Tencent executive responds to AI slowdown: Insufficient computing power has held back Hunyuan

Beating AI News (from Dongcha) – Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of Cloud and Smart Industries Group, published a lengthy article directly addressing the claim that "Tencent is slow in AI development." He acknowledged internal anxiety at Tencent, noting that rival firms have gained more momentum in recent years. More directly, Tencent’s overall computing power is "severely insufficient," which has slowed the training of its Hunyuan model and product development, "having a relatively large impact." However, Tang believes this is likely only the first kilometer of the AI marathon, "sticking it out for longer is more important than starting early." He previewed that the next-generation Hunyuan Hy4 will bring more breakthroughs, while arguing Tencent cannot compete solely on foundational models: algorithms set the upper limit, engineering determines implementation speed, and real-world scenarios such as WeChat’s office, documents, meetings, and enterprise services are Tencent’s biggest asset. Tang also rarely reviewed several mistakes: over the past year, Tencent’s AI assistant Yuanbao invested heavily in promotion for user growth, but "neither the model nor the product was ready" at the time, leading to unsatisfactory results. CodeBuddy and WorkBuddy were not pre-planned successful paths; early-stage DevTools operated at a loss for a long time and was almost cut. WorkBuddy later capitalized on the agent wave, rolling out over 40 versions in three months after launch. Tang said the team now uses AI extensively for coding, with humans mainly responsible for judgment, debugging, and oversight. He predicts the boundaries between product, development, and testing will become increasingly blurred in the future, and small teams will be able to accomplish tasks that only large teams could do before with AI’s help.