Nvidia’s CEO Says Artificial Intelligence Has Achieved Human-Level Abilities as Booming Chip Sales Drive Up Company Earnings

Nvidia’s CEO has declared artificial general intelligence already achieved, citing OpenAI’s GPT-6 Astra, even as leading researchers dispute the claim and contractual safeguards against premature announcements have been eliminated. The disagreement highlights both the absence of universal AGI benchmarks and potential conflicts of interest tied to major financial commitments in AI development.

• Jensen Huang stated current AI systems have “practically” achieved AGI, hardening his earlier two-year timeline prediction significantly.
• OpenAI and Microsoft removed their contractual AGI trigger in April 2025, allowing Microsoft payments to continue through 2030 regardless of technological progress.
• Anthropic CEO Dario Amodei and cognitive scientist Gary Marcus argue current systems fall far short of true human-level capability across diverse domains.

• April 2025 Month when OpenAI and Microsoft eliminated AGI trigger from contractual terms entirely
• October 2024 When companies first revised agreement to require independent expert verification of AGI claims
• 1997 Year physicist Mark Gubrud first coined AGI term, now claiming milestone has arrived

Nvidia Chief Executive Jensen Huang has declared that artificial general intelligence has been achieved, identifying OpenAI’s GPT-6 Astra as the system that crossed the threshold into human-level capability. His statement represents a hardening of measured remarks made at a G20 meeting in North Carolina, where he had expressed optimism about AGI arriving within two years. Huang’s declaration carries particular weight within the technology industry given Nvidia’s dominant position supplying graphics processing units and AI chips to nearly every major AI development effort.

In the next couple of years, we are going to achieve essentially what people call AGI. In fact, I would argue that we’re practically there today. It either means a lot or it doesn’t mean anything.

Jensen Huang, Nvidia CEO

Industry lacks universally accepted standards for measuring artificial general intelligence

The central problem underlying Huang’s announcement is the absence of agreed-upon definitions or tests for artificial general intelligence. AGI broadly refers to AI capable of performing intellectual work at human level across diverse domains, but researchers have proposed competing frameworks ranging from simple performance benchmarks on standardized tests to more complex definitions requiring autonomous goal-setting, long-term planning, and the ability to learn entirely new domains without human intervention.

The AGI concept emerged from academic computer science in the 1980s and 1990s as researchers sought to distinguish their ambitions from narrow, task-specific AI systems. Unlike deep learning systems trained to excel at chess or image recognition, AGI theoretically would match human flexibility, transferability of knowledge, and adaptability across completely novel problems. This distinction has grown increasingly important as companies have marketed increasingly capable large language models as approaches to AGI.

Recent large language models have demonstrated genuine advances in multimodal understanding, language comprehension, mathematical reasoning, and code generation compared to earlier generations. Systems like GPT-6 Astra show marked improvements in reasoning depth, contextual awareness, and the ability to handle complex, multi-step problems. However, these systems still require substantial human input, operate within constrained domains, and lack the flexible transfer of knowledge that humans demonstrate across radically different tasks and environments.

Without agreed-upon measurement standards, AGI declarations function as semantic exercises reflecting the declarer’s preferred definition rather than objective progress.

OpenAI and Microsoft eliminated contractual safeguards against premature AGI declarations

The original 2019 agreement between OpenAI and Microsoft contained a provision that would have terminated the partnership and cut off Microsoft’s access to future models if the OpenAI board formally declared AGI achieved. That arrangement created a powerful disincentive for declaration, introducing structural conflict of interest into any AGI announcement given Microsoft’s multibillion-dollar investment in the company and ongoing deployment of OpenAI systems throughout its product portfolio.

The companies revised their agreement in October 2024 to require independent expert verification of any AGI claim, establishing a panel tasked with evaluating whether systems had achieved human-level general intelligence. This modification was intended to preserve meaningful accountability while allowing legitimate progress announcements. However, the companies completely rewrote the agreement again in April 2025 to remove the legal trigger altogether. Under the updated terms, OpenAI’s payments to Microsoft now continue through 2030 regardless of technological progress, eliminating the financial penalty previously attached to AGI declarations.

Industry observers questioned the timing of the removal, noting it preceded Huang’s AGI declaration by several months. The elimination of contractual checks has raised concerns about whether financial incentives now dominate AGI assessment discussions. Microsoft’s commitment to fund OpenAI operations continues expanding even without achievement milestones, reducing structural pressure for accuracy in capability claims. Industry insiders debate whether the removal reflected genuine technical evolution making the trigger obsolete, or strategic removal of an inconvenient verification mechanism.

The removal of the trigger raises significant questions about whether other financial incentives now influence AGI declarations, particularly given that Nvidia supplies much of the hardware infrastructure powering AI systems and benefits directly from expanded computational spending.

Leading researchers set substantially higher thresholds for human-level artificial intelligence

Anthropic CEO Dario Amodei outlined a significantly higher threshold for AGI in February, several months before Astra’s release. Amodei suggested that if current systems represented true human-level capability, they would demonstrate the performance level of “the country of geniuses in a data center,” which he argued clearly has not arrived. Cognitive scientist Gary Marcus has echoed similar concerns about premature declarations, arguing that current systems still fail basic tests of general reasoning and adaptability that humans perform routinely.

Large language models, while impressive, remain brittle in ways human intelligence is not, failing catastrophically on tasks requiring common sense reasoning or novel problem-solving outside their training distribution. Current systems also lack key features of human intelligence including self-directed goal formation, the ability to acquire entirely new skills through observation and experimentation, and flexible reasoning across domains with minimal training. A human child can observe an unfamiliar tool, understand its purpose, and use it effectively; current AI systems struggle with comparable generalization challenges.

Academic researchers point to growing divergence between computational performance metrics and genuine understanding. Systems can achieve high scores on standardized tests while remaining fundamentally incapable of the adaptive reasoning humans deploy constantly. This gap between narrow benchmark excellence and broad capability represents the central concern for skeptics evaluating AGI claims.

Physicist Mark Gubrud, who first coined the term AGI in 1997, now says it has arrived, representing a significant shift in his assessment and adding a contrasting voice to the skeptics among academic researchers.

Governance bodies and academic institutions have called for standardized AGI assessment protocols, but implementation has been slow, leaving unresolved whether current systems represent a genuine inflection point or the latest cycle of overstated claims in an industry with strong financial incentives to announce breakthroughs. The debate reflects deeper questions about how technological progress should be measured and validated when billions of dollars depend on continued investment momentum.

BlockWest is a news publication. Nothing here is investment advice. Read our disclaimer and editorial policy.