Tencent does not usually move like a startup. For decades, the Shenzhen giant has been patient, diversified, and quietly dominant — the kind of company that makes more money from video games and WeChat ads than most countries make from oil. But this week, Tencent unleashed a barrage of AI news so dense it felt like a deliberate campaign.
In the space of three days, the company revealed that its Hy3 AI model saw API calls jump 68-fold after launch, that it spent $53 billion on AI hardware in a single quarter, that it could turn an instant profit renting that hardware but is choosing not to, and that it has backed Swedish vibe-coding unicorn Lovable at a $13.3 billion valuation. The message is unmistakable: Tencent is going all-in on AI, and it wants the world to know it.
The Numbers Behind the Bonanza
Tencent’s second-quarter earnings, reported on Wednesday, were a study in contrasts. Revenue hit 204.78 billion yuan ($30.36 billion), up 11% year-on-year and ahead of analyst estimates. Domestic gaming revenue surged 17%, driven by titles like Delta Force and Valorant PC and Mobile. Marketing services revenue rose 22%, powered by AI-driven ad recommendation systems. Net profit came in at 56 billion yuan, below expectations, but up 9% when adjusted for one-time items.
Then came the capital expenditure bomb. Tencent spent 52.8 billion yuan ($7.8 billion) in the quarter, a 65% jump from the previous quarter. Free cash flow turned negative, to the tune of 13.8 billion yuan. The company is pouring money into GPUs, data centres, and custom infrastructure at a pace that rivals the biggest spenders in Silicon Valley.
CEO Ma Huateng made the strategy explicit. “At the infrastructure level, we substantially stepped up our procurement of compute, which will enable us to convert usage of our applications and models into revenue going forward,” he said in a statement.
But here is the twist: Tencent’s executives admitted they could make money on that infrastructure almost immediately if they simply rented it out. Chief Strategy Officer James Mitchell told analysts that demand is so strong the company could recover depreciation costs “almost immediately.” President Martin Lau added that Tencent has offers for its compute capacity “at more than 30 percent profit compared to the price that we paid just a few months ago.”
Instead, Tencent is keeping the chips for itself.
Why Rent When You Can Build an Empire?
Tencent’s leadership is clear-eyed about the choice. Behaving like a neocloud would deliver a “decent return in an immediate timeframe,” Lau said. But Tencent is “playing a different game.” It is allocating a “very substantial proportion” of its new compute to building state-of-the-art models and deploying market-leading AI applications in China.
The bet is that superior intelligence will translate into superior economic returns over the longer term. Those returns will come from services like WorkBuddy, which Tencent describes as an agent swarm that can “plan, execute, and run tasks in parallel, handing back complete deliverables end-to-end in one flow.” CodeBuddy, its code generation tool, is already accelerating cloud migration projects and creating a virtuous loop that pulls more customers into Tencent Cloud.
It is a classic vertically integrated play. Own the chips, own the models, own the applications, and capture the margin at every layer. Amazon and Microsoft have tried similar strategies. The difference is that Tencent has 1.349 billion monthly active users on Weixin and WeChat, a distribution channel no Western AI company can match.
Hy3’s 68-Fold Surge
The centerpiece of Tencent’s model push is Hy3, the latest version of its Hunyuan foundation model family. Released in July, Hy3 is a 295-billion-parameter Mixture-of-Experts model with 21 billion active parameters per token and a hybrid fast-and-slow-thinking reasoning framework. It is open-weight and available globally through WorkBuddy, Tencent Cloud, and developer platforms.
The market response was immediate. According to the company, Hy3’s API call volume surged more than 68-fold compared to the previous generation within one week of release. It also topped the global large language model usage leaderboard on OpenRouter in its first week. That is not a modest improvement. That is a breakout.
Tencent’s earlier global rollout of Hy3 was already a warning shot to Western model providers. The Apache-licensed model offered high performance without the licensing restrictions that have made some Chinese open-weight models politically radioactive. Now the usage numbers suggest developers are actually showing up.
Hy4, Hy5, and the Long Game
Tencent is not stopping at Hy3. Martin Lau confirmed the company is developing Hy4, which will be larger and “able to beat AI models of an even bigger size in terms of performance.” A Hy5 is also in the pipeline. Lau promised that at some point Tencent will deliver a state-of-the-art model.
