AI models

The AI race just got a lot more interesting. Let me know your thoughts in the comments!

Timestamps:
00:00 - Tech Stocks Crash on Monday
00:27 - The Issue
00:53 - Questioning the statement of $5.6 mn training
01:49 - Revolutionary Mixture of Experts Architecture
02:13 - Benchmark Comparison
02:34 - Own Test with Microsoft Excel World Championship Case
07:44 - Another Test - LLM Chess
09:17 - Security Flaws
10:42 - Final Thoughts

🔗Sources:
Bloomberg $1.0 trn blow: https://www.bloomberg.com/news/articl...
US Trade Restrictions: https://apnews.com/article/technology...
OpenAI API Pricing: https://platform.openai.com/docs/pricing
DeepSeek API Pricing: https://api-docs.deepseek.com/quick_s...
DeepSeek Research Paper (R1): https://arxiv.org/abs/2501.12948
Training Costs (DeepSeek vs OpenAI): https://www.wired.com/story/deepseek-...
Groq Playground: https://console.groq.com/playground
DeepSeek Playground: https://deepseek.com/
Bigger is not Smarter (MS): https://techcrunch.com/2025/01/27/dee...
Satya Nadella Tweet on Jevson Paradox: https://x.com/satyanadella/status/188...
Marc Anderssen Tweet on DS-R1: https://x.com/pmarca/status/188271976...
LLM Chess Benchmark: https://dev.to/maximsaplin/deepseek-r...
Security Violation of DS: https://www.kelacyber.com/blog/deepse...
MEWC Case for Testing: https://fmworldcup.com/product/ges-20...
Quote "DeepSeek hiding 50,000 GPUs": https://www.livemint.com/ai/artificia...
Evidence on DeepSeek's A100/H100 GPUs [Mandarin]: https://mp.weixin.qq.com/s?__biz=Mzk0...
AI Chess Engine (Stockfish): https://stockfishchess.org/

👉For more AI News & Tutorials:

Subscribe:    / @maksimssics  
Connect on LinkedIn:   / maksims-sics  
Let's work together: https://flowbyte.ai/