DeepSeek vs ChatGPT: What’s the Hype?
When I first saw headlines about Nvidia’s $600 billion market drop, I thought it had to be a mistake. Turns out, it wasn’t. DeepSeek’s launch is disruptive for many reasons.
I think by now, we have all seen the news of DeepSeek and Nvidia. And if you haven’t, this is going to be your TL;DR on what’s happening in the AI space.
So What is DeepSeek?
DeepSeek is ChatGPT on steroids made in China. And you guessed it right, it’s cheaper—built on top of older Nvidia H800 chips1 and cost $6.5 million to build. For reference, this is 1/20th of the cost of ChatGPT, so it’s pretty cheap for a model that is comparable to ChatGPT in speed and accuracy.
How? DeepSeek is using R1 architecture that activates only the necessary parts of the model based on each query and optimizes GPU usage. This makes it faster, cheaper, and more energy-efficient, reducing the need for the high-powered GPUs that have primarily been fueled by Nvidia’s dominance. And before we can catch up on DeepSeek, Alibaba released an AI model that surpasses DeepSeek2.
Is It Worth the Hype?
DeepSeek probably sounds like another tech trend but it’s not. The R1 architecture could reshape the AI space in 3 ways:
Cost Efficiency: DeepSeek has proven that you can now run AI operations without needing heaps of GPUs. That’s a huge win for startups and mid-sized companies with tight budgets.
Open-Source Advantage: Developers can access and build on DeepSeek’s technology, accelerating innovation in the AI space. You can access it on GitHub and start building: DeepSeek R1 GitHub
Environmental Impact: With less power consumption, AI could finally become more sustainable.
How is it different from ChatGPT?
As a user, you may not notice a huge difference depending on your use case. In the backend, there are few key differences.
Architecture: DeepSeek uses a MoE (Mixture of Experts) architecture, where it activates only certain parts of its network depending on the task, making it super efficient. On the other hand, ChatGPT runs on a transformer-based architecture, processing tasks across its whole network.
Customization: Unlike ChatGPT, DeepSeek’s open-source nature enables more customization opportunities from the open-source community.
Response Style: DeepSeek shows it thinking while it processes the query in the backend—longer responses. ChatGPT’s response is condensed and to the point.
Use Cases: ChatGPT is great for more conversation style user agent, while DeepSeek works more for information retrieval and works like a highly optimized search engine.
Cost Efficient: If it takes OpenAI 20 chips to run and train the AI model, it takes DeepSeek 1 chip to run an AI model given R1 architecture—making cost-efficiency the key differentiator.
What This Means for Nvidia and ChatGPT
Nvidia’s market value took a massive $600 billion hit in a single day because DeepSeek disrupted their core business model. Fewer GPUs means less reliance on most advanced chips—which has been the case till now. It signals less reliance on Nvidia’s chips in the future which explains the stock dip.
What about ChatGPT? To be honest, ChatGPT will be fine. If anything, this will push them to innovate more cost-efficient ways of running AI models. Competition is not a zero-sum game. Though DeepSeek’s efficiency is serious competition, I suspect OpenAI will catch up. What really remains to be answered is the open-source nature. Would OpenAI ever consider using the “Open” in its name and become open-source or will it be CloseAI? Okay that was pun intended but you get the point.
DeepSeek has Critics
Despite the innovation, DeepSeek has received some harsh criticism from various news sources, such as this one, warning users to refrain from using DeepSeeK. The primary concern is that it’s developed and owned by China. The critics wonder how the data collected from this app will be used, who will use it and where it will be used (hinting at CCP). Others question the legitimacy of the cost-efficiency claims by DeepSeek and whether R1 architecture is really what it says to be. Regardless of the critics, the innovation can’t be denied. If this innovation was coming from a US based company, would the reaction be different?
What it means for the future of AI
We should expect DeepSeek’s rise to democratize AI where individual developers, smaller to mid-size companies and communities can access the open-source model and build on top of it—ultimately driving smart models and innovations. And finally, cheaper AI systems that are cost effective than what we have today. With Alibaba’s recent claim that its AI model surpass DeepSeek, we are definitely heading into that direction.
🔗 Let’s Connect
I hope you enjoyed reading this! If you want to see more content from me, I regularly post on YouTube, LinkedIn and more social channels at sundaskhalid.com.
✨ What’s New This Week?
There is way too much new stuff since last newsletter, so it’s going to be a laundry list. Ready? I finished my annual performance review at my day job (phewww). Outside of the day job, I did this very cool partnership with Meta AI, hired one new member on my social media team, recorded a podcast and did a live webinar talking about Salary Negotiation Masterclass, and took a day trip to Canada (drove 6 hours, which was tiring 🤯). But most important of all, I have been working out religiously and finally got my abs back (took 4 weeks of dedicated work). Lastly, I published 2025 Data Analyst roadmap on YouTube for everyone who is interested in entering and learning more about the data analytics field 📈