Beyond Nvidia: the open-weight, agent-safety and financing causes that just lost their organizer

Huang’s resignation is usually framed as a governance story and a national security story. TechCrunch argues the bigger loss is influence. He rallied hundreds of companies behind open-weight models, pulled more than 100 firms into an open agent safety effort and persuaded private equity giants to put up to $500 billion behind data center build-outs. Those causes don’t disappear when he leaves, but they lose their most effective champion just as each is under pressure, from proposed bans on Chinese open-weight models to a bond market wary of AI infrastructure debt.

He resigned after admitting an affair with Alice Huang, an executive at Singapore-based neocloud Megaspeed whom the U.S. government suspects of spying for China, and after reports of a second affair with a direct report (timeline). Industries are shaped by people who can pick up the phone and get a few hundred companies to sign the same letter.

Open models give enterprises real choice and push costs down. Nvidia has its own open models, a pending $12.9 billion deal for Hugging Face (Delangue’s response), and in July it organized industry support for open weights while frontier labs were reportedly urging restrictions on Chinese models. Without a heavyweight voice, the policy fight tilts toward companies that benefit from closing things up (analysis).

Arguing that the industry needs no regulation and will fix agent safety itself was always a bet. The Open Agent Safety Platform, OpenShell plus BlueField-4 Sentry (launch), was a reasonable start, and in Berlin he pitched agent security as an engineering problem (keynote). But self-policing depends on trust, and trust is what this scandal burned. If the chief spokesperson for self-regulation is out amid a security probe, lawmakers have an easier case for outside rules, which may not be the worst outcome.

This is where the risk compounds. TechCrunch describes a web of so-called circular deals, a $105 billion financing backstop tied to an OpenAI data center in Ohio, and a pitch that GPUs can serve as durable collateral because Nvidia stands behind their value. Arrangements like that lean on confidence in one company and one salesman. Remove him while roughly half of this year’s data center bonds are rated junk, and the structure looks more fragile. The market has already erased more than $1 trillion of Nvidia’s value (tally).

A new line of Nvidia-powered AI PCs due this holiday season loses its most visible promoter at launch. In Washington, Republican committee chairs have sought closed-door testimony (Congress), and Democrats take over key committees in January.

The AI build-out carried more key-person risk than most admitted. Whoever runs Nvidia next must sell the same vision without the same credibility, and the industry should build coalitions that don’t depend on one founder. More in our Nvidia coverage.