Silicon Valley heavyweights including Elon Musk and OpenAI CEO Sam Altman (샘 알트먼) have forecast that artificial intelligence will raise productivity and cut costs, bringing deflation. An analysis says that is far from reality.

Altman said, "We are at the point where unimaginably cheap intelligence is within grasp." Musk has argued that AI and robotics will create extreme abundance and lower costs. SoftBank chairman Masayoshi Son (손정의) also forecast prices would fall 40 percent, saying, "Labor that is unnecessarily hard and sweaty will no longer be needed."

On Aug. 12, economic media outlet CNBC reported that while companies are adopting AI more slowly than expected, trillion-dollar investment in data centres and AI infrastructure is creating supply chain bottlenecks and pushing up power and equipment prices. With costs and inflation pressures rising before AI's productivity effects fully appear, the U.S. Federal Reserve is facing new policy concerns.

Ronnie Chatterji (로니 채터지), OpenAI's chief economist, said companies must adopt the technology and actually create value for AI to affect the economy. He said more time is needed to confirm the effect in productivity statistics.

Goldman Sachs Research estimated U.S. capital spending related to building AI this year at $581 billion, or about 820 trillion won, and global spending at up to $1 trillion, or about 1,420 trillion won. It forecast U.S. spending at 1.8 percent of gross domestic product, rising to 2.8 percent in 2028. A U.S. Census Bureau survey found the share of U.S. companies using AI was 17 to 20 percent, and adoption was higher among larger companies.

Peter Boockvar (피터 부크바), chief investment officer at OnePoint BFG Wealth Partners, compared AI's productivity effects with the internet. He said it is still unclear whether generative AI can lift the economy to the level of the internet.

Similar concerns are emerging inside companies. Julie Averill (줄리 애버릴), former chief information officer at Lululemon, led AI adoption. She said changing people's behaviour and getting them to trust AI in a large organisation is a bigger obstacle than the technology itself.

Gaps in AI use among companies are also widening. She said so-called power user companies that actively use AI have an average per-user token usage that is 8 times higher than the average company. Just 3 months ago, the gap was only 2 times. She added that companies that redesign workflows and ways of working to fit AI are delivering bigger results.

Economists call such hard-to-automate work a "weak link." Charles Jones (찰스 존스), a Stanford University professor, said real jobs are bundles that mix tasks that are easy to automate and those that are not. That means overall productivity may not rise as fast as expected even if some tasks are replaced by AI. He cited that even if AI automates reading radiology images, tasks such as talking with patients or collaborating with colleagues remain. Jones is currently on leave from Anthropic.

Views inside the Fed also diverge on AI's economic impact. Fed Chair Kevin Warsh (케빈 워시) said last November that AI would be a disinflationary factor that raises productivity and U.S. competitiveness, but he has recently taken a cautious stance, saying it is hard to predict when and how much supply-side effects will appear. Warsh appointed Jones to a task force the Fed will use as a reference in judging AI's economic impact, and venture investor Marc Andreessen (마크 안드리센) is also participating in the task force to assess AI's effects on productivity and employment.

Minneapolis Fed President Neel Kashkari (닐 카시카리), by contrast, said massive investment in data centres has added a new demand factor to inflation. U.S. Bureau of Labor Statistics consumer price index data show that over the 2 years through June, U.S. household electricity bills rose 10.1 percent, exceeding the overall inflation rate of 6.3 percent. Constraints are also appearing in supply chains for servers, and JPMorgan Chase estimated that DRAM prices will rise 400 percent by the end of this year compared with 2024. Prices of computer software and accessories have also climbed 22.9 percent since June 2024.

AI may still have the potential to raise productivity and lower prices over the long term, but what is showing up in the economy now is massive investment costs, supply chain bottlenecks, and rising power and equipment prices rather than the deflation many expected. A key variable is how much companies can overcome adoption speed, organisational change and supply-side constraints before AI translates into actual productivity gains.