DeepSeek’s $70M Revenue Surge Exposes AI’s Margin Reality
DeepSeek’s 10x revenue jump hides a $100M+ loss. Builders face the same math: growth now, profitability later—if ever.
The story of the day
DeepSeek’s 475M yuan ($70.7M) revenue in seven months—a 10x leap from 2025—masks a 715M yuan net loss, per The Information. The Chinese lab is now chasing a 50B yuan raise at a 500B yuan valuation, proving AI’s capital hunger isn’t just a Western phenomenon. For builders, the takeaway is brutal: even aggressive growth doesn’t outpace infrastructure costs. The question isn’t whether DeepSeek can scale, but whether any LLM shop can escape the burn rate.
New tools & launches
Flow AI Video Generator — Text-to-video for creators, no pricing disclosed yet. [Hacker News]
Models & research
No new model releases or research papers met the signal threshold today.
Industry pulse
Paul Dix (InfluxData) argues AI can produce and refine million-line codebases—if verification systems and clear direction are in place. The caveat: this only works when the "oracle" (e.g., a reference implementation) exists. For most builders, that’s a luxury. [Simon Willison]
Stat of the day
DeepSeek’s $70.7M revenue (Jan–Jul 2026) vs. $103M net loss. The ratio: ~1.45x burn. [The Information]
Get the full analysis — subscribe to Gradient Pro.
