Commentary / Japan

Cheap power the key to AI-based business

by Takamitsu Sawa

It is considered inevitable that demand for electric power will increase dramatically as artificial intelligence replaces human intellectual labor with the progress of the Fourth Industrial Revolution. Human brains are extremely energy-efficient. When a person thinks in a concentrated manner, his or her brain consumes a mere 21 watts of electricity. But AI doing the same degree of intensive thinking requires over 10,000 times more electricity.

If that is the case, the international competitiveness of businesses will depend on factors concerning the supply and cost of electricity in their home country. How, then, does Japan stand with regard to power supply and cost?

Electricity rates in 2018 for industries were higher in Japan than in any other OECD member nations except Italy, and about three times higher than in Norway, the lowest among OECD countries. In countries like Canada and Norway, which are rich in water resources, the cost of generating power is far lower than in nations that rely mainly on thermal and nuclear power.

Based on its own quantitative assessment showing that the unit cost of nuclear power is the second-lowest after coal-fired power generation, Japan’s Natural Resources and Energy Agency has long promoted nuclear energy as a national policy, calling it a key, stable source of power that will help ensure an inexpensive supply of electricity and serve as a trump card toward combating climate change.

But since the catastrophe at the Fukushima No. 1 nuclear power plant caused by the March 11, 2011, Great East Japan Earthquake and tsunami, proponents of nuclear power in both the public and private sectors have been silent. The government’s Fifth Basic Energy Plan released in 2018 estimates that by 2030, nuclear power will be generating 20 to 22 percent of the nation’s total electricity supply. But it does not say a word about the prospect of restarting nuclear power plants idled following the Fukushima disaster, or about plans to build new nuclear plants or add new reactors at existing plants.

The sunk cost, or the cost required for construction, of a nuclear power plant is high. But once completed, its marginal cost to produce 1 kilowatt of electricity is around ¥1 — far lower than that of a thermal power station. Power companies’ desire to reactivate their nuclear reactors is quite understandable.

At this moment, however, no power company appears eager to build a new nuclear plant or add new reactors to existing facilities. That is because the cost of building a 1.2 million kilowatt-class nuclear plant has shot up about three times since the Nuclear Regulation Authority, created as an extra-ministerial bureau of the Environment Ministry to replace the Ministry of Economy, Trade and Industry’s Nuclear and Industrial Safety Agency, made the safety regulations for such plants much more stringent in the aftermath of the Fukushima fiasco.

As the increase in demand for power remains sluggish, private sector power companies apparently do not see any financial sense in investing in the construction of a new nuclear plant with an inherent risk of severe accident.

Following the Fukushima disaster, electricity charges in Japan rose by about 20 percent between 2011 and 2014. But rates have since returned to the level of 2009, when nuclear power accounted for roughly 30 percent of the overall supply. Although some idled nuclear plants have been reactivated since 2015, they generated a mere 3 percent of the total in 2018. So it does not appear that restarting the idled nuclear plants has contributed to a 20 percent decline in electricity charges.

The government’s Basic Energy Plan treats renewable energy as a principal source of power generation, estimating that renewables will account for 22 to 24 percent of the nation’s power supply in 2030. But in 2018, renewable sources excluding large-scale hydroelectric plants accounted for only 8 percent of the nation’s power supply. The ratio rises to only 16 percent even if large-scale hydropower plants are included.

Solar power occupies a predominantly large portion of renewable energy sources partly because of preferential treatment under the feed-in tariff system introduced in 2012 and the ease of installing photovoltaic panels on the rooftops of houses.

While the feed-in tariff system served to promote the introduction of renewable energy, it led to an increase of nearly ¥10,000 per year to the average household electricity bill. This is because the power companies passed on to consumers the gap between the selling and purchasing prices of power generated by renewable sources as a surcharge for promoting the use of clean energy. In view of the increase in customer bills, the feed-in tariff for solar power was abolished at the end of October, making it uncertain that renewables will be responsible for 22 percent of total power generation by 2030, as called for in the Basic Energy Plan.

The key hardware that supports the Fourth Industrial Revolution is the supercomputer, which consumes huge amounts of electricity. The Kei, which was lauded as Japan’s fastest computer from 2012 to 2019, reportedly uses the same amount of electricity as some 30,000 households combined. AI consumes a lot of power in the process of deep learning.

Take for example AlphaGo, an AI program that plays the board game go. In 2015, it became the first such program to defeat a Japanese professional player. It first learns go’s rules and joseki (studied sequences of moves whose results are considered balanced for both players), memorizes the records of all matches played in the past and then improves its skills by playing tens of millions of games against itself. In playing a match, AlphaGo uses about 1,000 servers to pursue the most appropriate moves based on its accumulated knowledge and experience. The servers for AlphaGo are said to consume 1.25 million kWh of electricity.

In a fully self-driving automobile, a server takes charge of steering, accelerating and braking on the basis of data on the driving environment collected by car-mounted cameras, sensors and radar that is transmitted through the cloud system. While the car itself uses a minimal amount of electricity, the server uses a lot.

In short, an AI program uses a huge amount of power in the process of deep learning. It operates by communicating via the cloud with servers at a data center. While in operation, it continues to guzzle electricity. Just as manufacturers in advanced industrialized countries moved their production bases to developing nations in East Asia in search of low-cost labor, data centers that require large volumes of electricity are bound to be located in countries that offer inexpensive power rates.

Faced with high domestic cost of electricity, Japanese businesses will import massive volumes of services known as cloud computing from data centers located in countries with lower electricity charges. This will inevitably erode their international competitiveness. It is imperative for Japan to take steps to propel structural reforms of the power industry in order to at least halve the domestic electricity charges from the current level.

Takamitsu Sawa is vice director of the International Institute for Advanced Studies in Kizugawa, Kyoto Prefecture.