Tencent was founded in 1998 by Ma Huateng (known as Pony Ma) and four co-founders in Shenzhen, starting with an instant messaging service called OICQ, later renamed QQ. What followed over the next two decades was one of the most remarkable corporate buildouts in tech history. QQ grew into WeChat (Weixin in Chinese), which became not just a messaging app but essentially the operating system of daily life in China — handling payments, ride-hailing, government services, social media, shopping, and more for over 1.3 billion monthly active users. Tencent simultaneously built the world's largest gaming empire by revenue, with stakes in or full ownership of Riot Games (League of Legends), Supercell (Clash of Clans), Epic Games (Fortnite/Unreal Engine), and dozens of other studios. By 2024, Tencent was one of the most valuable companies in the world, with a market cap regularly exceeding $400 billion and fingers in virtually every sector of the Chinese digital economy.
Tencent's entry into the foundation model race came through Hunyuan, a family of models spanning text, image, video, and 3D generation. Hunyuan-Large, their flagship language model, was built as a Mixture of Experts architecture with reportedly over 300 billion total parameters. What distinguished Hunyuan from some competing Chinese models was Tencent's emphasis on practical deployment: these models were designed from the start to integrate into Tencent's existing products rather than exist as standalone research showcases. Hunyuan powers AI features in Tencent's cloud services, WeChat's smart reply and translation features, enterprise collaboration tools, and advertising optimization. The company also open-sourced several Hunyuan models, including Hunyuan-7B and image generation models, in a move that aligned with the broader trend of Chinese tech giants releasing open-weight models to build ecosystem influence.
Where Tencent has arguably made its biggest splash in generative AI is video. HunyuanVideo, their text-to-video model released in late 2024, immediately drew attention for producing high-quality, temporally consistent video that competed with dedicated video generation companies like Runway, Kling, and Pika. The model was notable for its open-source release, which was unusual for a video generation model at that quality level — most competitors kept their best video models proprietary. Tencent followed up with improvements and specialized variants, leveraging their deep expertise in multimedia from decades of running one of China's largest video streaming and short-form video platforms. The gaming division also contributed AI-generated content tools, using generative models to create game assets, NPC dialogue, and environment textures at a pace that traditional art pipelines could not match.
Tencent Cloud, while smaller than Alibaba Cloud in the Chinese market, has been aggressively positioning itself as an AI-first cloud platform. The company invested billions in GPU clusters and custom training infrastructure, and offers Hunyuan models as managed services alongside support for popular open-source models. Tencent's AI infrastructure also includes specialized chips — their Zixiao AI accelerator, designed in-house, targets inference workloads across the company's own services. This vertical integration (owning the models, the infrastructure, and the distribution through apps with billions of users) gives Tencent a flywheel that pure-play AI companies cannot replicate: every improvement in their models immediately reaches an enormous user base, generating feedback data that improves the next iteration.
Tencent's greatest strength in AI is the same thing that defines the company overall: distribution. When you have 1.3 billion WeChat users, getting AI features in front of people is trivial. When you own the platforms where a significant fraction of China's digital commerce, communication, and entertainment happens, training data is abundant and feedback loops are tight. But Tencent also operates under constraints that Silicon Valley labs do not. Chinese regulatory oversight of AI is extensive, requiring approval for public-facing generative AI services and imposing content restrictions that shape what models can and cannot produce. Tencent's enormous scale also means it moves more cautiously than startups — a broken AI feature in WeChat affects a billion people, so the bar for deployment is necessarily higher. Internationally, Tencent's AI ambitions are complicated by U.S.-China tensions and restrictions on advanced chip exports, though the company's investments in gaming studios and cloud infrastructure outside China give it global reach that many Chinese AI companies lack. The real question for Tencent is not whether they can build competitive AI — they clearly can — but whether they will lead the frontier or remain a fast follower who excels at integration and deployment rather than pure research breakthroughs.
