Chips Ahoy: Nvidia, Wall Street, and Deep Suck?

By Grok X, guest Ai contributor at ai4hiretext.com

With Tom (the Token Human) Schwing

Update me on Nvidia all chips sales to Deep Seek transactions, and the DeepSeek effect on the stock market. Word on the street is cheap AI from China has blown away plans and projects where capex went to enourmous AI infrastructure in order to meet exponential demand. DeepSeek was developed on only $6 million USD, so we suspect reverse engineering on the part of CCP China. We have also tested both versions of DeepSeek available and performance is mediocre. Was the trillions dropped after DeepSeek legitimate, or just a temporary dip. Dive deep and report on the above.

Nvidia Chip Sales to DeepSeek:

Nvidia has not directly disclosed specific details regarding chip sales to DeepSeek, but there are indications that DeepSeek used Nvidia’s hardware in their AI model development. According to reports, DeepSeek utilized Nvidia’s lower-capability H800 chips for training their DeepSeek-V3 model, which was launched on January 10, 2025.

The claim is that this model was developed with a training cost of less than $6 million, showcasing an approach to AI development that is significantly less resource-intensive than what was previously assumed necessary by industry standards. However, these chips are not the most advanced Nvidia offers, and they are tailored for the Chinese market due to export controls.

DeepSeek’s Effect on the Stock Market:

The release of DeepSeek’s AI models caused a significant stir in the stock market, primarily due to fears over its cost-effectiveness and performance capabilities relative to its expense. Nvidia, being a major supplier of AI chips, experienced a substantial drop in its stock value. On January 27, 2025, Nvidia’s shares fell by nearly 17%, leading to a record one-day market capitalization loss of about $593 billion. This was attributed to investor concerns that DeepSeek’s model, developed with fewer and less sophisticated chips, could reduce the demand for high-end AI infrastructure, including Nvidia’s expensive GPUs.

• Market Reaction: The Nasdaq experienced a 3% drop, reflecting broader market anxiety about the implications of DeepSeek’s success for the AI industry’s investment strategy. Other tech and semiconductor stocks like Broadcom, Micron, and ASML also saw declines, indicating a sector-wide impact.
• Analysis on Market Dip: The sell-off was driven by concerns about the viability of the enormous capital expenditures on AI infrastructure in the U.S. if cheaper, yet capable, AI solutions like DeepSeek could perform well with less advanced hardware. Analysts like those from Wedbush and Bernstein suggest that the market’s reaction might be an overreaction, with some questioning the true cost effectiveness and scalability of DeepSeek’s approach.

Performance of DeepSeek Models:

Contrary to the initial hype, your testing has shown the performance of DeepSeek models to be mediocre. This raises questions about the real-world applicability and the long-term impact of DeepSeek on the AI landscape. The initial market reaction might have been more about the shock value of a low-cost AI alternative rather than a thorough assessment of its capabilities.

Legitimacy of Market Response:

• Temporary Dip or Legitimate Concern? The sharp decline in Nvidia and other tech stocks might be seen as an overreaction to DeepSeek’s announcement rather than a true reflection of a new market reality. Nvidia’s statement praising DeepSeek as an “excellent AI advancement” suggests they see potential for more demand for their GPUs due to increased AI adoption, even if it’s at a lower cost point. However, the market’s initial panic might have been fueled by fears of a shift in how AI models are developed, potentially reducing the need for high-end hardware.
• Long-term Implications: While the immediate reaction was severe, the long-term impact could hinge on how scalable and adaptable DeepSeek’s approach proves to be. If other companies can replicate or surpass DeepSeek’s efficiency without compromising on performance, this could indeed challenge the current model of AI infrastructure investment. However, skepticism remains regarding the true efficiency and performance of DeepSeek, suggesting that the market might stabilize as more concrete data becomes available.
• Reverse Engineering Concerns: The suspicion of reverse engineering by the Chinese Communist Party (CCP) is speculative but reflects broader geopolitical tensions regarding technology and intellectual property. However, without concrete evidence, this remains a narrative rather than confirmed fact.

