Groq raises $350M at $3.5B, half its 2025 value
The inference market just got a reality check, and Alibaba's Qwen team quietly shipped an open-source tool that wants to remember how you work.
Groq has closed a $350 million round at a $3.5 billion valuation — roughly half the $6.9 billion it commanded in September 2025, before Nvidia struck a licensing deal and poached much of its talent. Bloomberg, citing people familiar with the matter, reports the round was led by Disruptive, and that the markdown reflects how the chipmaker's own moves reshaped the competitive picture for Groq's custom LPU inference hardware. The down round is a sharp signal that the inference-market euphoria of 2025 has cooled: a company once pitched as the fast alternative to GPUs is now raising on defensive terms. If even a marquee name like Groq has to halve its price to get a check, the rest of the inference-hardware field should be watching its own burn rate.
Alibaba's Qwen Office team has open-sourced MyContext, a local-first tool that distills a "second you" from your Feishu and DingTalk work history. According to Zhidx, the project — Qwen Office's first open-source release, already past 1,000 GitHub stars — pulls in chats, docs, and meeting notes, then extracts your work patterns, routines, and red lines so an agent can draft replies in your voice. The clever bit is its safety posture: the model only generates, it can't send, and every conclusion must be tied to a source message, with the user's own edits flagged as unoverridable. It's a developer preview with rough edges (graph generation failed in the reporter's own run), but it points at where assistant software is heading — from doing tasks on command to quietly remembering how you work.
Groq's haircut suggests the inference gold rush is over — or just paused. Would you bet on custom AI silicon right now, or is the GPU the only safe horse? Tell us in the comments.
Sources: Bloomberg · Techmeme · Zhidx · MyContext (GitHub)
