Tencent’s Army-Navy Split: A Desperate AI Reckoning
The Chinese tech giant is dividing its workforce in half—one half to defend its core business, the other to explore AI-native products. Here’s what that signals about the pressure on China’s tech giants.

On November 11, 2026, Tencent held an online live stream for its 28th anniversary. Pony Ma proposed that 50% of employees should act like an “Army,” holding the ground and doing their duties, while the other 50% should act like a “Navy,” exploring AI Native and new territory. The next day, The Paper confirmed the remarks with a person close to Tencent, adding that this differs from the “cost reduction and efficiency improvement” round several years ago. That earlier round was relatively passive and focused on cutting costs; this one is a proactive move for the AI era, focused on efficiency gains.
When the news reached Maimai, the claim that Tencent planned to lay off 50% of its staff quickly appeared. Tencent responded clearly: this is a reallocation of resources to adapt to the AI era, and the layoff claim is false.
The rumor spread fast because it matched something employees could feel. A company publicly says that half its people will fight a new battle, that middle management must become flatter, and that those who get on board will no longer be coaxed or negotiated with. Even if it is not called a layoff, demand for old roles is shrinking and the threshold for new roles is rising. The important question about “Army/Navy” is not whether it is a euphemism for layoffs. It admits one thing: Tencent can no longer run the entire company on one timetable.
That sentence half a year ago, and this sentence half a year later
In May 2026, at Tencent’s shareholders meeting, Pony Ma was asked whether Tencent’s AI was lagging. He said: “At first, a year ago, we thought we had got on the boat, and later found that the boat was leaking. Now it feels like we have stood on it, but still cannot sit down. We still hope the boat can go a bit faster.” The boat leaks, but the people are on it; what remains is a question of speed. He added that Tencent may not be the fastest to seize the opportunity, but it insists on combining its own advantages and proceeding steadily. “We cannot watch others do it over there and just casually cross over to grab their turf.”
Half a year later, the wording of the 28th-anniversary speech changed. In the leaked internal speech, Pony Ma used the words “life-or-death,” “shifting from land warfare to sea warfare,” and “frankly speaking, it is a bit late now; we can no longer leisurely feel safe.” Bloomberg’s cited report offered another phrase: the industry has entered the “Age of Discovery.”
Between “leaking but on board” and “life-or-death” lies a set of numbers. iAnalysys’s October 2026 annualized Token revenue ranking: ByteDance/Volcano Engine about 30 billion yuan, Alibaba about 16 billion yuan, Tencent about 3.2 billion yuan, and Baidu about 2.5 billion yuan at the bottom. The gap between Tencent and ByteDance is nearly tenfold. The term “Navy” amounts to admitting that the latecomer window in the AI era may not open on the old rhythm. The Army continues to guard WeChat, games, and advertising, using AI to improve efficiency; the Navy enters the water now, rather than waiting for the land war to end before building ships.
Centralize authority first, then split
“Half Army, half Navy” is not an isolated move. Over the previous year, Tencent pulled back in its R&D system.
In December 2025, Yao Shunyu, a former OpenAI researcher from Tsinghua’s Yao Class, became Tencent’s chief AI scientist, reporting directly to President Martin Lau and concurrently heading the AI Infra Department and the Large Language Model Department. In March 2026, the nearly decade-old Tencent AI Lab was dissolved and its staff redeployed; Yao’s model system became the main trunk of Tencent’s large-model R&D. In July, the Hunyuan Multimodal Model Department and the Large Language Model Department were merged into the Foundation Model Department, with R&D for language, image, video, speech, and 3D placed under one person.
This timeline explains one thing: the underlying grammar of the AI war does not allow “federalism.” Tencent’s competitiveness over more than 20 years was built on independent decision-making by business groups, with the teams closest to the market making the calls. Compute must be centrally procured to dilute costs, data must be centrally fed to iterate models, and training and inference must be uniformly scheduled to form a combined force. Yao Shunyu’s management of Hunyuan is the R&D layer clenching scattered fingers into a fist.
The “Army/Navy” split adds a layer of manpower allocation on top of this centralization. Old businesses no longer pursue scale growth; after efficiency gains from AI, the Army holds them; the other half is pushed to the AI-native front line. Pony Ma’s description of personnel deployment was described as very blunt: it is not like the past, when there was still “coaxing, persuading, and negotiating terms.” Middle management was pushed to the fore first, because under the requirements of “flat and agile,” management bandwidth that is passed down layer by layer and attenuates at each level is exactly what the AI era least needs. An employee said Hunyuan has simplified its reporting line from “frontline—team lead—director” to “frontline R&D—project lead.” Dowson Tong’s earlier wording was more direct: AI-native product R&D teams have become flatter than before, and “teams will become smaller and smaller, with more ‘small teams of three to five’ completing R&D work.”
The Navy has a landing point
Seeing only anxiety and coercion would underestimate the texture of this adjustment. Tencent was not forced into the water without a point of leverage.
WorkBuddy provides a landing point. Launched in March 2026, by the end of October it had completed 9 billion task requests cumulatively; QuestMobile data showed September MAU of 32.353 million, ranking first in the office Agent track, with quarter-on-quarter growth of 647.1%, more than three times the second place. The enterprise version has landed in more than 50 industries, covering 50% of China’s top 500 companies and 60% of the top 100. Its playbook is very “Tencent”: it does not force binding to Hunyuan, opens access to external models, relies on the context of WeChat and WeCom to embed into workflows, and breaks through at a single point with light tool attributes.
This means the Navy’s exploration is not pure drift. The path Tencent has chosen is: at the model layer, admit the lag and keep chasing; at the application layer, use product capability and ecosystem distribution to first build a defensible position in the scenarios closest to monetization. Capital expenditure is increasing in sync. In Q2 2026, capex was 52.78 billion yuan, up 176% year on year and 65% quarter on quarter; in Q1 it was 31.94 billion yuan, and management explicitly said the increase would further expand in the second half. Annualizing the Q2 pace puts full-year compute procurement above 200 billion yuan. Money is being sent to the sea.
The layoff rumor points to job substitution
Back to “50% layoffs.” Tencent’s clarification is clear: this is a division of labor, not a cut. The rumor could resonate instantly because it points to a structural pressure.
Tencent’s total headcount was about 115,000 at the end of 2025, and it was still increasing in the first half; management also said it would recruit AI talent “quite aggressively.” On one side, old roles are being repriced under the logic of efficiency; on the other, new capabilities are being fought over at high prices in the market. The people in the middle layer, those who can neither get on the boat nor hold the ground, are the source of anxiety. Pony Ma said in the speech: “In the existing Army units, who can get on the boat and whether they can fight a sea war must be prepared in advance.” Those who cannot prepare: the organization will not stop and wait. This sentence is more worth scrutinizing than the number “50% layoffs,” and it is colder.
Coda
The “Army/Navy” signal is that Tencent admits it must live in two times at once. In one time, WeChat and games are still steadily producing profits, and someone needs to guard them; in the other, technological iteration is thinning the experience of “latecomer wins,” and someone needs to jump into the water now, even if they do not yet know where the other shore is.
Pony Ma said, “We have stood on it, but still cannot sit down.” This time, he lets half the people keep standing and the other half enter the water. Which half will find a landing point first, and which half will discover first that the boat is still leaking. There is currently no answer.
About the Creator
Jin
Writer of reamstories
https://reamstories.com/jin
Enjoyed the story? Support the Creator.
Subscribe for free to receive all their stories in your feed. You could also become a paid subscriber, letting them know you appreciate their work.
Comments
There are no comments for this story
Be the first to respond and start the conversation.