DeepSeek Nears $7.4B Raise: China’s Biggest AI Round Yet
2 min readDeepSeek is closing in on a roughly $7.4 billion funding round, which would rank as one of the largest startup raises ever recorded in China. The deal lands a year and a half after the Hangzhou lab shocked global markets with its R1 reasoning model, and it puts a Chinese open-source player squarely on the same financial playing field as OpenAI and Anthropic.
From Side Project to Strategic Asset
DeepSeek launched as a research arm of quant hedge fund High-Flyer in 2023, and spent most of 2024 quietly publishing efficient open-weight models. Its January 2025 release of R1, a low-cost reasoning model trained for a fraction of GPT-4’s bill, briefly knocked roughly a trillion dollars off US tech market cap when investors realized that frontier-grade results no longer required hyperscaler-only budgets.
That moment turned DeepSeek into a national champion. The Chinese government has since steered domestic compute, talent, and capital toward the lab, and its open-weight releases have become reference implementations for the rest of China’s AI sector.
What the Round Looks Like
According to PYMNTS, the round is led by a consortium of state-backed investors and large Chinese tech firms, and would value the lab at a level that finally puts it in conversation with the top US frontier labs. Proceeds are earmarked for a multi-year buildout of domestic GPU clusters, training infrastructure, and recruitment, with a heavy focus on agentic and multimodal systems.
The structure also keeps DeepSeek’s open-source posture intact. People close to the deal told reporters the lab intends to keep releasing model weights publicly, even as it pushes into commercial products around its hosted API and on-device assistants.
Why It Matters
The raise reshapes the geopolitics of AI funding. US frontier labs spent the first half of 2026 stacking compute deals worth billions per month, including Anthropic’s $1.25 billion monthly SpaceX agreement and Google’s $920 million monthly tab for SpaceX-hosted GPUs. A $7.4 billion war chest gives DeepSeek the runway to chase that scale on Chinese silicon, even under tightening US export controls.
It also raises the stakes on open weights. If DeepSeek keeps shipping competitive reasoning models for free download, the pressure on closed-source incumbents to justify their pricing only grows. Expect the next round of US releases to lean harder on agentic features and tool ecosystems where pure model quality is no longer the moat.