Jensen Huang Declares 'We Have Achieved AGI,' Sparking Fierce Industry Debate

NVIDIA CEO Jensen Huang declared that artificial general intelligence has already been achieved during a wide-ranging interview on the Lex Fridman Podcast, episode #494, published on March 23, 2026. The bold claim immediately ignited a fierce debate across the AI research community, with experts questioning both the assertion and the very definition of AGI.

What Huang Said

When Fridman asked how long it would take AI to innovate, find customers, and manage a team to build a billion-dollar company, Huang responded plainly: "I think it's now. I think we've achieved AGI."

He argued that it is "not out of the question" that a model like Claude could create a web service used by billions of people, generating massive value — even if that success proved short-lived. For Huang, the ability of current AI systems to perform at or above human level across a broad range of tasks constitutes AGI.

Researchers Push Back

The claim landed just days after multiple research teams published rigorous frameworks for measuring AGI — frameworks that suggest current systems fall well short.

Google DeepMind released a paper proposing evaluation across 10 cognitive faculties, including perception, reasoning, memory, learning, attention, and social cognition. Their framework requires median human adult performance across all domains — a bar no current model clears.

The Hendrycks-Bengio model defines AGI as matching "the cognitive versatility and proficiency of a well-educated adult." Under this framework, GPT-5 scores only 57% compliance, far from the threshold.

OpenAI itself reportedly defined AGI internally as technology generating at least $100 billion in annual profits — a threshold the company has not approached despite $13 billion in recent revenues.

The Jagged Profile Problem

Critics point to what researchers call the "jagged cognitive profile" of today's leading models. While frontier AI excels at math, factual recall, and code generation, it consistently struggles with experience-based learning, long-term memory, social understanding, and the kind of flexible abstract reasoning tested by benchmarks like ARC-AGI.

These visual puzzles require domain-agnostic learning and creative problem-solving — precisely the capabilities that differentiate human cognition from pattern matching at scale.

Why Definitions Matter

The lack of a universally accepted AGI definition enables companies to claim progress strategically. For NVIDIA, whose $4 trillion valuation depends heavily on the narrative that AI infrastructure spending is justified, declaring AGI "achieved" serves clear commercial interests.

For researchers focused on safety and alignment, premature AGI declarations risk inflating expectations, diverting attention from genuine limitations, and potentially reducing the urgency around safety research for truly transformative systems.

The Broader Context

Huang's interview covered NVIDIA's extreme co-design engineering, AI scaling laws, data center challenges, and even the possibility of AI data centers in space. The AGI segment, beginning around the two-hour mark, became the most widely discussed portion of a three-hour conversation.

The debate arrives at a pivotal moment. AI companies have raised record funding — OpenAI's $110 billion round and Anthropic's $30 billion Series G both closed in recent months — while questions about sustainable business models and genuine capability improvements grow louder.

What Comes Next

Whether or not current AI qualifies as AGI depends entirely on whose definition you accept. What is clear is that the gap between corporate claims and scientific consensus on AI capabilities continues to widen — and the stakes of that gap, measured in trillions of dollars of infrastructure investment, have never been higher.

Source: Fortune