China’s daily AI token calls rose from 100 billion at the beginning of 2024 to over 500 trillion in July 2026, Tencent senior executive vice president Dowson Tong said in a keynote at the Tencent Global Digital Ecosystem Conference in Shenzhen on 29 October 2026, citing data from national departments, according to ChainCatcher’s report of the speech.
By our arithmetic, 500 trillion tokens a day is about 11 times the nearly 46 trillion a day that Google’s first-party model APIs process. The two figures cover different things: one is a national total, the other one company’s API traffic.
Short explainer · China token calls
A vertical news Short narrated from the published headlines and opening lines of this piece and two related FourWeekMBA pieces.
Business Pill · Tokens per Watt
An evergreen Business Engineer explainer on what an AI data centre produces, measured in tokens per watt. It explains the general idea only and does not discuss Tencent or China’s token figure.
The key insight: Tencent’s Dowson Tong said China’s daily token calls passed 500 trillion in July, citing national data; by our arithmetic that is about 11 times the nearly 46 trillion a day Google’s first-party model APIs process, on different scopes.
Tong runs Tencent’s Cloud and Smart Industries Group. Some reports give his name as Tang Daosheng, with the same title. ChainCatcher reported that he said the daily token call volume in China had grown “from 100 billion at the beginning of 2024 to over 500 trillion in July of this year, a growth of over 5,000 times in two years”.
ChainCatcher’s headline attributed the figure to Tencent founder Pony Ma, but the body of its report attributes it to Tong, and so do we. The report says the figure came “according to data from relevant national departments”; it does not say which department or how calls were counted.
By our arithmetic, 500 trillion a day is about 347 billion tokens a minute. Alphabet said on 28 October that its first-party model APIs process nearly 32 billion tokens a minute.
AI tokens a day: China’s reported national total in July 2026 and at the beginning of 2024, as cited by Tencent’s Dowson Tong, against Google’s first-party model APIs as stated on 28 October 2026 (our arithmetic from nearly 32 billion a minute). The scopes differ. Sources: ChainCatcher report of the 29 Oct 2026 keynote; Alphabet, 28 Oct 2026.
Hy4 and Hy4 Flash
Tong also said Tencent will release two Hunyuan models, Hy4 and Hy4 Flash, TechNode and AASTOCKS reported. TechNode said the full Hy4 “will improve performance on tasks requiring sustained reasoning and execution, including real-world productivity workflows”, and adds visual understanding.
AASTOCKS reported that “Hy4 flash is positioned as a cost-effective and practical model”. TechNode said Hy4 Flash “will focus on cost efficiency while retaining visual understanding and AI agent capabilities”.
The Hy4 preview, released in August, “has 770 billion total parameters, with 49 billion activated during inference, and supports a context window of one million tokens”, TechNode reported. No release date or price was reported for either new model.
The Capacity Behind It
ChainCatcher reported Tong as saying that Tencent Cloud is “accelerating the construction of global data centers, having deployed over 2 million servers”.
He said that in the past year Tencent Cloud’s overseas AI demand “has achieved triple-digit growth, and the number of active international customers has doubled”, according to the same report. It added that Tencent Cloud had worked with the Hunyuan team to “increase the single-card throughput of domestic chips by 5 times” ahead of the model’s launch.
Where the Tokens Go: Office Agents
Tong pointed to Tencent’s WorkBuddy agent as an example of AI applications behind the token growth, saying such tools are moving from document work into businessanalysis, operations, market research and marketing, according to ChainCatcher.
The same day Tencent launched Buddy One, which ChainCatcher said links “WorkBuddy, Tencent Docs, Tencent Meeting, Tencent Cloud Drive, and ima into a unified AI office system”. Its VIP tier is listed at “700 yuan/month, which includes 50,000 universal points per month”, the report said.
WorkBuddy drew 85.34 million visits in September, the most of any office agent in China, according to research firm Analysys as cited by Beijing Business Today.
How the Lever Works
As we read the announcements, Tencent is pairing a flagship model with a cost-effective Flash version, selling capacity through its cloud, and bundling its office apps around the WorkBuddy agent. The reports describe the models in terms of practical use and cost-effectiveness.
As we read it, the national token figure is the demand side of that pitch: Tencent put a number on how much AI work is already running in China, alongside the products it wants that work to run on.
As we read the reports, Tencent paired three announcements on one stage: a national demand figure, two models with one aimed at cost-effectiveness, and cloud capacity of over 2 million servers.
ChainCatcher reported Tong as saying that over the past year Tencent Cloud’s overseas AI demand “has achieved triple-digit growth”, and that Tencent Cloud is “accelerating the construction of global data centers”.
Buddy One unifies the points system so WorkBuddy points can be used in Tencent Docs, ima and Tencent Cloud Drive, though points bought separately in each app are not merged, ChainCatcher reported.
Dowson Tong (Tang Daosheng), Tencent, 29 October 2026, as reported by ChainCatcher
“from 100 billion at the beginning of 2024 to over 500 trillion in July of this year, a growth of over 5,000 times in two years”
Three Implications
FOR INVESTORS Tencent gave a national token figure and a server count, over 2 million, but the reports carry no Tencent token volume, modelprice or release date.
FOR MODEL BUYERS TechNode reported that Hy4 Flash will focus on cost efficiency while keeping visual understanding and agent capabilities; no price was reported.
FOR WESTERN AI PROVIDERS By our arithmetic, China’s reported daily token calls are about 11 times Google’s first-party API volume, though the two figures measure different things.
The framework says of the token: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts”.
Tencent’s keynote, as reported, gave a national token count, cloud capacity and an office bundle built around the WorkBuddy agent. The framework does not discuss Tencent or China’s token figure.
What Is Not Established
The 500 trillion figure is national and was cited by Tencent from unnamed national departments; the reports do not say how a token call was counted or which providers were included. It is not a Tencent figure.
The comparison with Google is our arithmetic across different scopes: a country’s total against one company’s first-party API traffic, and July against Alphabet’s late-October figure. No release date, price or benchmark was reported for Hy4 or Hy4 Flash, and the quotes here come from media reports of the keynote, not a Tencent transcript.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Tencent’s Dowson Tong said China’s daily token calls passed 500 trillion in July, from 100 billion at the start of 2024, and that Hy4 and a cost-effective Hy4 Flash are coming. By our arithmetic, the national figure is about 11 times Google’s first-party API volume, on different scopes.
A note on sourcing. Figures and quotes come from media reports of Tencent’s 29 October 2026 keynote by TechNode, AASTOCKS, ChainCatcher and PANews; we did not see a Tencent transcript. The national token figure was cited by Tencent from national departments. The Google comparison and per-minute figure are our arithmetic; linked earlier pieces carry their own sources.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
