Does DeepSeek R1 can cause an AI bubble ?

Does DeepSeek R1 will cause an AI Bubble


Hello,


DeepSeek, a Chinese AI startup, has recently introduced its R1 model, which has garnered significant attention in the AI community. The DeepSeek R1 model is designed to be cost-effective, operating with less advanced hardware and offering services at a fraction of the cost of its competitors. Specifically, DeepSeek's R1 model charges approximately $0.55 per million tokens, compared to OpenAI's $15 per million tokens, representing a cost reduction of about 96%. ([businessinsider.com](https://www.businessinsider.com/meta-yann-lecun-ai-scientist-deepseek-markets-reaction-inference-2025-1?utm_source=chatgpt.com))

In terms of performance, DeepSeek R1 is competitive with OpenAI's o1 model. On the Massive Multitask Language Understanding (MMLU) benchmark, which evaluates knowledge across 57 subjects, OpenAI o1-1217 scores 91.8%, while DeepSeek R1 scores 90.8%. However, in the Graduate-Level Google-Proof Q&A (GPQA) benchmark, which measures the ability to answer general-purpose knowledge questions, OpenAI o1-1217 achieves 75.7%, surpassing DeepSeek R1's 71.5%. ([datacamp.com](https://www.datacamp.com/blog/deepseek-r1?utm_source=chatgpt.com))

When compared to OpenAI's more advanced o3 model, DeepSeek R1 falls short in performance. The o3 model outperforms other AI models in key benchmarks, including a leading Codeforces Elo rating of 2727 and 96.7% accuracy on the AIME test. It also excels in the GPQA-Diamond Benchmark with 87.7%, surpassing competitors like DeepSeek R1, V3, and OpenAI o1. ([analyticsvidhya.com](https://www.analyticsvidhya.com/blog/2025/01/openai-o3-vs-competitors-performance-and-applications/?utm_source=chatgpt.com))

The significant cost advantage of DeepSeek R1 has disrupted the AI industry, leading to a substantial drop in Nvidia's market value. Microsoft has quickly adapted by testing and deploying R1 on Azure and GitHub, aiming to leverage R1 to reduce AI costs and increase consumption. ([theverge.com](https://www.theverge.com/notepad-microsoft-newsletter/603170/microsoft-deepseek-ai-azure-notepad?utm_source=chatgpt.com))

However, this development raises concerns for DeepSeek. The open-source nature of their model allows competitors like OpenAI to study and potentially adopt their cost-effective optimizations. This could enable OpenAI to enhance their models rapidly, potentially diminishing DeepSeek's competitive advantage. Moreover, the open-source nature of DeepSeek's models allows for easier and cheaper integration into various applications, challenging the dominance of expensive, closed models from companies like OpenAI, Google, and Microsoft. ([vox.com](https://www.vox.com/technology/397330/deepseek-openai-chatgpt-gemini-nvidia-china?utm_source=chatgpt.com))

So, while DeepSeek R1 offers a cost-effective alternative in the AI market and is competitive with OpenAI's o1 model, it lags behind the more advanced o3 model. The open-source strategy, while promoting accessibility, may inadvertently aid competitors in improving their models, potentially undermining DeepSeek's position in the market.


Other than that, DeepSeek R1 alone is unlikely to cause an **AI bubble** in the traditional sense—where excessive speculation leads to an overinflated market that eventually crashes. However, its **drastic cost reduction** could accelerate the AI industry's transformation and put downward pressure on the profitability of AI service providers.

Here’s why DeepSeek R1 is **not likely** to create an AI bubble:

1. **Lower Costs Do Not Inflate Expectations**
- AI bubbles form when there is **irrational investment** driven by hype, rather than sustainable economic fundamentals.
- DeepSeek R1 lowers inference costs significantly, but it does not make exaggerated claims about AI capabilities—it's still weaker than OpenAI's o3.

2. **Open-Source Models Reduce the Risk of Overvaluation**
- Closed-source models from OpenAI, Google, and Anthropic can be valued based on exclusivity. DeepSeek's open-source nature means no single entity controls it, **reducing monopolistic hype**.

3. **Market Disruption vs. Speculative Frenzy**
- The impact of DeepSeek R1 is more about **forcing competitors to adapt** rather than causing unsustainable AI speculation.
- Its cost reduction is real, tangible, and leads to **cheaper AI access**, but this is a shift in pricing strategy rather than a sudden hype-driven surge in investments.

### What Could Still Create an AI Bubble?
While DeepSeek R1 itself is not likely to create a bubble, **excessive investment in AI startups** that promise unrealistic breakthroughs—without real technological progress—could still fuel an AI bubble. If companies over-invest in training massive models without considering efficiency, or if investors blindly chase "the next ChatGPT" without understanding real capabilities, a bubble could form.

