AI token spending index falls 20% from May high

A closely watched gauge of what the market pays for AI inference has dropped sharply from its spring peak, fueling a bruising rotation out of semiconductor and memory stocks that has wiped hundreds of billions of dollars in value across global markets this week.

The Silicon Data LLM Token Expenditure Index, which tracks the expenditure-weighted average price paid per million large-language-model tokens, is down nearly 20% from its May high after nearly doubling since its December inception, Bloomberg reported. For stock investors, the decline raises the possibility that AI companies are losing pricing power as enterprise customers grow more cost-sensitive — and that expectations for a sustained AI revenue boom may be misplaced.aiweekly

A Widening Gap Between Investment and Revenue

The token index's drop arrives amid broader warnings about unsustainable spending patterns. In a March special report, Allianz Research noted that AI capital expenditure is expanding roughly 46% faster than revenue — a divergence that exceeds the 32% gap observed during the 2001 telecommunications infrastructure bust. The five largest U.S. technology companies have collectively committed between $660 billion and $690 billion to capital expenditure in 2026, nearly doubling 2025 levels.globalgeopolitics+2

Adding to overcapacity concerns, Reuters reported on July 1 that Meta ↘ 6.98% is building a cloud business — internally called "Meta Compute" — to sell excess AI computing capacity to outside developers. The service would let customers access AI models hosted on Meta's infrastructure or rent raw GPU capacity, putting it in direct competition with Amazon Web Services Amazon.com, Inc. ↘ 4.57%, Google Alphabet Inc. ↘ 3.08% Cloud, and Microsoft ↘ 2.23% Azure.reuters+1

Memory and Chip Stocks Bear the Brunt

The resulting selloff has been punishing. On July 2, SK Hynix fell 14.6% and Samsung Electronics dropped roughly 9% in Seoul, dragging South Korea's Kospi sharply lower. CNBC reported the declines came as Asia's largest chipmakers bore the brunt of the global tech selloff after the Nasdaq Composite slumped overnight. The broader AI semiconductor complex suffered its worst two-day move since last year's Liberation Day tariff shock, according to reporting from multiple outlets.seekingalpha+3

Token prices have collapsed more than 90% since 2023, yet total AI spend has roughly doubled since last year, complicating the picture: cheaper tokens may simply reflect efficiency gains rather than weakening demand. Still, the convergence of falling token prices, a Meta-driven supply overhang, and the widest capex-to-revenue gap since the dot-com era has left investors questioning whether the AI trade's best days are behind it.navyaai+1