AI News

ByteDance's 2029 goal and Tencent's "half the workforce": how the two companies are reallocating AI resources

TMTPost · AGIUpdated Nov 14 19:10

AI summary

TMTPost reviewed ByteDance's and Tencent's AI resource investments: according to secondhand accounts, Zhang Yiming said at an internal meeting of ByteDance's AI team that catching up with the frontier would take until 2029 or 2030 at the earliest, and ByteDance raised its 2026 AI infrastructure budget to more than 200 billion yuan; in his speech on Tencent's 28th anniversary, Pony Ma proposed that half the workforce defend the core business and half be invested in AI Native, which was rumored to mean a 50% layoff, and PR director Zhang Jun said this was a misunderstanding and a rumor. The article argues that both companies treat time and manpower as resources they can actively allocate, and proposes verification criteria such as whether AI revenue is reported separately and whether the depreciation period for capital expenditure is disclosed.

Why read this

AI score 62

The article places ByteDance's timeline and Tencent's workforce reshuffle within the same framework, offering observable indicators that can be checked year by year.

Topics

Tags

Body · AI translation

Measuring an internet company used to be mainly about user time spent. Around this metric, the industry recently had a debate that lasted ten days, focused on data definitions and conceptual boundaries.

According to QuestMobile monthly tracking estimates cited by third-party institutions, among the top 50 mobile apps by usage time in July, ByteDance-affiliated apps accounted for 40.9% of time and Tencent-affiliated apps for 29.1%; this is the conventional basis outsiders use to judge the two companies. This definition covers only leading mobile apps and excludes long-tail apps and PC, which is the starting point of this debate.

Not long afterward, the two companies successively disclosed their respective AI investments. According to reports, Zhang Yiming gave a deadline for catching up to the frontier at an internal meeting of ByteDance's AI team: 2029 or 2030. In the same period, a rumor that Tencent was "cutting 50% of staff" trended on social media; PR director Zhang Jun called it a misunderstanding and a rumor. The rumor came from Tencent's 28th-anniversary speech, in which Pony Ma proposed using half the workforce to improve efficiency in core businesses and the other half to explore AI Native.

ByteDance: Converting the distance to the frontier into a year

According to reports, Zhang Yiming said at an internal meeting of ByteDance's AI team that ByteDance is capable of competing with OpenAI and Anthropic, but that truly catching up to the frontier would take until 2029 or 2030 at the earliest. The report also mentioned that the pre-training datasets for the Seed series models are several orders of magnitude smaller than those of top US rivals, and that ByteDance does not rely on distillation to trade for short-term leaderboard results. On November 12, another media outlet obtained the same account from multiple informed sources; these are all secondhand accounts, and ByteDance has not officially confirmed them.

This deadline corresponds to both the gap and the pace: model catch-up is measured in years, while application iteration is measured in weeks. Zhang Yiming said at the meeting that an AI assistant benchmarked against Meta Muse would be launched within weeks; ByteDance plans to use application and engineering capabilities to maintain product rhythm while making long-cycle investments at the underlying layer.

To judge whether such statements are serious, one can look at how much money the company is putting behind them. ByteDance raised its 2026 AI infrastructure budget to more than 200 billion yuan and in September signed a syndicated loan of about $29.6 billion. According to media disclosures, ByteDance's net profit in the first half of 2026 was about $20 billion, down by a single-digit percentage year on year, while cash flow from its main business remained stable; the budget's execution is supported by profit and cash flow.

Whether 2029 is accurate cannot be verified right now. What can be confirmed is that this is the first time ByteDance has publicly given a timeline for catching up to the frontier; previously no company was willing to disclose its gap to the frontier, and outsiders could only guess from leaderboards.

After the year was announced, progress had a reference point; the funding side, the selection side, and talent considering joining all gained another basis for comparison.

Whether this coordinate can work depends on whether ByteDance continues to disclose progress. If it is only a one-time statement, by 2029 not everyone may remember it. Victory or defeat is still determined by products and models; what the public year changes is the evaluation method, and outsiders can from now on compare ByteDance's progress year by year.

Tencent: It is resource allocation

Tencent's "half the people" remark came from the 28th-anniversary Linktime online livestream on November 11. According to a transcript of the speech disclosed by media, Pony Ma said in the speech that Tencent had previously done asset-light businesses and that this was the first time it was doing such an asset-heavy business, so the financial and organizational approaches were different. He proposed that half the people be like the army, guarding core businesses and improving efficiency, while the other half be like the navy, exploring AI Native.

This ratio was simplified into layoffs in circulation. The most direct evidence for testing this reading is the total number of employees: as of June, Tencent had 115,927 employees, compared with 111,221 in the same period a year earlier, a net increase of about 4,700 over the year.

Lu Shan, who heads the Technology Engineering Group, proposed not using KPIs from existing businesses to constrain AI innovation teams and giving innovation room for error tolerance. The first sentence explains the reason for compressing layers: the decision chain is too long to catch the window of opportunity. The latter sentence shows that AI innovation teams do not yet have corresponding assessment methods, and whether they are doing well will be difficult to judge uniformly in the short term.

