The AI industry is reeling from the emergence of Chinese AI company DeepSeek (this link may not work when you visit, as they are battling a cyberattack), which has developed a high-performance large language model (LLM) at a fraction of the cost of Western counterparts. With its latest open-source release under the MIT license, DeepSeek is shaking up the AI landscape, triggering market reactions and fueling speculation about the future of AI competition.

DeepSeek: AI Powerhouse at a Bargain?

DeepSeek website home page seen through a magnifying glass
DeepSeek has taken the AI industry by surprise. Source: @ARAMYAN/Adobe Stock

DeepSeek’s flagship DeepSeek-R1 model reportedly cost just $5.6 million to train, a mere fraction of the $100 million+ budgets of AI giants such as OpenAI, Google, and Anthropic. Despite this, DeepSeek-R1 achieves benchmark results comparable to ChatGPT while costing significantly less to run.

Unlike many AI models that rely on brute-force computation, DeepSeek has optimized training by focusing on reinforcement learning and step-by-step reasoning, improving accuracy without excessive computational overhead. The model breaks down complex problems into smaller steps, similar to how long division is easier when worked through on paper.

Hardware Efficiency and Local Deployment

While U.S. export restrictions have prevented China from accessing Nvidia’s top-tier AI chips (such as the H200 and future B100), DeepSeek has worked around these limitations. Instead of relying on ultra-high-end hardware, the company trained its model using 2,048 Nvidia H800 GPUs, which are less powerful than restricted H100s but still efficient for AI workloads. The H800 is “a modified version of H100 specifically sold in the Chinese market due to export regulations” according to NVIDIA.

Further adding to its disruptive potential, DeepSeek has released distilled versions of its model, making it possible to run on consumer-grade hardware. One such version was recently demonstrated running locally on a Raspberry Pi with an SSD and a Hailo module, achieving 200 tokens per second. See Jeff Geerling pull this off:


The H800s are less powerful than the restricted H100s, but still capable of large-scale AI training. The necessity of using these lower-end AI chips forces Chinese researchers to optimize their training methods, relying more on software efficiency rather than brute-force computation. Despite this, DeepSeek has managed to achieve competitive AI performance, proving that strategic algorithm design can compensate for hardware limitations.

Market Reaction and Industry Impact

DeepSeek’s rapid rise has already caused ripples in global markets. Following its launch, the Nasdaq Composite index plunged nearly 3% in one day, with tech stocks — particularly those of AI-heavy companies such as Nvidia, Microsoft, and Alphabet — suffering losses. The AI industry is now facing the reality of high-quality, low-cost open-source competition that could challenge the revenue models of proprietary AI services.

The Nasdaq Composite took a dive on Monday morning, January 27th. Source: Google.

A Shift Toward Edge AI?

One of the most exciting aspects of DeepSeek’s approach is its potential impact on IoT and edge computing. With lightweight AI models capable of running on low-power hardware, DeepSeek is paving the way for real-time, local AI processing in sectors like energy management, automation, and embedded systems. This could reduce reliance on cloud-based AI services, lower operational costs, and enhance privacy by keeping data processing on-device.

The Future of AI Competition

With DeepSeek proving that high-end AI models can be trained for a fraction of traditional costs, the broader implications are clear:

  • Open-source AI is becoming increasingly competitive.
  • Efficiency and algorithmic improvements may matter more than raw compute power.
  • China’s AI ecosystem is evolving rapidly, despite hardware constraints.

Whether DeepSeek continues to disrupt the market remains to be seen, but one thing is certain: The AI race just got a lot more interesting.

Subscribe
Tag alert: Subscribe to the tag Artificial Intelligence and you will receive an e-mail as soon as a new item about it is published on our website!