Jensen Huang declares AGI has arrived

Silicon Valley loves a good hype cycle, but it hits differently when the man selling all the shovels decides to declare the gold rush over.

Nvidia CEO Jensen Huang took to X to congratulate OpenAI on launching its new powerhouse model, Astra. Four years after ChatGPT, Huang wrote, "AGI has arrived." And just in case anyone forgot who made it possible, he pointed out that Astra was trained on Nvidia silicon and teased that another 400,000 GPUs are rolling out next.

OpenAI leadership went all in too. President Greg Brockman told reporters, "Welcome to the AGI era," adding that looking back, history might mark this exact model as the tipping point.

What Astra actually does

OpenAI framed Astra as its most intelligent and aligned model to date, designed to take on heavy professional workloads with real judgment.

Leaked demos and early benchmark runs show why people are paying attention. Astra leans hard into direct computer use: it can grab your cursor, navigate spreadsheets, fill out complex tax forms in a browser, and even turn a schematic into a printable circuit board design.

On the demanding ARC-AGI 3 benchmark, which drops agents into unfamiliar rule-based puzzles where most frontier models struggle below 10%, Astra reportedly scored over 60% under maximum reasoning effort. It also aced the Exploit Bench cybersecurity test with a perfect 100% score.

Still, early real-world leaderboards tell a more nuanced story. On LiveBench, Astra landed in third place, and on the Artificial Analysis index, it trailed Claude Fable 5. It is undeniably a beast of a model, but hardly an omnipotent mind.

Victory by corporate fiat

Unsurprisingly, researchers pushed back against the victory lap.

AI critic Gary Marcus posted a blistering Substack note on Sunday evening, arguing that Huang jumped the gun without offering a shred of evidence or an operational definition. To Marcus, declaring AGI via social media felt like an attempt to hijack a serious scientific question by corporate decree. He noted that across his own 10-point AGI rubric, Astra satisfies only one or two criteria.

Even the creators of the ARC Prize declined to join the celebration. While calling Astra's scores a major leap forward, they emphasized that passing tightly bounded benchmark puzzles does not prove general intelligence in the messy, open-ended real world.

Even Sam Altman sounded skeptical of the label recently, calling AGI a poorly defined concept that resembles an irrelevant marketing term.

Yet for Nvidia, that marketing term pays handsomely. OpenAI publicly calls Nvidia the foundation of its infrastructure. Just last month, Nvidia posted $96.2 billion in quarterly revenue, powered by an $89 billion data center haul.

Why it matters

For builders and creators, Astra is a genuine leap in practical utility. Its ability to navigate desktop interfaces, build 3D assets in Blender, and solve complex terminal tasks shows that software agents are finally escaping the chatbot text box. It is worth testing the moment you get access.

For the industry, Huang's declaration is textbook narrative control. By pronouncing AGI arrived on current hardware and immediately teasing 400,000 new GPUs, he sets up Nvidia as the inevitable tollbooth for the post-AGI world. It forces every competing chipmaker and frontier lab to stay on the treadmill.

We think the move is brilliant marketing, but shaky science. When tech giants unilaterally declare victory without clear standards, they risk turning what should be a rigorous scientific pursuit into a PR exercise. Astra pushes the frontier forward, but redefining AGI on a whim ultimately muddies the water for everyone.