Latest News

Japan, a country with limited resources, was bracing itself for a surge in AI energy. DeepSeek increased the stakes.

The rise of Chinese startup DeepSeek, which has shocked the world by upending conventional wisdom about the future energy demands of the AI sector, comes just weeks after Japan announced details of its landmark energy plan designed in part to keep pace with the expected AI surge.

Tokyo only recently re-evaluated its long-held predictions that its electricity consumption would decline with the ageing of its population and began to account for increased AI-driven usage by data centres and chip makers.

The government released a new document in late December.

A draft

The basic energy plan of the government, which is a major document that's reviewed every three years, projected an increase in electricity production between 10-20% b 2040, and cited these factors.

Analysts say that while Tokyo is unlikely to rush to revise its forecasts, DeepSeek models' apparent leaner designs have triggered an extensive rethinking of AI energy requirements. The world's resource-poorest major economy, they argue, would be remiss if it ignored this.

Andrew DeWit is a professor at Rikkyo's School of Economic Policy Studies, Tokyo.

DeepSeek launched last week a free AI Assistant that, according to the company, uses less data and costs a fraction of what other services do. It had surpassed its U.S. competitor ChatGPT by Monday in the number of downloads on Apple's App store, which triggered a selloff across tech stocks.

Investors weighed DeepSeek models that appeared to be more energy efficient, and their stock prices fell.

As analysts analyze DeepSeek's impact, a new view is emerging: its success could lower the entry barrier in a sector dominated largely by Silicon Valley giants. It may also catalyze a higher demand for electricity from new AI-based entrants.

It is a serious problem for Japan. The country produces only 13% of the energy it needs domestically, the second-lowest ratio among all 38 OECD nations, behind only Luxembourg.

If AI is cheaper to develop than expected, it will accelerate mass adoption rather than slowing. It would actually increase the demand for power in the country, said Yuriy Huber, CEO of K.K. Yuri Group is an energy consulting and research firm based in Tokyo.

The Japanese officials took their time in adjusting power demand estimates, even though AI boom was evident two years ago. He said that he expected them to closely monitor the new developments.

The Trade Ministry, which is responsible for the long-term planning of energy in the country, did not respond immediately to a comment request.

Unnamed senior officials at the ministry said that despite the impressive nature of DeepSeek, it did not improve existing systems. He thought the markets had overreacted.

Tepco, Japan’s largest power company by revenue, said that it closely monitored DeepSeek’s potential impact on electricity demand but had yet to assess the full effects.

'BITTER LEARNING'

Grid monitors in Japan have predicted for years that future electricity demand will decline due to energy-efficient equipment adoption and a shrinking populace.

In 2024 it revised its outlook, reflecting an increase in overall demand, driven primarily by a new 5,14 million kWh power demand expected from data centres and chips plants by 2034.

The government has also stated that AI-related energy requirements are a reason for restarting nuclear reactors. This is a sensitive topic in a nation that experienced one of the worst nuclear disasters in the world when the Fukushima nuclear plant was crippled in 2011 by an earthquake and tsunami.

Mika Ohbayashi of the Renewable Energy Institute, Tokyo, stated that DeepSeek was "clear evidence" that AI could become more efficient, and require less power.

She criticised Japanese officials for linking AI energy to nuclear power and said that the government should instead do more to promote renewable energy.

Analysts say that DeepSeek technology must be thoroughly tested to determine if it will lead to a future with more or less AI energy demand.

DeWit from Rikkyo university said that Japan had made mistakes in its energy technology preparations in the past. He cited the years leading up to the burst of its economic bubble in the late 1980s.

"Japan, at the time, was the leader in the chip market. They thought they would be number one. So they developed the power system. As the bubble burst in the 1990s, the power demand didn't materialize.

"They have learned bitter lessons." "Policy makers should take this very seriously," he said. (Reporting and editing by Kim Coghill; Additional reporting and reporting by Tim Kelly, Yuka Obayashi and John Geddie)

(source: Reuters)