DeepSeek Crashed Energy Stocks. Here’s Why It Shouldn’t Have.
1 min read
Summary
In 2024, energy stocks soared and there were a flurry of deals as companies predicted a significant growth in demand for energy to power data centres.
However, along came DeepSeek’s R1, which claimed it had trained its new AI model with fewer chips and less computing power, sending energy stocks tumbling and causing a ripple of concern through the industry.
But what impact will DeepSeek really have on future energy demands, and should companies be rethinking their plans to scale up their energy capacity?
Training an AI model is different to using it, with most computing power and energy consumed in the training phase, with fewer resources deployed when using the model, when it answers queries or generates content.
DeepSeek has created a new “reasoning” model which does use more energy during the inference phase, by testing different formulas to get a correct answer, and improving its output.
While these new “reasoning” models may be more energy intensive, they are likely to become increasingly more efficient over time.
Therefore, the overall impact of AI on energy consumption looks set to increase significantly.
