token demand

China Telecom

2026-09-12 09:17:35

China Telecom think tank projects China’s annual token consumption at 10 quintillion in 2026

A report from the China Telecom Research Institute says demand for tokens, or word units used in AI processing, is rising sharply in China as the focus of the artificial intelligence industry shifts away from competition over foundation models and computing power toward large-scale agent deployment and application monetization. The report, titled Development Research Report on AI Infrastructure in the Agent Era (2026), describes AI as entering a new stage in which agents become the main form of deployment. According to the report, agents are making complex AI technology less visible to end users and more widely accessible, allowing both consumers and businesses to complete complex end-to-end tasks without needing specialized technical expertise. Rao Shaoyang, head of the Industry and Enterprise Strategy Research Institute at the China Telecom Research Institute, said China’s annual token consumption is expected to reach 10 quintillion in 2026 and exceed 3,500 quintillion by 2030, with a compound annual growth rate close to 12x.

810

AI Infrastruc

2026-08-28 02:47:40

Analysis says AI infrastructure cycle may outlast the internet era as token demand keeps expanding

An analysis cited by ChainCatcher argues that demand for AI semiconductors and infrastructure could last longer than it did in the internet era, driven by the combined effect of user penetration and rising token usage per person. The note says user penetration has already moved past 50%, while future growth is expected to come more from per-capita token consumption, which it describes as still being in an early stage. It also contrasts AI with the internet subscription model, saying inference carries a much higher marginal cost per visit and therefore requires heavier infrastructure investment. On commercial use cases, the analysis says the next phase of annual recurring revenue growth after developer coding is likely to come from broader programming, where non-programmers use coding infrastructure to complete non-coding tasks across industries. It adds that these broader programming tasks account for roughly 60% to 70% of ARR. The note also cites enterprise usage data for OpenAI and revenue mix data for Anthropic to support the view that token demand on the user side is still growing and that capital providers and beneficiaries remain closely aligned.

990

AI pricing

2026-08-07 15:20:08

China AI Models Move to Time-Based Pricing as Token Demand Surges

China's large language model providers are moving from flat per-use fees to tiered and time-based pricing as token demand explodes, according to CCTV Finance via Odaily. In the week from July 27 to August 2, weekly token call volumes hit 28.13 trillion, keeping China at the No.1 spot globally for a 14th consecutive week. The numbers underline a fast-growing market: platform operators report token demand is surging and that peak periods are already stretching supply. That mix of runaway demand and periodic compute shortages is now reshaping how model vendors bill their customers. Kimi, for example, has introduced four monthly subscription tiers priced between 39 and 559 yuan. DeepSeek, on the other hand, says it will raise prices for its open model API and is expected to adopt 'peak-valley' pricing, with daily peak windows from 9 a.m. to noon and 2 p.m. to 6 p.m. charged at double the standard rate.

1920

Jordi Visser

2026-07-15 13:02:02

Jordi Visser says AI selloff was a speed crash, not a bubble burst, as crypto gains statecraft role in the U.S.

Macro analyst and AI Macro Nexus founder Jordi Visser argues that the sharp correction in AI and semiconductor stocks between May and June 2026 does not mark the end of the AI trade. In his view, the move was a classic “speed crash” inside a structural bull market, designed to flush out crowded retail positions and quantitative leverage rather than signal the start of a cyclical bear market. Visser backed that call with several market indicators. He said Morgan Stanley’s tech momentum factor saw its 60-day volatility jump to 87, Samsung shares dropped 21% in 10 days for their biggest decline since COVID-19, and 87% of S&P semiconductor components moved into oversold territory on a 14-day RSI basis. Even so, 86% of the 100 AI names he tracks remained above their 200-day moving averages, while NVIDIA regained its 20-day moving average and flashed a MACD buy signal. He also tied the long-term AI buildout to memory constraints and token consumption. Visser said consumer AI agents could drive a 24x increase in token demand once interface bottlenecks are solved, while memory shortages may persist beyond 2030. On crypto policy, he pointed to U.S. Treasury Secretary Scott Bessent’s July 2026 speech, support for stablecoins and tokenization, and progress on the CLARITY Act and GENIUS Act as signs that the U.S. is repositioning crypto as a tool of financial statecraft.

1840