Jensen Huang Claims AGI Has Been Achieved

On March 23, 2026, during a two-and-a-half-hour conversation on the Lex Fridman podcast, NVIDIA CEO Jensen Huang made a statement that is bound to stir discussions in the tech community. When Fridman asked how far we are from an AI system that can create, develop, and operate a tech company worth over a billion dollars, Huang replied succinctly, “I think it’s now. I think we’ve achieved AGI.”

This assertion carries significant weight coming from the head of the world’s most valuable company, NVIDIA, which is valued at approximately $4.3 trillion as of late March 2026. Its GPUs power nearly all mainstream large models, from OpenAI’s ChatGPT to Google Gemini. If anyone has a direct understanding of AI systems’ real capabilities, it is Huang.

However, Huang quickly narrowed his statement. He seized on a phrasing gap in Fridman’s definition: “You said a billion dollars, but you didn’t say ‘forever.’” Following this line of thought, he explained that an AI could potentially create a simple web application that suddenly goes viral, generating a brief revenue spike of over a billion dollars before quickly failing. He noted that many such websites emerged during the internet bubble, where their technical complexity was not greater than what today’s AI could produce, and many users would lose interest within months. He concluded that the probability of 100,000 such agents building a company like NVIDIA is zero.

The word “zero” shifted his earlier claim of “AGI has been achieved” to a completely different context. According to Huang, AI can now luck into a temporary commercial success, but it is far from being able to sustain a complex enterprise, manage supply chains, handle compliance audits, or lead thousands of people. He effectively raised the bar for AGI to the level of a one-time commercial success, claiming we have reached that, while admitting that the kind of AGI everyone truly expects is still far off.

This reflects a long-standing dilemma in the AGI discussion: both parties often use the same terminology but point to completely different expectations. One interpretation sees AGI as a threshold that can be reached by selecting appropriate metrics, while another views it as a cross-domain, sustainable capability that can operate autonomously in an open environment.

By the first standard, large language models can already pass bar exams, write complex code, and match or exceed human performance on various academic benchmarks. By the second standard, the most advanced AI systems still produce hallucinations, make basic errors in multi-step logical reasoning, and lack true contextual understanding, indicating that we are still far from achieving “general” intelligence.

Looking back at Huang’s previous public statements, his timeline for AGI has also evolved. At the end of 2023, during the New York Times DealBook Summit, he stated that if AGI is defined as software that performs competitively in various human tests, it could be achieved in about five years. In March 2024, at the GTC conference, he reiterated the five-year timeline, provided that AI performs better than most people by over 8% in legal exams and logical tests. He specifically mentioned that he was reluctant to make predictions without a clear definition of AGI.

Now, he has dropped the five-year timeline altogether, asserting that it is “now,” but has also narrowed the definition from “outperforming in all human tests” to “creating a briefly popular application.”

In contrast, many tech executives have been deliberately avoiding the term AGI in recent months. Sam Altman, in a funding statement at the beginning of 2026, downgraded AGI to a “milestone on the road.” Demis Hassabis has repeatedly stated that AGI is a “vague term.” Many are coining alternatives like “advanced AI systems” or “cutting-edge intelligence” to retain the grand narrative without being tied to a concept lacking consensus.

Huang, however, is doing the opposite by bringing AGI back to the forefront.

In Silicon Valley, AGI has always been more than just an academic concept or PR jargon; it relates to real financial contract terms. The new agreement signed between OpenAI and Microsoft in October 2025 stipulates that Microsoft’s exclusive IP rights to OpenAI’s models and products last until “AGI is declared achieved.” Once AGI is recognized, revenue sharing, API exclusivity, and other terms will need to be recalculated.

To prevent unilateral actions, the agreement includes an independent expert committee for verification. Media reports citing leaked documents indicate that earlier versions of the agreement defined the trigger for AGI as “OpenAI developing an AI system that generates at least $10 billion in profit.” Thus, in contractual terms, AGI is a financial metric.

Therefore, when the CEO of the world’s largest AI chip supplier publicly states, “AGI has been achieved,” it stirs not only public opinion but also an entire set of business relationships.

The podcast contained more than just AGI discussions. Huang spent considerable time discussing how NVIDIA is transitioning from single-chip design to a focus on “rack-level” and even “data center-level” systems engineering. He emphasized the need for GPU, CPU, high-bandwidth memory (HBM), networking, optical interconnects, power supply, cooling, and software stacks to be optimized as a cohesive whole, a concept NVIDIA calls “extreme co-design.” He mentioned having over 60 direct reports, each an expert in their respective technical fields, and this ultra-flat structure is designed to facilitate this interdisciplinary engineering.

Huang also outlined his four laws of AI expansion: pre-training, post-training, expansion during testing, and agent expansion. The essence is that the industry has previously worried about running out of high-quality data for pre-training, but the subsequent three dimensions have taken over, leading to increased computational demands during inference, and agents will continuously spawn sub-agents to run sub-tasks, further amplifying computational needs. All paths point to the same conclusion: the growth of intelligence depends on computational power.

This conclusion is a perfect business narrative for NVIDIA. Huang stated that by the end of 2027, NVIDIA will generate at least $1 trillion in revenue from the sales of Blackwell and Vera Rubin chips, “and supply will definitely be short.” In this context, declaring that AGI has arrived, while defining it as an activity that requires massive computational support, has a clear purpose.

An interesting segment of the podcast was Huang’s discussion on the distinction between intelligence and humanity. He noted that all 60 of his direct reports are stronger than him in their respective fields and have received better education. “To me, they are all superhumans.” Yet, he, who washed dishes at Denny’s in his early years, coordinates the entire organization among these superhumans. “You have to ask yourself, how can a dishwasher sit among superhumans?”

He aimed to convey that intelligence and humanity are two different terms. Empathy, willpower, generosity, and the ability to endure pain do not fall within the realm of “intelligence.” “I believe these are the superpowers, while intelligence is about to be commoditized.”

He used radiologists as an example. AI researchers initially predicted that radiologists would be the first profession to disappear, as computer vision had already surpassed humans in reading images around 2019 to 2020. However, now every radiology platform uses AI, and the number of radiologists has actually increased, with a global shortage still ongoing. As AI has sped up image reading, hospitals can accommodate more patients, thereby increasing the demand for radiologists. Huang believes the same logic applies to software engineers: “Your professional purpose is related to your job tasks, but they are not the same thing.”

This statement naturally sparked a significant reaction on social media. Hacker News saw hundreds of replies within hours, with many noting that NVIDIA’s Q1 2026 financial report is about to be released, GTC 2026 has just concluded, and the expectation of “at least $1 trillion” in chip revenue has been announced. Declaring that “AGI has been achieved” at this juncture seems unlikely to be coincidental.

However, Huang likely does not care how others interpret it. Having been a CEO for 34 years, he is the longest-serving leader of a tech company, and his tolerance for controversy may be higher than most people imagine. He remarked on his management style during the podcast, which also aptly describes his AGI statement: “How hard can it be?”