Sacred Valley Casually I: Artificial Intelligence, $320bn, and Four Cities of Lima

What do AI, $320bn, and Four Cities the Size of Lima Have in Common?

$320bn is the amount that the four leading AI applications companies (Amazon, Microsoft, Google, and Meta) are planning to spend on AI-related capital expenditures this year—19% more than Peru's GDP in 2023 ($270bn) and about 1.2% of the United States’ 2023 GDP ($27.4tr).

These companies consumed a total of 72 TWh of energy in 2021—including the energy required to operate their AI projects—enough to power four cities the size of Lima. That consumption is only growing.

Tech firms are ramping up their investments in artificial intelligence, with the combined capital expenditures of the leading four (Amazon, Microsoft, Google, and Meta) projected to reach $320 billion in 2025—a 30% increase from 2024 and double what they spent just two years ago. This scale of investment rivals the GDP of entire nations, such as Colombia ($360 billion in 2023) or Chile ($330 billion in 2023).

These investments are set to transform a wide range of industries. Sectors like healthcare, finance, defense, transportation, and logistics will undergo major shifts through automation, advanced analytics, and AI-driven business models.

• In healthcare, AI will revolutionize diagnostics, personalized medicine, and drug production.
• In finance, it will enhance algorithmic trading, expand access to investment, grow the distribution of wealth management products, and strengthen fraud detection.
• Economically, countries at the forefront of AI innovation will shape global trade, much like oil-rich nations have dominated energy markets in the past.
• In defense, AI will shift global power dynamics, giving technologically advanced nations a strategic edge.

Beyond the brainpower needed to develop AI, it also requires sizeable infrastructure—especially data centers that run complex machine-learning models. These facilities consume vast amounts of energy, with some large data centers using as much electricity as small cities.

The combined energy consumption of Amazon, Microsoft, Google, and Meta’s data centers reached approximately 72 terawatt-hours (TWh) in 2021—enough to power four cities the size of Lima or 1.4 times the annual electricity consumption of New York or Paris (50-53 TWh).

How Much Energy Do AI Applications Consume?

Training GPT-3, an earlier model, required about 1,287 MWh—equivalent to Cusco’s annual energy consumption.

Once deployed, large-scale AI operations serving millions of queries daily might use between 87,600 and 438,000 MWh annually. More complex AI tasks, such as simulating protein folding for drug discovery or powering fully autonomous vehicle fleets, would demand even more computational power and energy.

Whoever finds a way to generate that amount of power efficiently, safely, and sustainably will secure a seat at the AI revolution table.

Sources:
Business Insider, "The CapEx of Meta, Google, Microsoft, and Amazon," 2024
World Bank, "GDP Data for the United States, Peru, Colombia, and Chile," 2024
CEIC Data, "Electricity Consumption in Lima and Cusco," 2023
IEA, "Data Centres and Data Transmission Networks," 2021
Estimated energy consumption for AI training, based on industry data

Photo: Reuters.com