Quick Thoughts: DeepSeek +1 day

The introduction of DeepSeek’s R1, a Chinese-based artificial intelligence (AI) reasoning model, represents a significant advancement in the AI industry. The market’s reaction was swift, with the tech sector declining amid concern that this less-expensive model might disrupt the more established AI players. To learn more about DeepSeek and the implications, I spoke with Franklin Equity Group CIO Jonathan Curtis, an expert on AI and its implications for investors. Here, I share highlights of our conversation.  

What is DeepSeek R1?

By integrating three well-known efficiency techniques, R1 has reportedly achieved substantial improvements in large language model (LLM) information processing efficiency (“compute efficiency”) at dramatically lower cost. This development has broad implications for the future of AI. Notably, R1’s open-source nature means its methods are available to everyone and replicable. Leading AI labs such as Meta, OpenAI, and others are expected to adopt similar approaches to remain competitive. While this advancement marks progress in the language space, the journey toward artificial general intelligence (AGI) remains challenging, requiring breakthroughs in multi-modal capabilities, generalized knowledge and physical-world modeling—areas that demand substantial investments in model training.  

Disruptive near-term implications

• We believe recently announced investment plans by large AI hyperscalers, such as Meta’s and OpenAI’s significant Stargate initiative, already reflect an industry-wide anticipation of efficiency advancements like those offered by DeepSeek with R1.
• R1 is unlikely to deter major labs and hyperscalers from pursuing more powerful frontier AI models.
• Reduced costs associated with this DeepSeek’s R1 model will likely increase developer activity and inference compute demand.
• Investors should expect heightened volatility in the AI complex as markets adjust to potentially shifting compute demand dynamics.
• Infrastructure providers may face increased uncertainty.
• Upcoming earnings calls are expected to provide valuable insights into how leading AI companies plan to adapt to these changes, particularly regarding shifts from pre-training (building models) to post-training (using models) compute optimization and the level of capital expenditures required to support both.
   

Positive in the long term

• R1’s efficiency gains are expected to drive greater demand for AI infrastructure and broader adoption of AI technologies.
• Historical parallels, such as the transformative impact of Dense Wavelength Division Multiplexing (DWDM) in the optical transport space—which significantly increased fiber bandwidth, lowering bandwidth costs—suggest that efficiency improvements can catalyze disruptive innovation.
• Just as DWDM paved the way for next-generation internet-based services like social media and Web 2.0, R1’s reduced compute costs could enable accelerated application of AI in software applications and other real-world use cases.
   

Summary

DeepSeek’s R1 highlights both near-term volatility and long-term growth opportunities in the application of AI across the economy. We believe investors should monitor hyperscaler capex trends and earnings updates for clues on how companies are responding to these efficiency advancements. Ultimately, the reduced cost of creating intelligence through innovations like those built into R1 positions the AI industry for a new phase of transformative growth, shifting the focus towards model application. In our view, this is just part of a long journey in which AI transforms our economy. 

Special thanks to Jonathan Curtis, CIO-Franklin Equity Group, for sharing his expertise and knowledge on this topic.