Organizational moves had already been happening: in May, multiple departments in the WeChat Group piloted a project-owner system, and in July the IEG Ecosystem Development Department abolished L1 and L2 management ranks. Capital expenditure expanded in tandem; according to Tencent's interim results, capital expenditure in the first half of the year was 84.72 billion yuan, including 52.78 billion yuan in the second quarter alone, up 176% year on year. Pony Ma said in the speech that those who choose to adapt to the new approach must be decisive.

The risk of "half the people" ultimately lies in delivery. If the half directed toward AI Native cannot produce demonstrable assets within a reasonable period, the layoff narrative may become a true prophecy on its own without anyone pushing it. By comparison with peers, Zuckerberg's 2022 "year of efficiency" also rearranged resources through an internal speech, and the subsequent layoffs and AI pivot unfolded from there.

From user time to resource allocation

ByteDance and Tencent treat time and manpower as resources that can be actively allocated, rather than expenses that naturally grow with existing businesses. User time is still generating cash revenue; whether this revenue stays in existing businesses or is redirected to AI may determine the two companies' next stage.

This allocation logic is not limited to ByteDance and Tencent. On November 12, Robin Li said at Baidu World 2026 that the next act of AI is the intelligent organization, and he reduced future roles to two categories: Builders responsible for creating new things and Sellers responsible for connection and trust. Baidu disclosed that the average daily number of active agents at its headquarters increased from 5,000 to 37,000. The three companies' judgment is consistent: what AI will change is not only products, but also the companies themselves.

Organizational change is converging in direction: Tencent is compressing its middle management, Baidu in April scrapped the T, P, E, and M job-level labels it had used for years, and ByteDance CEO Liang Rubo's June all-hands letter wrote "deep into the front line" into managers' performance reviews. The logic behind this is that AI has compressed the value of passing information up and down the chain, and the thicker the middle layer, the slower a company's response to technological change.

The yardstick of the capital markets is also adjusting. Some institutions call this year's third quarter a quarter for verifying the profit inflection point of Hong Kong-listed tech stocks, while cautioning that the profit recovery mainly comes from the scaling back of food-delivery subsidies and that no substantive improvement has yet appeared on the demand side.

The consensus expectation compiled by institutions is that Tencent's adjusted net profit in the third quarter will fall about 5% year on year. Another institution judges that the actual result may be better than this expectation and gives an estimate of 66 billion yuan in capital expenditure for the quarter. Tencent's price-to-earnings ratio is about 14 times, in the bottom range of the past decade. Taken together, the three sets of numbers lead to the conclusion that the market's assessment cycle for internet companies is lengthening: AI spending is suppressing short-term profit, the P/E ratio has been pushed to a historic low, and the market has begun pricing companies according to the payback cycle of AI investment, with quarterly profit no longer the only coordinate.

For the industry, such adjustments are necessary. Heavy assets, long cycles, and organizational reshuffling are all moves unfamiliar to internet companies, and the way they are measured needs to be updated accordingly. This set of criteria can be summarized in three points: whether AI revenue is disclosed separately, whether the depreciation period for capital expenditure is disclosed, and whether the staffing and computing-power allocation for new businesses is made public. The various standards at the company level, industry level, and practitioner level ultimately all rest on these.

The industry level still lacks a unified framework for verifying effects. Token call volume and the number of active agents are only process indicators; only the proportion converted into revenue reflects results, which is why discipline in measurement standards matters. ByteDance's Volcano Engine has a 49.5% share of China's public cloud MaaS market by call volume, and more than 40% by revenue.

Resource flows also leave traces. Tencent's headcount is still increasing year on year; the change is happening in reporting lines and job definitions. How many people and how much computing power the new battleground gets will ultimately show up in hiring, financial reports, and organizational adjustments. These clues all have corresponding observation windows. Tencent releases its third-quarter financial report on November 18; the actual capital expenditure figure, changes in free cash flow, and whether AI revenue is disclosed separately will provide a checkable reading for the "half of the people."

As for ByteDance, whether the AI assistant benchmarked against Meta Muse is released as scheduled, the Seed main model's version update has stayed at 2.1 since June, when the next version will appear, and whether the developer for the second phase of the Brazil data center has been determined are all on the timetable.

Whether the two sets of numbers can be delivered has only one common precondition: the cash generated by the old businesses must keep covering the consumption of the new businesses. Every step of investment ByteDance has scheduled through 2029 relies on the cash flow of its advertising business. Tencent is directing half of its people toward new directions, with the cost coming from its already firmly established gaming and fintech businesses.

This also means that what determines how long this investment can last is not the 40.9% or 29.1% on the rankings, but the gap between cash inflow and consumption.

(This article was first published on the TMTPost APP, author | AGI-Signal, editor | Zhao Hongyu)

Source

TMTPost · AGI