AI Capital Spending to Hit $1 Trillion Globally, Fed Forms Task Force

Large-scale infrastructure construction in the artificial intelligence sector is exerting short-term upward pressure on inflation, while the productivity gains it promises have yet to appear in macroeconomic data. US spending on AI now equals 1.8% of gross domestic product, and that share is estimated to reach 2.8% by 2028. A Census Bureau survey published in May found that 17% to 20% of US businesses reported using AI, with adoption rates significantly higher among large enterprises than smaller firms.

Soaring construction costs have already rippled across multiple sectors. US household electricity prices rose 10.1% in the two years through June, outpacing the 6.3% increase in overall prices over the same period. Server chips used for advanced models are in tight supply, and manufacturers cannot ramp up capacity quickly enough to meet demand. DRAM memory prices are estimated to rise 400% by the end of this year compared with 2024. Consumer price index data show that computer software and accessories costs have risen 22.9% since June 2024, including a 17.4% increase over the past year.

Internal adoption within companies also faces hurdles. Executives at firms that have deployed AI at scale note that the technology itself is ready, but the difficulty of implementing it across large organisations has been underestimated. For retailers, using AI to forecast product sales is far more complex than deploying chatbots; changing employee behaviour and building trust in the models are the main challenges. Data show that heavy AI users deploy the technology at eight times the density of average firms, up from a two-fold gap just three months ago. The application gap between frontier and mainstream companies is widening rapidly.

Economists use the term "bottleneck tasks" to describe jobs that are difficult to automate. AI excels at routine work such as reading radiology images, but most jobs comprise multiple tasks, some of which resist automation. In 2016, technology experts predicted radiologists would become redundant within five to ten years, yet their numbers have continued to grow because AI tools automate only a small portion of a radiologist's work, while tasks such as patient communication and collaboration with colleagues remain bottleneck tasks.

The Federal Reserve is divided over AI's economic impact. The Fed chair said in November that AI would be an important disinflationary force, boosting productivity and competitiveness. However, at the July rate-setting meeting, officials voted to hold rates in a range of 3.5% to 3.75%, with some publicly expressing concern that the economy may need to be restrained to curtail AI-driven price increases. The Minneapolis Fed president, explaining his dissent in favour of higher rates, said massive data-centre investment adds a new demand factor to US high inflation. The Fed chair acknowledged in July that AI spending is laying the groundwork for future growth, but "the exact timing and scale of the supply-side effects remain difficult to predict".