DeepSeek
Company History
DeepSeek History: Founding, Growth & Key Milestones
Founded 2023 in Hangzhou, China
When was DeepSeek founded?
DeepSeek was founded in 2023 in Hangzhou, China by Liang Wenfeng, making it 3 years old.
- Founded
- 2023
- Founding place
- Hangzhou, China
- Founders
- Liang Wenfeng
- Headquarters today
- Hangzhou, Zhejiang, China
- Milestones documented
- 9 (2023–2026)
- Current CEO
- Liang Wenfeng
| Year | Event |
|---|---|
| 2023 | DeepSeek spun out of High-Flyer |
| 2023 | First open models |
| 2024 | DeepSeek-V2 introduces MLA |
| 2024 | DeepSeek-V3 launched |
| 2025 | DeepSeek-R1 and chatbot launch |
| 2025 | V3.1 and V3.2 |
| 2026 | DeepSeek-V4 family |
| 2026 | First outside funding and IPO preparation |
| 2026 | Huawei Ascend software toolkit |
- Where was DeepSeek founded?
- DeepSeek was founded in Hangzhou, China in 2023. wikidata.org
- Who founded DeepSeek?
- DeepSeek was founded by Liang Wenfeng.
- When did DeepSeek start?
- DeepSeek was founded in 2023.
- How old is DeepSeek?
- DeepSeek is 3 years old as of 2026.
Founding Story
Liang Wenfeng studied electronic information engineering at Zhejiang University and co-founded High-Flyer in 2015 to trade with machine-learning models. By 2021 he was buying Nvidia GPUs in bulk, reportedly about 10,000 A100s before US export controls tightened. In April 2023 High-Flyer announced an AGI research lab; on 17 July 2023 it was spun off as DeepSeek. Venture investors initially passed, doubting a quick exit, so High-Flyer and Liang funded it themselves.
Founders
Liang Wenfeng
Liang Wenfeng founded DeepSeek in 2023. Leveraging the massive GPU clusters and algorithmic expertise he built at his quantitative trading firm, High-Flyer, he aimed to build a purely Chinese, open-source AI rival to OpenAI. Under his leadership, DeepSeek rapidly trained the highly efficient DeepSeek-V2 and DeepSeek-Coder models, aggressively undercutting Western AI API pricing and establishing the company as a dominant force in China's generative AI ecosystem. Liang Wenfeng was born in 1985 in Zhanjiang. Education: Zhejiang University; Zhejiang University. Liang was born in 1985 and attended Zhejiang University. In 2025 he was named to Time magazine's list of the 100 most influential people.
Key Milestones
DeepSeek spun out of High-Flyer
High-Flyer announced an AGI lab in April 2023; it became the independent company DeepSeek on 17 July 2023.
Source: en.wikipedia.orgDeepSeek-V2 introduces MLA
May 2024 release introduced Multi-head Latent Attention and DeepSeekMoE and triggered a Chinese API price war.
Source: github.comDeepSeek-V3 launched
December 2024: 671B-parameter MoE model; final training run reported at ~$5.6M of H800 GPU time.
Source: github.comDeepSeek-R1 and chatbot launch
January 2025 reasoning model rivalled OpenAI o1; the app hit No. 1 on the US App Store and Nvidia fell ~17% in a day.
Source: github.comV3.1 and V3.2
V3.1 added a hybrid thinking/non-thinking mode; V3.2 introduced sparse attention for long-context efficiency.
Source: api-docs.deepseek.comDeepSeek-V4 family
24 April 2026: V4-Pro and V4-Flash released as MIT-licensed open weights with a 1M-token context; V4-Pro reached general availability in August 2026.
Source: api-docs.deepseek.comFirst outside funding and IPO preparation
June 2026 Series A of ~50 billion yuan (~$7.4B) at a valuation near $50B; Bloomberg reported IPO preparations in July for a 2027 debut.
Source: techcrunch.comHuawei Ascend software toolkit
30 September 2026: open-sourced programming tools for Huawei Ascend chips, an alternative to Nvidia's CUDA stack.
Source: qz.comKeep researching DeepSeek
DeepSeek Company History FAQ
What were the major corporate milestones of DeepSeek in 2023?
High-Flyer announced an AGI lab in April 2023, it became DeepSeek on 17 July 2023, and in November 2023 it released DeepSeek-Coder and DeepSeek-LLM 67B.
When did DeepSeek introduce Multi-head Latent Attention (MLA) with DeepSeek-V2?
In May 2024, DeepSeek released DeepSeek-V2, introducing Multi-head Latent Attention (MLA) and DeepSeekMoE, slashing KV cache memory by 93% and cutting API pricing to industry-low levels, sparking a major AI price war.
When did DeepSeek release the flagship DeepSeek-V3 model?
In December 2024, DeepSeek launched DeepSeek-V3, a 671B-parameter MoE model trained for under $6 million on 2,048 H800 GPUs, matching GPT-4o and Claude 3.5 Sonnet across reasoning, math, and coding benchmarks.
What occurred during the landmark release of DeepSeek-R1 in January 2025?
In January 2025, DeepSeek released DeepSeek-R1, matching OpenAI's o1 reasoning capabilities through pure reinforcement learning, and released distilled reasoning models under the MIT license, creating a worldwide sensation.
How did DeepSeek-R1's release impact global technology markets in early 2025?
DeepSeek-R1's extreme compute efficiency demonstrated that frontier reasoning can be achieved with significantly less hardware, triggering a historic single-day market repricing across global chipmakers and cloud providers.
When did DeepSeek top the global Apple App Store charts?
In January 2025, DeepSeek's mobile AI assistant app surged to become the #1 most downloaded free application on the Apple App Store in the United States and dozens of countries worldwide, surpassing ChatGPT.
What did DeepSeek release in 2026?
The V4 family on 24 April 2026 (V4-Pro and V4-Flash, MIT-licensed, 1M-token context), V4-Pro general availability in August, an experimental vision model, and Huawei Ascend chip tools on 30 September 2026.
How did DeepSeek start?
DeepSeek was founded in Beijing in 2023, directly backed by the massive Chinese quantitative hedge fund High-Flyer Quant. The founders essentially utilized their immense financial resources and massive existing GPU clusters to pivot from financial algorithms directly into building massive Large Language Models.
Related Strategic History
Compare founding milestones and corporate evolution
Explore DeepSeek
Core profile pages, founders and related research hubs for this company.