Question 1

How much money has DeepSeek raised in total?

Accepted Answer

With its June 2026 Series B of $7.4 billion at a $59 billion valuation — led by Tencent with participation from state-backed Chinese technology funds — DeepSeek has raised approximately $7.5 billion in total. The company was founded as a research initiative within High-Flyer Capital Management, a Hangzhou-based quantitative trading firm, and initially operated without formal external funding, benefiting from High-Flyer's existing NVIDIA GPU infrastructure. The June 2026 round is the largest single AI funding event in Chinese AI history and one of the largest globally, trailing only OpenAI's 2025 $40 billion round. The capital is earmarked for training infrastructure expansion, talent acquisition, safety research, and international enterprise go-to-market expansion.

Question 2

Why does DeepSeek need $7.4 billion if open source AI is supposed to be free?

Accepted Answer

Open source AI is free to use, not free to build. DeepSeek's models are released under MIT licenses that allow anyone to download, run, and modify them without fees — but creating those models requires massive upfront capital. Frontier training runs cost $50 million to $500 million in compute alone. PhD-level researchers cost $500,000 to $2 million annually each. Safety and alignment research requires dedicated teams and infrastructure. DeepSeek's widely cited $6 million V3 training cost was the marginal cost of one training run on compute DeepSeek already owned from parent company High-Flyer Capital — not the economic cost of building the capability. The $7.4 billion round is capital for the next generation of frontier capability, and its scale is consistent with every other frontier AI lab regardless of license type.

Question 3

What is the relationship between DeepSeek and High-Flyer Capital?

Accepted Answer

High-Flyer Capital Management is a Hangzhou-based quantitative hedge fund that built one of China's largest private GPU clusters — estimated at 10,000 NVIDIA A100s acquired before U.S. export controls tightened — to power its algorithmic trading models. DeepSeek was founded in 2023 as a research spin-off within High-Flyer, led by Liang Wenfeng, High-Flyer's founder. DeepSeek's structural advantage was access to High-Flyer's GPU infrastructure at zero marginal cost, which is why the $6 million V3 training cost understates total economic cost — the hardware investment was already sunk into High-Flyer's balance sheet. The June 2026 funding round marks DeepSeek's transition to an independent entity with its own capitalization, building its own compute base rather than operating on inherited assets from the parent fund.

Question 4

How does DeepSeek's valuation compare to other AI companies?

Accepted Answer

At $59 billion, DeepSeek's valuation sits near Anthropic's most recent valuation range and represents roughly one-fifth of OpenAI's $300 billion valuation. Mistral AI, the closest European open-source peer, is valued at approximately $6 billion — roughly one-tenth of DeepSeek's figure. The premium versus Mistral reflects DeepSeek's frontier model capability matching GPT-4o on most benchmarks, plus the geopolitical premium attached to Chinese AI independence at scale with state investment signaling both capital commitment and strategic national priority. Cohere, the enterprise open-weights provider, is valued at approximately $5 billion — DeepSeek at $59 billion implies the market values compute-efficient frontier open-source models at a substantial premium to earlier open-source-only positioning.

Question 5

What does DeepSeek's $7.4B round mean for OpenAI and Anthropic?

Accepted Answer

The direct competitive pressure falls on API pricing margins. DeepSeek's API is priced 80-95% below OpenAI's equivalent tiers for comparable capability — a differential sustainable through compute efficiency, state subsidy, and willingness to operate at lower margins during growth. For OpenAI, the risk is margin compression on its enterprise API business as cost-sensitive developers route commodity workloads to DeepSeek-compatible endpoints; OpenAI's response is accelerating application-layer products like ChatGPT, Codex, and Operator that are less directly commoditizable than raw model API access. For Anthropic, the safety-premium thesis remains intact but DeepSeek's own $500 million safety research allocation begins to challenge the safety moat over a 2-3 year horizon. For Google, frontier open-source inference displaces Google Cloud API revenue when developers run DeepSeek models on alternative infrastructure.