Open-source AI without Jensen (Interconnects) and the post-Huang week
TL;DR. Nathan Lambert argues Jensen Huang's resignation (11-12) is a major setback for US open models. Nvidia still has a business reason to back open weights (keeping its biggest customers from concentrating power and building their own chips), but Huang was the founder-style patron who funded the zone and used his power to stop bad outcomes, especially sloppy state-level regulation. Lambert's economics: with compute scarce and inference startups starved of chips, open models are still served at large margins (DeepSeek's DeepInfer showed this), so they are not yet a commodity, and closed labs with scale and chip-to-datacenter integration win. He quotes Dylan Patel: the US problem is talent, not money, and Moonshot is as skilled as US labs, held back only by compute. Inkling 2 Small is the best Western open-weight model in a while but far behind Kimi K3.1 Pro. Thinking Machines' next round at about $40B is reportedly done. His remedy: a truly frontier US open-model player, plus hyperscaler allies (Zuckerberg, Jassy, Nadella).
Same-window context
- A reinstatement petition passed 100,000 signatures, including 5,248 self-identified Nvidia employees (The Information).
- The Information's weekend column says the board nearly did not act and moved only at general counsel Tim Teter's insistence; Reuters says Huang told Teter the Megaspeed dealings were "standard business" (The Information, AI Weekly).
- TechPowerUp reports allegations that Nvidia engineered scarcity by funding startups that then bought its GPUs (amplified by Gary Marcus). Allegation only.
- Goldman sizes AI infrastructure tied to Nvidia, Broadcom and AMD systems at about $3.8T through 2028: Nvidia 63% of dollars but 47% of gigawatts, Broadcom 21% and 31%.
- Amodei calls compute buying Anthropic's "deepest business dilemma", with commitments near $901B before the IPO (The Decoder).
Relation to prior pages
Follows Nvidia without Jensen (11-13) (SemiAnalysis: $530B guarantee book personal to Huang) and AI financing stress (11-14). Lambert's essay is the first piece in the wiki to price the open-model consequence rather than the financing one.
Source: Interconnects (raw/gmail/2026-11-15-newsletters.md, raw/rss/2026-11-14-interconnects-ai-open-source-ai-without-jensen.md).