A Value Investor’s Thoughts On DeepSeek And The Next Phase Of AI

DeepSeek, a Chinese artificial intelligence research lab, recently introduced DeepSeek-V3, a powerful Mixture-of-Experts (MoE) language model. With 671 billion total parameters and groundbreaking efficiency, it rivals closed-source models while demanding significantly less computing power. More recently, DeepSeek-R1, an open-source reasoning model, captured attention for its ability to autonomously generate complex thought chains, purportedly reducing computing power consumption by up to 95%.

Instead of locking away this innovation, DeepSeek posted R1 on GitHub, allowing researchers and developers to explore its capabilities freely. While this openness sparked enthusiasm in tech circles, the stock market had a different reaction—AI-related stocks tumbled on fears that lower compute costs could disrupt investment theses.

Overreaction In AI Stocks

Despite the excitement surrounding DeepSeek, I argue the market’s swift reaction overlooked key realities:

1. Development costs are still significant. While DeepSeek suggests substantial efficiency gains, its claims do not account for the massive prior research, ablation studies and infrastructure investments required to bring such models to fruition. Training AI models is an expensive and resource-intensive process, and savings on inference alone do not render existing models obsolete.
2. Demand for AI hardware remains strong. Even with efficiency improvements, scaling AI workloads remains constrained by supply chains, power generation limitations and human capital. A single data center can consume as much energy as 50,000-plus homes, and there are persistent shortages in critical semiconductor components. The idea that AI infrastructure spending is suddenly unnecessary is misguided.
3. Not every AI model will be commercially viable. While AI technology is advancing rapidly, monetization remains a challenge. Many AI models, even if technically impressive, will not generate meaningful revenue if customers aren’t willing to pay for their incremental benefits. Investors must differentiate between companies with scalable, profitable applications and those pursuing AI for its own sake.

Looking Ahead

Despite the excitement surrounding DeepSeek, I think the market’s swift reaction overlooked key realities:

AI continues to drive technological advancements, but not all companies will benefit equally. The most successful firms will likely be those that can effectively harness AI by developing proprietary models, integrating AI into enterprise solutions, or building application-layer technologies that address real-world challenges.

Companies that secure reliable computing resources and optimize efficiency in a power-constrained environment ought to gain a competitive edge, particularly as data center demands grow. Additionally, firms that can navigate evolving regulatory landscapes and competitive pressures—especially in response to open-source developments like DeepSeek-R1—will be better positioned for long-term growth.

While AI stocks may experience further turbulence, we continue to believe that opportunities are still plentiful for well-capitalized leaders in semiconductors, cloud computing and enterprise AI solutions over the long term.

For more on my AI thinking, including specific stocks our team likes, check out our latest Special Report.

Disclosure: Please note that shares of the stocks mentioned are owned by asset management clients of Kovitz Investment Group Partners, LLC, a SEC registered investment adviser. For a list of stock recommendations like these made in The Prudent Speculator, visit theprudentspeculator.com.