Artificial Intelligence · Profile from AI Research Companion
DeepSeek
The Hangzhou model lab that Liang Wenfeng spun out of his quantitative hedge fund High-Flyer in July 2023. It was funded for almost three years by the fund's profits and its stock of Nvidia GPUs, many bought before US export controls. It became the most famous Chinese AI company in the world in January 2025, when its open-weight reasoning model R1 and a claim that its V3 base model's final training run cost only ~$5.6m (company claim) helped wipe ~$589bn off Nvidia's market value in a single day. DeepSeek's business is unusual. Its staff is small (~160–200 people in 2025, reported). It released its best models under the MIT licence, priced its API far below US rivals, did little marketing, and for most of its life had no outside investors. In 2026 that changed. DeepSeek took its first outside money (~RMB 50bn, ~$7.4bn, at a reported ~$52bn post-money (FT) to up to ~$59bn (Reuters), mid-2026), raised prices for the first time (August 2026), reportedly reached a ~$1bn annualised revenue run rate (September 2026, reported) and began preparing a Shanghai listing. Its controversies are as large as its achievements: how much its models really cost, data-security bans in a dozen or more countries, allegations that it distilled US models and obtained smuggled Nvidia chips, and its shift under state encouragement to Huawei hardware.
At a glance
- Founded
- High-Flyer AGI lab announced 14 April 2023; incorporated as Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co. on 17 July 2023 (company filings, reported)
- Headquarters
- Hangzhou, Zhejiang, with an office in Beijing
- Ownership / structure
- Private. Liang Wenfeng held ~84% through two holding entities in 2024 (business registry, reported), the rest held by High-Flyer-linked partners. First outside round mid-2026 (~RMB 50bn; Liang reportedly put in ~RMB 20bn himself; Tencent ~RMB 10bn, CATL ~RMB 5bn, plus JD.com, NetEase and China's national AI fund, reported); second round ~RMB 100bn (~$15bn) raised and 'set to complete in the coming days' (Reuters, 22 October 2026) at ~RMB 500bn (~$74bn) pre-money, ~RMB 600bn (~$84bn) post-money if fully closed (Bloomberg, Reuters; final close not confirmed — re-checked Nov 2026)
- Leadership (2026)
- Liang Wenfeng (founder, controlling shareholder and CEO; co-founder of High-Flyer). DeepSeek publishes few executive titles; its research papers list hundreds of contributors
- Products
- Open-weight language models: DeepSeek-Coder and DeepSeek-LLM (Nov 2023), MoE (Jan 2024), Math (Apr 2024, introduced GRPO), V2 (May 2024), V3 (Dec 2024), R1 (20 Jan 2025), R1-0528 (May 2025), V3.1 (Aug 2025), V3.2-Exp (Sep 2025), V3.2 and V3.2-Speciale (1 Dec 2025), V4 preview (24 Apr 2026), V4-Flash (31 Jul 2026) and V4-Pro (13 Aug 2026) releases, V4.1-Flash (10 Sep 2026); free chatbot app and website; paid API
- Scale
- ~160–200 staff (2025, reported); app briefly No. 1 on the US App Store (27 Jan 2025); ~130m monthly active users in China (QuestMobile, June 2026, reported); annualised revenue run rate ~$1bn (The Information, September 2026; unaudited, reported)
- Market value
- No public valuation until 2026: ~$10bn target in early talks (April 2026), ~$52–59bn in the first round (May–July 2026), ~$74bn pre-money in the second (October 2026) (all reported)
- Signature moves (incl. failures)
- High-Flyer and the Fire-Flyer clusters (2015–23); V2 and the China price war (2024); V3 and the cost claim (2024); R1 and the market shock (2025); Open Source Week and the 545% margin claim (2025); data-security bans abroad (2025–26); distillation and chip-smuggling allegations (2025–26); the Huawei Ascend shift and the R2 delay (2025–26); V4, the price rise and first outside funding (2026)
9 strategic moves analysed
Each one gets a full strategy file in the companion: context, the decision, the financing, what happened, and the lesson.
2015
High Flyer and the Fire Flyer Clusters
Founding (a hedge fund builds a GPU fleet, then spins out an AGI lab)
2024
DeepSeek V2 and the China Price War
Model release and pricing
2024
DeepSeek V3 and the Cost Claim
Model release (open weights, with a technical report that disclosed a training cost)
2025
Data Security Bans Abroad
Regulatory action (foreign privacy orders, government-device bans, app-store delisting requests)
2025
Distillation and Chip Smuggling Allegations
Controversy and regulatory scrutiny (accusations by US labs, officials and Congress; denials by DeepSeek and Nvidia)
2025
Huawei Ascend Shift and the R2 Delay
Compute strategy (moving from Nvidia to Chinese accelerators, under state encouragement)
2025
Open Source Week and the 545 Percent Margin Claim
Launch (open-sourcing of infrastructure code) and disclosure (inference economics)
2025
R1 Open Reasoning and the Market Shock
Model release (open-weight reasoning model, MIT licence) and consumer app launch
2026
V4 Price Rise and First Outside Funding
Model release, pricing change and funding round (first outside capital; IPO preparation)