DeepSeek’s energy-efficient AI sends energy market reeling: The US power market saw a record one-day drop after China’s DeepSeek launched its rules-bending, energy-efficient open-source AI model, potentially changing the trajectory of worldwide energy demand, Reuters reported Monday.

The better alternative: The Chinese startup claimed DeepSeek V-3 is “comparable” to closed-source AI models like OpenAI’s GPT-4o, yet cost some USD 5.6 mn, the firm said in a technical report. It also only used about 2k Nvidia chips for its model, making its model much more energy-efficient in comparison to those developed by leading Western tech players, which use as many as 16k chips to train their software, The New York Times reported Monday.

Future energy demand modelling could require a revamp: DeepSeek’s rise “calls into question the significant electric demand projections for the US,” with AI accounting for 75% of US energy demand forecast through 2035, Jeffries analysts told Bloomberg on Tuesday. DeepSeek’s offering may consume as little as 10% of the electricity used by leading AI models, which deflated the share values of several energy players whose shares have recently rallies on the back of an AI-powered projected demand boom, including nuclear energy, uranium producers and natural gas pipeline operators, Bloomberg reported separately on Wednesday. For example, the US energy provider stronghouse Constellation Energy — whose shares jumped about 100% in 2024 over rising demand on its nuclear energy from US data centers — saw its share price drop by 20%, Reuters reported on Monday.

AI is not exactly climate-friendly: Google’s greenhouse gas emissions rose 48% within the past five years thanks, in part, to its data centres integrating AI services. Microsoft reported a 30% increase in emissions for the same reason.

That’s why tech players are pursuing nuclear energy: Microsoft opted to go nuclear, inking a 20-year power supply agreement with Constellation Energy, to sustain its AI services. Tech giant Meta is also seeking to power its AI offerings with 4 GW of nuclear energy as early as the 2030s.