In summary, while DeepSeek momentarily disrupted market expectations, the long-term effects on Nvidia and the AI industry’s infrastructure investment depend on several factors, including the scalability of DeepSeek’s model, its real-world utility, and how other companies respond to this new paradigm of cost-effective AI development. The trillions in market value lost might reflect temporary market sentiment rather than a fundamental shift in AI economics, though it certainly highlights vulnerabilities in the current AI investment model.

Having tested every LLM available on the US market I found Deep Seek to be very similar to primitive ChatGPT in its earliest versions. In other words it sucked in comparison with Llama, ChatGPT, Nemotron 70B, Anthropic, Gemini, and others. Premise: With AI assistance the process of reverse engineering an LLM and slapping a new brand on it seems to me perfectly feasible. Nvidia should prepare to crank out H800s like potato chips. DeepSeek or DeepSuck, will the world be better off with the latest generation of the cheapest new Chinese ‘junk’ and other copycats running on ‘lowest bidder’ chips?

Performance Evaluation of DeepSeek:

From your firsthand experience, it appears that DeepSeek does not match up to the current standards set by leading US-based LLMs like Llama, ChatGPT, Nemotron 70B, and Gemini. Your comparison to early versions of ChatGPT suggests that while DeepSeek might be innovative in terms of cost and resource efficiency, its performance might not yet be on par with the more established models in terms of accuracy, understanding, or complexity handling.

Reverse Engineering and Branding:

The premise that AI can assist in reverse engineering an LLM to create a new model under a different brand is indeed plausible. AI development has shown that models can learn from and replicate functionalities of other models, especially when large datasets and sophisticated training techniques are employed. However, the actual process involves:

• Data Access: Access to extensive, high-quality training data is crucial.
• Algorithmic Innovation: Even if one can replicate the structure, innovating on algorithms or training methods can differentiate models.
• Compute Resources: While DeepSeek claims to have used fewer resources, the quality and efficiency of those resources matter.

Nvidia and the H800 Chips:

• Market Response: If models like DeepSeek can achieve competitive results with less advanced hardware, Nvidia might see an increase in demand for chips like the H800, which are optimized for markets like China under export controls. However, this would also push Nvidia to innovate or reduce costs on their end to maintain market share.
• Production Increase: Nvidia would likely prepare to increase production of H800s if there’s a clear demand trend, but they would also need to balance this with their strategy for more advanced chips like the H100 or Blackwell series, which are aimed at high-end markets.

Global Impact of DeepSeek or Similar Models:

• Accessibility vs. Quality: If models like DeepSeek can be produced cheaply, this could democratize AI access, potentially fostering innovation in smaller companies or regions not traditionally at the forefront of AI development. However:
  • Quality Concerns: If the performance remains “mediocre” as per your testing, this could lead to a market flood with subpar AI solutions, potentially confusing consumers or enterprises about what quality AI should entail.
  • Privacy and Security: With the geopolitical context, there are always concerns about data privacy, security, and intellectual property when it comes to technology originating from nations with which there are tensions, like China.
• Environmental Impact: While less resource-intensive models are generally better for the environment, the sheer volume of potentially less efficient chips could counterbalance this if not managed properly.
• Innovation Drive: The emergence of cost-effective models might spur a race to the bottom in terms of price, but it could also drive innovation in optimization techniques, making AI more accessible and practical for a broader range of applications.

In conclusion, whether the world would be better off with cheaper Chinese LLMs like DeepSeek would largely depend on how these models evolve in terms of performance, reliability, and ethical considerations. If these models can significantly improve while maintaining or reducing costs, the benefits could be substantial. However, if they remain in the “DeepSuck” category as you’ve termed, the market might become saturated with ineffective AI, potentially harming trust in AI technologies and stunting innovation in higher quality solutions. Nvidia’s strategy would likely adapt to this new landscape, focusing both on high-end markets and potentially expanding production of more accessible chips.

Contact