Prioritizing Stability Over Speed

At this year's shareholders' meeting, Ma Huateng made a candid, albeit somewhat awkward, comment regarding Tencent's AI endeavors.

"A year ago, we believed we had embarked on the right path, only to discover later that our vessel was taking on water; now, we find ourselves standing, yet unable to fully settle down."

Such frankness is a rarity, even among statements from internet industry executives. For a company with a market capitalization exceeding HK$4 trillion, for its leader to confess misjudgments to shareholders reflects either immense self-assurance or a genuine sense of urgency.

Most likely, it's a blend of both.

01

The 'Scissor Gap' in the Financial Statements

Tencent's Q1 2026 earnings report revealed a 9% year-over-year (YoY) revenue growth and a matching 9% YoY increase in Non-IFRS operating profit. For China's premier internet company by market cap, these figures are commendable, albeit slightly underwhelming.

However, if we exclude the investments in new AI ventures, the company's Non-IFRS operating profit growth would surge from 9% to 17%. In essence, AI endeavors have nearly halved Tencent's profit growth.

What does it take to transform a 9% growth into 17%? Q1 R&D investment soared to RMB 22.54 billion, marking a 19% YoY increase. Capital expenditures reached RMB 31.94 billion, up 16% YoY, accounting for nearly 40% of the full-year 2025 capital outlay. Selling expenses jumped from RMB 7.5 billion per quarter to RMB 11.3 billion.

The bulk of this additional spending was directed towards the training and inference of the Hunyuan large model, marketing efforts for Yuanbao, and R&D for Agent products like CodeBuddy and WorkBuddy.

Without delving into the specifics, the scale of AI investment in a single quarter roughly equates to 60% of the quarterly net profit.

In the current Chinese internet landscape, this approach seems somewhat counterintuitive. We've grown accustomed to narratives of 'cost-cutting and efficiency gains' and 'profit first' from major players. After all, ByteDance's ad revenue growth has decelerated in its financial report for the same period, and Alibaba's Q4 e-commerce revenue even experienced a YoY decline. Yet, Tencent has responded with a massive quarterly expenditure.

Labeling it a gamble would not be an overstatement. However, after reviewing this earnings report and the conference call transcript, my impression is that Tencent's AI moves resemble a meticulously calculated wager rather than a spontaneous rush.

02

The Costs Involved

Ma Huateng emphasized at the shareholders' meeting: "We can't simply dive in because others are doing so and attempt to seize their territory. We've tried that before, with limited success."

In simpler terms: We have a history of this.

From pursuing search during the PC internet era to competing in short-form video during the mobile internet boom, Tencent harbors painful memories of 'chasing trends.' This time, their AI strategy is clear: don't chase trends, set our own pace.

This sense of rhythm manifests in three key ways.

Firstly, there's the 'blood swap' in talent strategy. Previously, Tencent's AI core talent predominantly consisted of academic returnees over 45—impressive backgrounds but weak in execution. A 2024 personnel shakeup nearly replaced the entire first-generation AI team.

By late 2025, Yao Shunyu, a post-95, emerged as Tencent AI's youngest leader, ushering in a wave of youthification and organizational restructuring across the team. The Hunyuan team was split into Large Language Model and Multimodal Model departments, while core infrastructure departments like AI Infra and AI Data were established.

Secondly, there's the 'tear down and rebuild' of infrastructure. In April 2025, Tencent overhauled the Hunyuan R&D system. As Ma Huateng put it, past issues stemmed from insufficient infrastructure and inadequate model iteration frequency.

The first fruit of this rebuild, the Hy3 preview model, was released and open-sourced in April this year, boasting 295 billion total parameters and 21 billion activated parameters. Since April 28, it has topped OpenRouter's weekly token consumption rankings for three consecutive weeks, with usage exceeding the previous Hunyuan 2 by over 10 times.

