In a development that has sent shockwaves through the artificial intelligence industry, Chinese startup DeepSeek has achieved what many thought impossible: creating a world-class AI model for less than $6 million, a fraction of the hundreds of millions typically spent on computing power alone by major players like OpenAI (ChatGPT) and Anthropic (Claude).
THE MARKET IMPACT
This development has raised concerns about the future of established players in the AI industry. Nvidia, whose high-margin GPU business has been central to the AI boom, faces particular scrutiny. Shares of chipmaker Nvidia have lost $730 billion in value since November 2024. As the news of DeepSeek’s success broke, NVDA shares were down more than -17% today.
Shares of Nvidia are part of the “Magnificent 7,” a group that makes up roughly 33% of the S&P 500’s value. Nvidia alone is 6.5% of the S&P 500’s total market capitalization, so the dramatic selloff on Monday had spillover effects for the broader market.
DAVID VS. GOLIATH: HOW DEEPSEEK STACKS UP AGAINST AI CHAT GIANTS
Deepseek spent only $50 million on 2,048 Nvidia H800s (not even H100s) and only $5.576 million in training costs on its DeepSeek V3 model. The largest AI companies in the world have spent greater than $600 billion, including $325 billion slated to be spent in 2025 capex alone.
The company's latest offering has garnered praise from industry leaders including Microsoft CEO Satya Nadella and venture capital’s Marc Andreessen, with the latter calling it “one of the most astonishing and impressive breakthroughs I have ever seen.”
- Deepseek V3’s training uses 30% less memory
- It took Meta 11 times longer to train its Llama 3.1 model than Deepseek took to train V3 (achieved 14.8 trillion token pre-training in just 2.788 million GPU hours vs Meta’s Llama 3.1 model at 30.8 million hours)
- DeepSeek is about 53 times cheaper for API access relative to OpenAI’s ChatGPT (at $0.14 per million tokens vs OpenAI's $7.50 rate for the same amount of data
WHY DEEPSEEK MATTERS TO AI STOCKS & THE BROADER MARKET
With the launch of DeepSeek’s R1 model on January 20, 2025, the company’s app reached 1.6 million downloads by January 25. The DeepSeek app rose to the #1 spot in iPhone’s app stores across the U.S., UK, and China.
The demonstration that world-class AI models can be built with fewer, less powerful chips challenges the assumption that cutting-edge AI development requires massive investments in premium hardware.
The breakthrough has particular significance given it comes just days after the announcement of the ambitious $500 billion Stargate Project, a US-based AI infrastructure initiative backed by OpenAI, SoftBank, and Oracle. Just last week, President Donald Trump called the Stargate Project "the largest AI infrastructure project in history.”
HOW DOES THIS IMPACT YOUR PORTFOLIO?
The quick takeaway for investors? As Wall Street has increasingly piled into the largest tech stocks via the Magnificent 7, it’s become more important than ever to remain disciplined and diversified.
That’s our bread and butter at Hedgeye. We provide a go-anywhere Macro strategy so you can protect and grow your portfolio.
"There are a lot of other things to buy,” Hedgeye CEO Keith McCullough told viewers of The Macro Show this morning. “There should be other opportunities born out of this," McCullough explains. "Under no certain terms do I have to be long (Nvidia CEO) Jensen Huang at peak margins and peak demand."
What do you need to know to protect and grow your portfolio?
REQUIRED READING (& VIEWING)
- You can read Global Tech analyst Felix Wang’s “Deepseek Black Swan” take on NVDA here.
- You need to watch this from The Macro Show, “Pros Don't Panic: Keith McCullough on QQQ, NVDA”