AI's Initial Impact: Rising Costs Precede Promised Savings
Artificial intelligence is expected to eventually lower costs across various sectors, from software development to customer service and logistics. However, the immediate future involves significant upfront investment, leading to increased expenses. Major tech companies like Amazon and Microsoft are projected to spend around $410 billion on capital expenditures by 2026, with other giants such as Alphabet, Meta, and Oracle contributing to a total nearing $700-800 billion annually. This massive spending fuels demand for physical infrastructure, including land, construction, millions of computer chips, transformers, cooling systems, and power grids, requiring a large workforce of engineers, technicians, and construction workers.
The surge in demand for electricity, driven largely by data centers, is straining existing power grids. In the US, electricity consumption is projected to grow by approximately 2% in both 2026 and 2027, with data centers potentially accounting for half of the increased demand by 2030. The expansion of power grids and the manufacturing of essential components like transformers are lagging behind this demand, causing prices to rise. Electricity and water have become critical bottlenecks for AI infrastructure development.
Furthermore, the demand for copper, a key component in cables, transformers, and cooling systems, is also increasing, impacting its price. The lengthy process of opening new mines contrasts sharply with the rapid deployment of data centers, exacerbating supply-demand imbalances. This initial phase of AI development is characterized by a demand shock, leading to inflation before productivity gains are realized.
Central banks are monitoring this situation, as the initial phase of AI implementation is expected to boost economic activity and inflation, while the productivity benefits are anticipated to materialize later. Long-term projections suggest AI could add between 0.1 to 0.8 percentage points to global potential growth. Locally, the cost of infrastructure upgrades for data centers may be passed on to consumers through general electricity tariffs, prompting some US states to demand consumption commitments from data farms.
In Israel, data center server farm connection requests have reached approximately 27,000 megawatts, significantly exceeding the country's average consumption of around 9,000 megawatts. While 1,500 megawatts have already been approved for connection, this demand highlights the potential for AI to increase living costs before delivering savings.
Looking ahead, increased investment in hardware and infrastructure is expected to eventually lower computing costs and improve efficiency. As AI tools become more integrated into business processes, leading to faster document preparation, credit assessments, and predictive maintenance, the cost per unit of output will decrease. Businesses will likely pass some of these savings on to consumers through lower prices. The labor market may also see a moderation in wage increases in certain professions as AI tools become more commonplace, although workers with AI skills currently command a premium.