### Conclusion
DeepSeek R1 **is disruptive**, but it does not fit the pattern of past tech bubbles. Instead of overhyping AI, it **forces cost reductions**, making AI more accessible. The real risk for an AI bubble still comes from **irrational investment and hype**, not from DeepSeek’s pricing model.


Other than that, DeepSeek R1 might **ultimately benefit OpenAI** more than itself. OpenAI can analyze DeepSeek R1’s optimizations and **rapidly integrate** similar cost-saving techniques into its **more advanced o3 model**, making it both **cheaper and stronger**.

### Why OpenAI Can Adapt Rapidly
1. **Open-Source Nature of DeepSeek R1**
- Since DeepSeek R1 is open-source, OpenAI (and others) can **study its optimizations** and integrate them into their own models.
- OpenAI has **more resources, talent, and infrastructure**, allowing them to iterate on these optimizations **much faster** than DeepSeek.

2. **o3 is Already More Powerful**
- OpenAI’s **o3 model is significantly better** in benchmarks than DeepSeek R1.
- If OpenAI can **match DeepSeek’s efficiency optimizations**, it will retain its **performance advantage** while dramatically **lowering costs**.

3. **First-Mover Advantage in Monetization**
- OpenAI already has **enterprise adoption** with companies integrated into its ecosystem (Azure, ChatGPT, Codex, etc.).
- If OpenAI slashes costs using DeepSeek’s methods, **DeepSeek loses its biggest selling point (cost savings)** while OpenAI retains its performance lead.

### The Real Risk for DeepSeek
By open-sourcing its model and exposing its optimizations, DeepSeek may have **given OpenAI a free cost-reduction playbook**.
- OpenAI can use **DeepSeek’s efficiency tricks** while keeping its superior model performance.
- **DeepSeek lacks OpenAI’s scale and customer base**, making it harder to compete once OpenAI adapts.

### Conclusion
DeepSeek R1’s cost-cutting innovation **benefits the AI ecosystem**, but paradoxically, it might **help OpenAI more than it helps China**. OpenAI can absorb these optimizations and **quickly erase DeepSeek’s only advantage (low cost), while still leading in capability**. **This makes DeepSeek’s long-term impact uncertain, as it may have just accelerated OpenAI’s improvements.**


Thank you,
Amine Moulay Ramdane.








Comments

Post a Comment

Popular posts from this blog

Is USA becoming an oligarchy ?

Is USA becoming an oligarchy Hello, So here is my question to the AI (Artificial intelligence) from OpenAI called GPT-4o: So i invite you to look in internet at how a part of people of USA are now saying that USA has become an oligarchy in the Era of president Donald Trump, but i think that technically it is difficult in Democracy and especially in this Era of sophistication of information to become an oligarchy, since every act of Donald Trump presidency is being scrutinized carefully by for example people and by the media and by academia, so in this kind of standard it is difficult to become an oligarchy in USA. And here is what GPT-4o answered me: "Your observation about the challenges of establishing an oligarchy in the United States, especially given the current era of information sophistication and media scrutiny, is insightful. Indeed, the extensive oversight by the public, media, and academia serves as a robust check against the concentration of power....
Read more

About the efficiency of Decentralization

About the efficiency of Decentralization Hello, And here is my question to the AI (Artificial intelligence) from OpenAI that we call GPT-4o: What were the main weaknesses of Japan lack of competitiveness in the technology sector , is it that it was much less specialized, and too much centralized , as opposed to the US way of doing ? And here is what GPT-4o answered me: "Yes, those factors contributed significantly to Japan's challenges in the technology sector. Here's a detailed breakdown: ### 1. **Lack of Specialization** - Japan's technology sector historically focused on **hardware**, particularly in consumer electronics, rather than diversifying into software or platforms. - The U.S. embraced a more specialized approach, with companies focusing on niches such as operating systems (e.g., Microsoft), processors (e.g., Intel), and later software and cloud services (e.g., Google, Amazon, and Apple). - Japanese firms, such as Sony, Panasonic, a...
Read more

About the open-source model vs. the proprietary model

Open,source vs proprietary Hello, I invite you to look at the following new video from a known guy in tech called techlead about DeepSeek AI: DeepSeek AI CRASHES tech stocks, Bitcoin, Memecoins. https://www.youtube.com/watch?v=8ex2HnnJ6I0 And here is my following thoughts and critics about it: And here is my question to the AI (Artificial intelligence) from OpenAI that we call GPT-4o: Can you look in internet at the articles about the new AI from China called DeepSeek R1 , and tell me more what you think ?. And here is what GPT-4o answered me: "DeepSeek, a Chinese artificial intelligence company, has recently introduced its latest model, DeepSeek-R1, which has garnered significant attention in the tech industry. This model is notable for its high performance in reasoning tasks, rivaling leading models like OpenAI's o1, while being developed at a fraction of the cost. DeepSeek achieved this by utilizing less advanced hardware and innovative train...
Read more