DeepSeek will not compete with Tencent or ByteDance for consumer revenue, instead channeling resources into artificial general intelligence research, founder Liang Wenfeng told investors in a leaked four-hour meeting.
DeepSeek will not compete with Tencent or ByteDance for consumer revenue, instead channeling resources into artificial general intelligence research, founder Liang Wenfeng told investors in a leaked four-hour meeting.

DeepSeek will not compete with Tencent or ByteDance for consumer revenue, instead channeling resources into artificial general intelligence research, founder Liang Wenfeng told investors in a leaked four-hour meeting.
DeepSeek's decision to bypass the consumer AI market removes a competitive overhang from Tencent and ByteDance, as the startup redirects resources toward artificial general intelligence research rather than commercial products.
"DeepSeek does not aim to become the next ByteDance or Tencent," Liang Wenfeng, founder of DeepSeek, said during the meeting. "Maximizing commercial returns is not an imminent goal."
The company will not prioritize investment in areas such as 3D generation, video generation or world models, Liang said, adding that products are not the firm's current main focus. Instead, DeepSeek hopes to increase the probability of achieving AGI by reducing spending on non-core businesses. On the China-US AI gap, Liang said the difference lies mainly in resources rather than talent, and he aims to narrow it through higher computing efficiency.
The strategy leaves Tencent and ByteDance with one fewer well-funded rival in the consumer AI market, where Tencent's WorkBuddy office agent has already captured more than 20 million monthly visits on China's PC side, ranking first in market share, according to Analysys data. Tencent shares rose 2% on the day of the leak.
The AGI-first approach contrasts sharply with the direction of DeepSeek's peers. At the Shanghai WAIC conference this month, Tencent heavily promoted WorkBuddy and its CoderBuddy coding agent, while Alibaba's Qwen integrated three office agent products — QoderWork, Wukong and MuleRun — into a dedicated workplace suite. StepStar, one of China's "Six Rising AI Stars," launched STEPX, a large-model-native AI terminal brand, and its Step AOS operating system for agents, signaling a full pivot from pure model development to hardware and agent products.
Meanwhile, the model arms race continues behind the scenes. Kimi released its K3 model on July 16 with 2.8 trillion total parameters and roughly 32 billion activated, scoring above Zhipu's GLM-5.2 on multiple public benchmarks. Zhipu's market value fell 42.47% over two trading days following the K3 launch, closing at HK$890.5 on July 20. Tencent also released its HY3 model, a 295-billion-parameter open-weight system under the Apache 2.0 license, scoring 75.8 on the Swaybench Multilingual test and 57.9 on Swaybench Pro.
DeepSeek's decision to sit out the consumer product race carries implications for investors. Tencent, trading with short selling of $5.02 billion and a short ratio of 16.85%, faces reduced competitive pressure from a startup that could have challenged its WorkBuddy franchise. ByteDance, which did not exhibit at WAIC but showcased its Doubao Phone 2.0 prototype at ZTE's booth, also benefits from one fewer rival in the AI assistant market. For DeepSeek, the bet is that AGI breakthroughs will ultimately deliver more value than near-term consumer revenue — a thesis that may take years to validate.
This article is for informational purposes only and does not constitute investment advice.