Booming US demand for AI driven by falling token prices

Business adoption of AI is booming, rapidly outpacing the take-up of the internet in the 2000s and 2010s, such that the eventual productivity payoff from AI could be delivered sooner than widely expected. The US experience shows AI is being used in all industries, although most intensively in the IT, professional services, and finance sectors. However, Ramp transactions data show that extraordinarily strong demand for AI tokens from US tech-intensive businesses is mostly a function of average token prices falling at an even faster rate, halving over recent months. As a result, growth in the dollar spend on AI tokens by US tech-intensive firms – while still very strong – has slowed sharply, which, if sustained, would raise questions about the return on the massive investment undertaken by AI providers.

• AI usage by US and EA businesses ramping up, although it is lagging many other advanced economies.
  Business use of AI varies considerably across countries and is actually at the lower end of the scale in the US and the euro area, at 25% of US businesses and 21% of EA non-financial businesses. However, some of the differences between countries can reflect differences in survey questions and sampling.

• AI is being used across all industries in the US, but most extensively in the IT, professional services and finance sectors.
  In the US, AI has made inroads across the entire economy, although there is a wide variation of industry use of AI. The heaviest users of AI are the IT, professional services, and finance sectors, which is not surprising, as they have long accounted for the bulk of investment in software. At the other end of the distribution, there is a long tail of industries with a below-average take-up of AI.

• Business adoption of AI is occurring at a much more rapid rate than take-up of the internet, such that the eventual boost to productivity could emerge sooner than widely assumed.
  The productivity payoff from AI could be delivered more quickly than the earlier payoff from the use of the internet given that US data show that business use of AI is growing at a much faster pace than take-up of the internet over the 2000s and 2010s.

• Private-sector data overstate the economy-wide use of AI in the US, but provide the right signal on the direction.
  Private-sector estimates of business use of AI in the US – such as the transaction-based data published by Ramp that are monitored by the Fed – are heavily skewed towards companies that are more intensive users of AI, although they still provide valid insights into the rate of take-up of AI.
  
  
• US business use of AI is booming among heavy users of technology, but only because the price of AI tokens has halved over the past three months.
  With that caveat, the Ramp transactions data show that business use of tokens for the top AI models continues to grow at an exceptionally rapid rate, increasing by about 70% over the past three months in log terms. However, this phenomenal growth has been supported by average token prices falling at a much faster pace, down about 50% over the past three months in log terms.
  
  
• Falling prices mean that the dollar value of US tech-intensive business spending on AI has slowed sharply, albeit to a still strong rate, which is problematic for AI providers if sustained.
  With exceptionally strong demand for AI tokens propped up by rapidly falling average token prices, growth in US tech-intensive businesses’ dollar spending on tokens has slowed sharply, according to the Ramp data. Log growth over the past three months is still strong at about 20%, but this is the slowest growth since a brief volume-driven contraction late last year, and well below the average 60% growth over most of this year. It is too soon to know whether this slowdown will be sustained, but it raises questions about the return on investment for AI providers.