Agi Huang Company
AGI: Already Here, Meaningless, or Arriving by December?
Jensen Huang declared this week that artificial general intelligence has already arrived — and in the same breath called the milestone 'senseless.' Sam Altman said OpenAI expects to reach AGI by the end of 2026. Those two statements are less contradictory than they appear, and the gap between them exposes how much definitional work the acronym is doing.
The classic formulation of AGI — a system capable of performing any intellectual task a human can — has no agreed technical benchmark. OpenAI's own internal definition has shifted, and in Altman's current framing the threshold has moved toward something more like: a system that can autonomously generate economic value equivalent to a highly productive knowledge worker. By that narrower definition, current systems are arguably crossing it. Huang's dismissal of the label as 'senseless' is not a denial of the capability but a rejection of the category — once a system outperforms humans across domain after domain, the debate over whether it qualifies as 'general' becomes philosophical rather than technical. From a chip-maker's perspective, what matters is compute demand, and compute demand is rising regardless of nomenclature.
Nvidia's decision to ship its first H200 processors to China — apparently a small, one-time batch — while simultaneously stating it expects no future high-end chip sales into that market is an unusual preemptive disclaimer for a public company. It suggests Nvidia is either genuinely foreclosing China as a market for its advanced AI chips or signaling compliance with anticipated export control tightening before formal announcement. The Tencent angle directly confirms the pressure those controls are creating: a senior Tencent executive publicly acknowledged that the company's computing capacity is 'severely insufficient,' even as the company's first-half capital spending in 2026 already exceeded all of its 2025 spending combined. Hearing a major Chinese tech company confirm, in those words, that chip restrictions are biting is described as genuinely notable.
SoftBank is separately in talks to acquire a majority stake in 1X Technologies, the Norwegian humanoid robotics company, continuing Masayoshi Son's consistent three-year conviction about workplace-ready robots. A majority stake would position 1X to access SoftBank's portfolio companies as potential customers — a significant distribution advantage. Meanwhile, a Wake Forest study found that 89.6 percent of leaked AI agent credentials — the authentication tokens that allow AI systems to take autonomous actions on behalf of users — are immediately exploitable by bad actors. As AI agents move from answering questions to executing code, managing files, and interacting with financial systems, the researchers argue that a leaked agent credential represents a qualitatively different threat than a leaked website password, with potentially persistent access to everything the agent was authorized to do.