The strategy is deliberate vertical integration. Tencent is designing its products specifically to work with Hy4 and beyond, so that mutual optimization between model and application makes the combined system more powerful than anything built on third-party models. It is the same logic that has made Apple’s silicon-software stack so formidable — except Tencent is doing it across cloud, messaging, gaming, advertising, and enterprise software.
WeChat is the obvious weapon. Tencent has started a small-scale prototype test of Xiaowei, an AI assistant embedded inside WeChat, in recent weeks. If Xiaowei rolls out at scale, it would instantly become one of the most widely used AI agents on Earth.
The Lovable Bet
While Tencent’s own model push is the headline, its investment in Lovable is arguably the more intriguing move. Lovable, a Stockholm-based “vibe coding” platform that lets users build software without traditional programming, raised $400 million in Series C funding at a $13.3 billion valuation.
The round was led by Menlo Ventures and co-led by EQT’s Scaleup Europe Fund. Tencent joined as a new investor alongside Balderton Capital, Carmignac, Kaszek Ventures, LTS Growth, World Innovation Lab, and Regent. The valuation doubled in roughly seven months.
Lovable’s numbers are staggering for a company launched in November 2024. Users have created more than 60 million projects, and Lovable-built applications receive more than 900 million visits each month. It is one of the fastest-growing software products in the world, and Tencent now has a seat at the table.
For Tencent, the investment is both financial and strategic. Lovable is a distribution engine for AI-generated applications. If those applications run on Tencent Cloud, use Hy models, and connect to WeChat ecosystems, the investment pays off many times over. It is also a way for Tencent to build influence in Europe without triggering the political scrutiny that would come with an outright acquisition.
What It Means for the Global AI War
Tencent’s bonanza is a reminder that the AI race is not just a US-China contest between OpenAI and DeepSeek. It is a multi-front war in which capital, compute, distribution, and applications all matter. Tencent has advantages in three of those four categories: capital from its gaming and ad empires, compute from its massive capex program, and distribution through WeChat and WeChat Work.
What it lacks is the global narrative. American and some Chinese AI companies have captured the attention of developers and investors with headline benchmarks and charismatic founders. Tencent has historically preferred to let its numbers do the talking. This week, the numbers are finally loud enough to be heard.
Investors are still skeptical. Tencent’s stock is down around 26% year-to-date, and the shares dipped further after the earnings call. The market is asking the same question it is asking of every AI spender: will all this capex actually produce returns? Tencent’s answer is that it could already monetize the hardware if it wanted to, but it is choosing to build something bigger.
🔥 Hot Takes
1. Tencent just proved the “AI capex bubble” thesis is half wrong. The bear case says tech giants are overbuilding compute with no clear path to returns. Tencent’s executives explicitly said they could rent their chips at a 30% profit tomorrow. They are not overbuilding because they are desperate. They are overbuilding because they see a chance to vertically integrate the entire stack and lock in China’s enterprise AI market before anyone else can.
2. Lovable is Tencent’s backdoor into the global developer brain. Buying or controlling a European AI company would trigger regulatory fireworks. Buying a small strategic stake in a hype-cycle darling is quiet, legal, and potentially more valuable. If Lovable becomes the default way non-coders build software, and those apps increasingly run on Tencent’s infrastructure, Tencent wins without ever putting its name on the product.
3. WeChat + Xiaowei is the most underhyped AI agent launch in the world. Everyone watches ChatGPT, Gemini, and Claude. But if Tencent rolls out a native AI assistant to 1.35 billion WeChat users, it will instantly become the most deployed AI agent in history. The interface is already built. The users are already trained. The distribution is already owned. The only question is whether Tencent can execute before regulators or competitors force it to slow down.
Bottom Line
Tencent’s week of AI announcements was not a coincidence. It was a coordinated statement of intent: the world’s most profitable gaming and messaging company is transforming itself into an AI infrastructure and applications powerhouse. With a $53 billion hardware splurge, a breakout model in Hy3, an even larger Hy4 on the way, and a strategic foothold in Lovable, Tencent is building a full-stack AI empire that rivals anything in Silicon Valley. The question is no longer whether Tencent will be a major player in global AI. The question is how long it will take investors to price that in.