Inside Liang Wenfeng's Master Plan: Why DeepSeek Will Never Churn Profits, Build a Super App, or Close Its Models

Open Source is Non-Negotiable

Liang firmly committed DeepSeek to keeping its most advanced models open-source, arguing that open-source development and commercial monetization are not mutually exclusive . He described open-source as "ceding profit" — a deliberate trade-off that yields internal benefits (employee pride, company cohesion) and external benefits (the broader AI community benefits) . He explicitly rejected going closed-source, saying restraint itself is a strategy to maximize the probability of achieving AGI .

The AGI Roadmap: Step by Step

Liang stated that DeepSeek's sole long-term vision is AGI — not building a consumer platform, a super app, or maximizing near-term revenue . He laid out a staged technical roadmap:

• CoT (Chain-of-Thought): the "staircase step" of 2025 .
• Agent: the current staircase step for 2026 .
• Continuous learning: the next major problem to solve after Agent .
• Singularity → Embodied intelligence: the longer-term progression beyond continuous learning .

He explicitly ruled out investing in 3D generation, video generation, or world models as distractions from the AGI path .

Philosophy of 'Restraint as Strategy'

Liang rejected monopoly-style profit maximization. His API pricing philosophy was straightforward: price to recover hardware costs in roughly ten months — a reasonable profit, not a profit-maximizing one . He noted that demand in this price range is inelastic — even doubling the price would not meaningfully reduce token consumption — but he chose not to do so . When DeepSeek once cut a model's price to one-quarter, the team celebrated; affordability itself was the goal .

On monopoly, he said the company is not fighting for consumer traffic, not chasing B-end market trends, and not trying to dominate any vertical . His core argument: "The more restrained you are, the more likely you are to succeed" . The only monopoly he cares about is the one that matters for AGI progress — access to computing power, which he called DeepSeek's biggest constraint, noting they operate at roughly 1/20th the compute of U.S. frontier labs with about 20,000 H-equivalent GPUs, trailing the U.S. by 12-18 months .