Nathan Lambert left Ai2 and is repositioning Interconnects around three explicit goals: clarifying frontier model development, building an open model ecosystem, and constructing institutions to support both. The blog has crossed 70,000 subscribers, roughly 900 of whom pay, and the LLC bank account has hovered near zero after reinvestment. He chose not to go full-time on writing, rejecting a path similar to SemiAnalysis or Stratechery, because commentary alone cannot build the open-science movement he wants.
Lambert has signed advising agreements with two companies: Arcee AI, which builds open-weight models, and Mercor, which he connects to transparent evaluations and neutral post-training research. He frames these not as conflicts of interest but as access to close the gap between his goals and how frontier labs actually operate. His full-time work is expected to remain in the nonprofit sector. Disclosures are maintained live at interconnects.ai/about.
The piece is worth reading in full because Lambert is unusually direct about the financial mechanics of running an independent technical publication, the specific trade-offs he rejected, and why he believes raw high-voice writing outperforms polished newsletters in an era of AI-generated content. He has a stated target of 1,000 paid subscribers by summer 2026, and his reasoning for staying independent despite family financial pressure is more concrete than most writers admit publicly.
[READ ORIGINAL →]