Thirdly, there's cautious commercialization. During the conference call, Tencent management explicitly stated that the AI era shouldn't blindly pursue Daily Active User (DAU) growth. Unlike traditional internet models, AI services entail high computing and inference costs—"purely chasing DAU and user duration isn't economical." The key lies in identifying high-value usage scenarios where users are willing to pay.

This judgment diverges from industry norms.

Alibaba's Wu Yongming recently remarked, "Almost no GPU cards are idle," setting a five-year cloud and AI commercialization revenue target of $100 billion annually, with capital expenditure commitments far exceeding the previously announced RMB 380 billion. ByteDance's AI capex plan even rose from RMB 160 billion to RMB 200 billion.

In comparison, Tencent's 'slowness' seems somewhat out of place.

But there's a rationale behind this slow pace. At a time when everyone discusses 'thousand-card clusters' and 'ten-thousand-card clusters,' Tencent's calmness raises a more fundamental question: When all large model vendors are burning money on GPUs, who will be the first to get users to open their wallets?

03

Anticipating the Chemical Reaction of WeChat AI

Currently, advertising appears to be the closest scenario to monetization. Q1 Tencent marketing services revenue grew 20% YoY, with the earnings report explicitly attributing this to AI-driven upgrades in ad recommendation models. However, this merely enhances efficiency in existing businesses rather than creating new revenue streams.

WorkBuddy ranks first in DAU among domestic efficiency AI agents, while CodeBuddy demonstrates decent retention in code generation scenarios. Nevertheless, these products are still in the user accumulation phase, with large-scale commercialization pending.

Tencent's judgment during the call is that C-end AI subscriptions have a limited ceiling in the Chinese market, while B-end API calls and computing power leasing represent clearer monetization paths.

This aligns with Dowson Tong's previous views. The head of Tencent CSIG explicitly stated that as the industry matures, the capability gap between mainstream large models is narrowing. Enterprises' core demands are no longer just having the best model but maximizing its utility.

In other words, Tencent isn't betting on creating the 'strongest model' but on its ecosystem maximizing the model's value. WeChat's 1.3 billion Monthly Active Users (MAUs) and millions of mini-programs represent Tencent AI's true differentiation. The gray-scale testing of WeChat AI agents starting mid-year will be the first test of this logic.

Historically, WeChat took at least four to five years from its 2011 launch to achieve true commercialization explosion. The transition from agency to self-developed games also involved prolonged growing pains. Tencent clearly allows sufficient time for its AI bet.

Tencent management stated during the earnings call that they don't measure Return on Investment (ROI) by single quarters or products but manage from an 'asset portfolio' and 'full lifecycle' perspective.

Behind this phrasing lies unwavering strategic determination and displayed confidence—Q1 Tencent's net operating cash flow reached RMB 101.4 billion, a new historical high for the period. In essence, Tencent can truly 'afford to burn.'

The question remains: With the right direction, is the speed sufficient?

Ma Huateng's remark about 'still can't sit down, hoping the boat speeds up' reveals Tencent's internal mindset. Admitting lag isn't terrible, nor is acknowledging anxiety. What warrants caution is using 'slow strategy' as an excuse to mistake conservatism for deep consideration, or packaging lag as calculated maneuvering.

For Tencent, the paradox of this AI arms race is that technology requires time to mature, but the market won't wait for you to finish maturing before moving on. After Alibaba, ByteDance, and Baidu, whether Tencent's 'not seeking speed but stability' represents wise restraint or another form of hesitation will require time to answer.

At least for now, Tencent is betting on long-term ecosystem integration and scenario penetration. The costs of this choice are already visible in the income statement; however, the returns remain absent from the profit and loss statement. The final outcome hinges on whether Hunyuan and WeChat can create a chemical reaction.

This might be the only Chinese internet giant truly positioned to achieve deep coupling between a 'super app' and self-developed large models. The upside of this narrative cannot be measured by a single quarterly report.

— END —