DeepSeek's Liang Wenfeng Pledges Sole Focus on AGI Over Profits

In a recent leaked recording of a nearly 4-hour closed-door investor meeting, DeepSeek founder Liang Wenfeng rarely detailed the company's strategy. He explicitly stated that DeepSeek was not created to make money; the company's sole mainline is to advance along the path to AGI and increase the success rate of AGI, rather than chasing short-term product growth or commercialization. Since Chinese AI entrepreneurs rarely share their visions publicly at length, this recording provides a rare window into DeepSeek's strategic thinking.

Confirmed

In the recording and leaked internal talks, DeepSeek demonstrated a highly pure AGI orientation and strategic determination:

• Strategic Priorities and Roadmap: Liang Wenfeng believes current AI opportunities are merely "sesame seeds," and the real massive opportunity lies beyond AGI. The company's evolution path roughly follows: Chain of Thought (CoT) reasoning, Agents, and Continuous Learning. He thinks it is difficult to reach AGI directly with the current route; models need to continuously absorb new knowledge over longer time scales like humans, rather than relying solely on one-time intensive training. These goals rank ahead of product and growth.
• Tech and Product Trade-offs: Liang explicitly pointed out that 3D generation, video generation, and world models are not on the company's main path. Whether to value these directions depends on the definition of "intelligence" (e.g., whether it includes holographic cabins or robotics). Furthermore, the company views LLM "hallucinations" as a product issue, but it is currently a low priority. To advance core goals, DeepSeek's current priority is developing a Coding Agent (Liang was previously conservative about this direction but has changed his stance).
• Open Source and Restraint: Liang emphasized that open source is DeepSeek's core agenda. The company will definitely open-source its strongest models, believing closed-source holds no essential advantage. They build competitiveness by open-sourcing strong models, recovering only hardware costs via APIs, and focusing on engineering efficiency. He also stated that "restraint" is the way to survive, and large model labs attempting to "swallow the world" will eventually be eliminated. This restraint is also viewed externally as a unique Eastern philosophy of tech development.
• Business Model and Computing Power Alternatives: Liang revealed a current profit margin of about 6x and a payback period of about 10 months. Regarding computing power, he believes the CUDA ecosystem barrier is continuously weakening, and Huawei's Ascend has the opportunity to become a viable alternative; even if costs are higher, domestic substitution will be an irreversible trend given computing constraints.
• Team Organization Philosophy: Liang disclosed that the company is not a traditional hierarchical organization, but "vision-driven and organized around the vision," going so far as to say there is "no organization, only vision."

Unconfirmed

Although media reports suggest the leaked closed-door investor meeting records are highly likely to be authentic, DeepSeek has not officially confirmed the content.

Why it matters

Amid the current industry frenzy to push for commercialization, monetization, and the rapid application of large models, DeepSeek has shown a distinctly different strategic determination. Betting all resources on the evolution of underlying intelligence and AGI not only deeply impacts its own product rhythm (such as delaying hallucination fixes and heavily investing in Coding Agents) but also provides a highly valuable, non-mainstream reference sample for understanding the future development path of Artificial General Intelligence.

2026-07-23 ~ 2026-07-25 · 18 related posts

• DeepSeek says hallucinations are a product issue, but coding agents are the priority — teortaxesTex · 2026-07-23
• DeepSeek founder says 3D, video, and world models are off the main path — Scobleizer · 2026-07-23
• [source] DeepSeek founder Liang Wenfeng says the company exists to maximize AGI odds, not profit — 智东西 · 2026-07-23
• DeepSeek post says AGI opportunities will dwarf today’s AI market — zephyr_z9 · 2026-07-23
• Wenfeng says DeepSeek will keep open-sourcing its best model — basedjensen · 2026-07-23
• DeepSeek’s Liang Wenfeng says AGI matters more than products or growth — MagicZhang · 2026-07-23
• A nearly four-hour talk gives a rare look into DeepSeek founder Liang's thinking — pstAsiatech · 2026-07-23
• Liang says the company is “vision-driven,” not organized by hierarchy — pstAsiatech · 2026-07-23
• Liang Wenfeng’s investor interview lays out a clear case for open-weights AI — inductionheads · 2026-07-23
• DeepSeek’s reported AGI roadmap puts continual learning ahead of products and growth — SwordfishGreedy1945 · 2026-07-23
• Liang Wenfeng’s DeepSeek investor call doubles down on open source and restraint — teortaxesTex · 2026-07-23
• [source] DeepSeek founder says open source, low margins and coding agents are the road to AGI — AlchainHust · 2026-07-23
• Liang Wenfeng says AGI needs continuous learning, not just bigger training runs — zephyr_z9 · 2026-07-24
• Takeaways from DeepSeek Founder Liang Wenfeng's Investor Meeting — cpaik · 2026-07-24
• DeepSeek founder Liang Wenfeng says CUDA’s moat is weakening and Ascend may emerge as an alternative — sudoraohacker · 2026-07-24
• DeepSeek’s Liang Wenfeng says AGI, not profits, is the company’s real mission — MarvinTBaumann · 2026-07-24
• DeepSeek is still not multimodal, and that gap makes little sense, says commentator — dotey · 2026-07-24
• [source] DeepSeek Founder: Open Source Doesn't Hurt Revenue, No Overtime Needed — dr_alphalyrae · 2026-07-25