DeepSeek kicked off the round in July targeting a 500 billion yuan ($74 billion) valuation and is set to raise more than 80 billion yuan ($11.9 billion), with CATL and Tencent among the largest committers, according to Reuters sources. The round is expected to wrap up in October and was first reported by Bloomberg. DeepSeek has partnered with Huawei on programming tools optimized for Huawei’s Ascend AI chips, positioning a China-native software stack outside Nvidia’s ecosystem. It released the DeepSeek-V4.1-Flash model last month and has hired CITIC Securities to prepare for a STAR Market IPO, though timing and valuation are uncertain.
In the US, Etched is reviewing funding offers ranging from $40 billion from top-tier investors to $50 billion, according to TechCrunch sources, just a couple of months after raising $700 million at a $21 billion valuation. The startup claims its transformer-optimized chips process more tokens faster and at lower cost than Nvidia’s, has poached engineers from Nvidia, and operates a 10-megawatt data center in Silicon Valley plus a Taiwan facility.
The open-model race is accelerating alongside the funding. Reflection AI debuted Beam, a 501-billion-parameter open-weight mixture-of-experts model with 23 billion active parameters trained on 23.8 trillion tokens; the two-year-old startup has raised about $4.7 billion at a $25 billion pre-money valuation and locked up Nvidia GB300 compute through SpaceX and Nebius through 2029. Mistral launched a 1-trillion-parameter model it says beats some Chinese rivals. The funding terms assume AI compute demand keeps outrunning supply; the Etched figures are single-sourced and should be treated with caution.