The Contrarian
Founder of DeepSeek, who can show that a frontier model is built on open weights and efficiency instead of the largest compute budget.
He builds open-weight models and argues that discipline and openness beat scale and secrecy.
On 24 April 2026 DeepSeek released a preview of V4 as open weights, in Pro and Flash versions, adapted to run on Huawei’s Ascend chips.
Market Status reads Liang Wenfeng as The Contrarian. The friction this archetype resolves is the belief that frontier capability requires the largest compute budget and a closed model.
He is the founder of DeepSeek, which was separated from the AI research lab of the hedge fund High-Flyer, and he leads both. DeepSeek released R1 under an MIT licence in January 2025 and the V4 preview under open weights in April 2026. External financing, a first for the company, was reported to have opened in mid-April 2026.
His stated aim is AGI built openly. In a late July 2026 investor meeting, as summarised, he said that DeepSeek will release its strongest models and will not open-source a weaker one while keeping a better one for itself. He also said the talent gap is fundamentally a compute gap and that DeepSeek is about two years behind the United States.
He champions efficiency, yet he told investors that chip scarcity caps what DeepSeek can buy and that money is not the constraint. V4 was designed for domestic chips and was still trained, at least in part, on Nvidia hardware. The transcript behind the investor-meeting summary has not been confirmed by DeepSeek.
Against the journey, Liang is Building. An open-weight model is a public reference that any organisation can inspect and run, and the Data Economy’s open standards were the Referenceability that let strangers rely on a shared protocol. Model weights that anyone can audit are a candidate for the same role at the model layer, with the open licence as the boundary. His work helps introduce an inspectable model layer for the Technology, Legal and Risk Management functions, while the question of which chips it runs on sets who can afford to use it.
Released a preview of DeepSeek V4 as open weights, in Pro and Flash versions, adapted to run on Huawei’s Ascend chips. Quartz
DeepSeek opened its first external financing, as reported by insiders. Reported secondhand. ChinaTalk
In an investor meeting, said all differences between labs can be attributed to compute resources. Reported from a transcript DeepSeek has not confirmed. NYU Shanghai RITS
Entries are added as new public statements appear. Last updated 19 October 2026.
Funding the build, for readers who can direct capital toward the containerization.
Developing the micro-container that holds the gap, for readers who can make the technology.
Helping the micro-container reach scale, for readers who can set standards, convene consensus or move the political and regulatory conditions.
Note: The perspectives shared on GreenDeveX.com are the author's independent analytical interpretations of trust as a historical medium of exchange. This commentary is intended for conceptual exploration and does not constitute peer-reviewed academic research or commissioned institutional findings.