The cost of powering intelligence

The cost of powering intelligence | Daily Tribune

We know though, that given current technology, this is unlikely. Powering intelligence is one of the key puzzles that AI builders are trying to solve.

Creating a world that runs largely on AI is not without energy costs. Computer chips and semiconductors that will process all the computations when we use AI platforms for content generation and automation using agents and bots use 3.5 to 7.0 times more power than non-AI computer chips. This is why data centers being operated and planned by AI platforms are even more massive than before and are called hyperscalers.

Hyperscalers are cloud service providers that offer computing, networking, and data storage services on a global scale. Because of their size and growth, data centers of this category require more power (and capacity is measured in mega or gigawatts), and by transitivity, more heat, which means an equivalent cooling capacity to manage this heat.

Because of their massive size, hyperscalers require a significant amount of water as a coolant, which in some areas raises a sustainability and water stress issue. Based on an AI query, a typical hyperscaler can consume water equivalent to one and a half Olympic swimming pools per day.

It is said that one of the value propositions of the SpaceX IPO of Elon Musk is that with a successful reusable rocket platform, they can build data centers in space. These “space” data centers would be powered by direct sunlight, which will not have the disadvantage of earth-based solar generation, which is affected by cloud cover and weather. In effect, this is “free” and abundant energy for a hyperscaler.

The gold rush that is AI is also a gold rush to more reliable power. It is critical for the Philippines to consider the strategic need for more power generation. We are no longer just building for residential, industrial, and commercial electricity demand. We need to consider incorporating power generation for technology — our technology.

Much of the data center capacity being built are in the US and China, while other countries are also in a rush to build their own. It is not just a matter of more power but also a resilient power system to ensure that computers and servers are working 24/7 and not affected by crashes due to blackouts.

While we need to incorporate AI requirements into our national strategy, we must also recognize that power efficiency is its Achilles Heel and the “Excalibur” of the other AI or actual (human) intelligence.

Humans are still the most power-efficient source of intelligence. We can run mathematical computations quickly with a cost of only 12-20 watts of power to our brains. At the current level of technology, humans have a natural competitive advantage.

Another AI-based movie franchise, The Matrix, imagined a world where solar energy is no longer available and the machines resorted to farming humans to serve as batteries. Hypothetically, a billion humans can only generate 100 gigawatts of heat energy, which, even at our current world population, is not enough to power what the world needs. The current total installed generating capacity across the globe is close to 9,000 gigawatts of energy.

Our national strategy must balance the need to develop more power generation capacity sustainably with investments in human capital and food security. Furthermore, improving human intelligence through massive educational reform and research has more strategic value in the long run.

At the end of the day, AI requires constant investments in power plants worth hundreds of billions of dollars, while all we humans need is a juicy cheeseburger or a fully loaded rice meal to keep going.