Introduction

The United States Department of Defense in 2013 put out a net assessment briefing to Congress regarding a concept of warfare that was developed by the Chinese PLA (People’s Liberation Army) titled the “Three Warfares”. This briefing described a non-kinetic system of warfare designed to employ psychological, media and legal means to demoralize, degrade and dissuade an adversary from pursuing actions contrary to China’s interests. This framework has been greatly advanced by the Chinese over the last decade with AI being added as another tool the Chinese can use to cause disruption to perceived adversaries. We believe that China welcomed the announcement of Deepseek V3 as an AI disruptor and made sure to emphasize the low cost of model training against our U.S based AI companies. The news announcement sent shocks through the sector and caused NVIDA, who supplies the most powerful graphics cards and processors for AI training down 17% during the trading session and left a lot of technical damage to the broader markets in its wake. This is the type of psychological and information warfare operation that is meant to demoralize and shock our country and capital markets system. That being said, the market was able to recover relatively quickly from the opening salvo and stabilize, but we expect that as tensions rise between the United States and China that overall market volatility is likely to rise as well. Investors need to be mindful of the risks of being overly concentrated in a given position or sector, even if it may be leading the market in terms of performance. 

As the artificial intelligence (AI) arms race continues to accelerate we expect these types of announcements to be used strategically to incite volatility and shock amongst US and global investors.  The truth that carries weight however is that within the artificial intelligence ecosystem, DeepSeek AI represents a new formidable player and has broader implications for the sector as a whole.  The technology represents a shift in AI architecture, model training, and hardware utilization. This article examines some of the underlying operational differences between DeepSeek AI and its competitors, the timeline of Deepseek’s rise and some of the core AI mechanics, hardware choices and capital expenditure amongst the current large players in the market.  Furthermore, we assess how DeepSeek AI’s innovations could bolster the broader AI ecosystem and create a net-positive investment outcome for U.S. AI companies and investors.

The Rise of DeepSeek AI: A Timeline of Breakthroughs

DeepSeek AI's rise has been rapid, driven by strategic investments and an aggressive innovation cycle:

• May 2023: Chinese hedge fund High-Flyer establishes DeepSeek AI as a dedicated AI research entity.

• November 2023: DeepSeek Coder, an open-source code-focused AI model, is released.

• Early 2024: DeepSeek launches a 67-billion-parameter large language model (LLM), intensifying competition with OpenAI and other players.

• May 2024: Introduction of DeepSeek-V2, notable for its cost-effective training methodologies.

• January 2025: DeepSeek-R1, a reasoning model targeting complex scientific problems, is unveiled, solidifying its leadership in efficient AI training.

Despite its rapid ascent, DeepSeek AI has achieved its milestones with significantly lower costs than its competitors. The company claims it trained its LLM for under $6 million, a stark contrast to OpenAI’s reported $100 million investment in training GPT-4. This cost efficiency is largely attributed to model distillation, where smaller models learn from larger ones, reducing computational overhead and making AI training more accessible.

AI Infrastructure: Comparing DeepSeek AI, OpenAI, Microsoft, and Meta

DeepSeek AI has disrupted conventional AI hardware reliance by integrating neuromorphic computing and quantum-inspired AI processors, reducing computational costs by up to 50%. DeepSeek AI’s approach reduces dependence on NVIDIA’s dominance in the AI chip market, signaling broader implications for semiconductor companies. As it Combines standard transformer architectures with an additional retrieval mechanism that enables the model to access and integrate relevant information from external sources dynamically. The advantages to this are increased efficiency in specific tasks while reducing hallucination in AI responses. An AI hallucination is the generation of false or misleading information and is often attributed to limitations in model training data and processing errors. 

This is different to OpenAI’s Chat GPT, which uses large scale reinforcement learning that is then tuned by human feedback to optimize responses and increase alignment to topics the user intended. This is seen as a more robust method, as it produces higher quality, human-like outputs. The disadvantage though is the human capital required to label data and provide feedback. Each of the above listed models has their distinct drawbacks and advantages as they diverge from each other, becoming more robust, specialized or efficient with each iteration. There is an argument to be made, that no one AI model or development will become the sole dominant player in the AI race.  

Capital Cost Analysis: Investment Requirements & Cost Efficiency

While DeepSeek AI has managed to undercut its competitors in capital expenditures, it remains far from a “cheap” player in AI development. The revised estimates place DeepSeek's total training costs at around $63 million, significantly higher than its initial claims of under $6 million.

Developing and deploying AI at scale is inherently capital-intensive. Model training, infrastructure costs, and ongoing optimization require substantial investment, and all major AI players continue to ramp up capital expenditures to stay competitive. While DeepSeek has shown that efficiency gains are possible, AI research and deployment remain an expensive endeavor. For investors, this highlights a long-term bullish trend in AI CAPEX, as companies must continuously invest in cutting-edge infrastructure and research to maintain leadership.

Conclusion - The Bull Case for AI

Despite recent market volatility triggered by DeepSeek's advancements, the long-term outlook for AI remains robust. The ability to develop cost-effective AI models is expected to accelerate the integration of AI across various sectors, enhancing productivity and fostering innovation. Investors, while adjusting to the immediate market disruptions, recognize the transformative potential of AI and continue to view the sector as a cornerstone of future economic growth. For investors, DeepSeek's emergence underscores the importance of a diversified approach to AI investments. While hardware providers have traditionally been focal points, the evolving landscape suggests that software and service-oriented AI companies may offer compelling opportunities. Additionally, the emphasis on cost-efficiency and accessibility in AI development highlights the potential of startups and emerging players who can innovate within constrained resources.