The prior issue of this series discussed the term premium freak-off. Now we have a freak-off of a different sort. The release of the DeepSeek AI chatbot is making markets rethink the Mag7/AI fueled trade and wondering if it has been shot in the head.
The macro implications are enormous, given the gargantuan sums of cash committed to AI in this capex-cycle.
Microsoft: In 2024, Microsoft announced plans to invest $80 billion in 2025 to build AI-enabled data centers, train AI models, and deploy AI and cloud-based applications.
Meta: Mark Zuckerberg stated that the company will invest $60 to $65 billion in AI infrastructure capex.
Amazon: Amazon announced plans to increase its capex in 2025, surpassing the approximately $75 billion allocated in 2024. Amazon's capital spending increased from $48.4 billion in 2023 to an estimated $75 billion in 2024.
Alphabet (Google): Alphabet has also signaled plans to enhance its AI capabilities, spending $13 billion per quarter.
As the public whisper around DeepSeek is old news since December, and likely earlier privately, this is the noise.
The signal is in contextualizing the DeepSeek macro impact using a few simple principles from economics and philosophy.
Jevons’ Paradox
Jevons’ Paradox is an economic observation that increased efficiency in using a resource tends to increase, rather than decrease, its overall consumption.
It was first articulated by the English economist William Stanley Jevons in his 1865 book The Coal Question.
Jevons noticed that improvements in the efficiency of coal-powered steam engines did not reduce the demand for coal; instead, they made coal-powered processes cheaper and more accessible, leading to an increase in coal consumption.
Jevons’ Paradox suggests in the case of DeepSeek's R1 model, the model's ability to run efficiently on less powerful GPUs could lead to a proliferation of AI applications and make AI more accessible at scale.
While the individual applications might use less powerful hardware, the total GPU usage could rise due to increased adoption and new use cases.
We have seen this with energy efficiency and consumption, where improvements in vehicle fuel efficiency lowers the cost of driving, leading people to drive more or purchase larger vehicles.
The same with data and internet usage. As internet technology has become more efficient, the cost of transmitting data has decreased, resulting in massive increases in data consumption (e.g., streaming video or gaming).