The artificial intelligence trade has produced little return for more than three months, according to Jurrien Timmer, Director of Global Macro at Fidelity Investments, as token spending and GPU rental rates decline while memory-chip prices continue rising. Fidelity data show the Silicon Data LLM token expenditure index fell 49% over 50 days, while rental rates for Nvidia H100 and A100 GPUs declined after earlier gains of 56% and 16%, respectively. Timmer said the AI infrastructure cycle is inflationary and has an uncertain return for companies investing trillions of dollars in computing capacity, although he believes AI could eventually raise the economy’s non-inflationary speed limit through a productivity surge. U.S. hyperscalers are projected to spend about $916 billion on capital expenditures over the next 12 months, while combined 2027 spending by Alphabet, Amazon, Microsoft, Meta Platforms, Oracle and Space Exploration Technologies is expected to exceed $1 trillion. A global memory-chip shortage is expected to persist until 2028 as manufacturers expand capacity. Ark Invest CEO Cathie Wood separately said AI, robotics, energy storage, blockchain technology and multiomics sequencing could drive a deflationary boom and give real GDP growth of 7% or more over the next five years, compared with roughly 3% since the Industrial Revolution. She said AI inference costs had fallen by more than 99% annually while inference-token demand rose 25-fold in 2025. Real GDP grew at a 1.5% annual rate in the second quarter after expanding 2.1% in the first.