AI News Briefing — June 17, 2026

Chinese AI lab DeepSeek raised over 50 billion yuan (about 7.4 billion dollars) in its first-ever external funding round at a roughly 50 billion dollar valuation. It marks a major capital influx for a leading open-weight model developer that had previously stayed self-funded.

Leaked audited financials show OpenAI revenue grew from 3.7 billion dollars in 2024 to 13.07 billion in 2025, approaching nearly 2 billion in monthly revenue by year-end, but expenses (especially R&D) far outpace income as it preps for an IPO. The documents reveal multi-billion-dollar annual losses.

AI Business covers SpaceX's 60 billion dollar Cursor acquisition as a push into agentic coding, gaining access to Cursor's developer workflow and user analytics. It frames the deal as expanding SpaceX's developer offerings.

Qwen released Qwen-Robot-Suite, three embodied-AI foundation models built on its vision-language backbones: RobotManip (VLA manipulation on Qwen3.5-4B), RobotWorld (a language-conditioned video world model), and RobotNav (navigation at 2B/4B/8B), with two shipping public GitHub repos. It targets fragmented robotics data with a unified open suite.

Microsoft patched a max-critical M365 Copilot flaw that researchers showed could exfiltrate 2FA codes and other sensitive email data via prompt injection. The case underscores the unsolved problem of LLMs failing to separate trusted user instructions from malicious third-party content.

OpenAI introduced Deployment Simulation, a method that predicts AI model behavior before release by simulating deployment using real conversation data to improve safety and evaluation accuracy. It aims to catch behavioral issues pre-launch.

The Decoder reports OpenAI spent 34 billion dollars over the past year, far more than the prior year, underscoring the enormous cash burn behind its growth ahead of a potential IPO. It complements leaked financials showing expenses outpacing revenue.