Jensen Huang: NVIDIA has achieved AGI!

AI Era Report 

During the recent earnings call, Jensen Huang declared: NVIDIA has achieved AGI!

Soon after, he made a stunning prediction—

In the future, NVIDIA may need only 40,000 human employees while simultaneously commanding 400,000—or even 4 million—tireless AI Agents working around the clock!

If NVIDIA is the first to achieve AGI, what would that mean?

NVIDIA already owns the world’s largest computational power base. With AGI realized, an unprecedented “compute-powered R&D” moat would be fully established.

Once AGI unlocks Recursive Self-Improvement (RSI), NVIDIA will be internally incubating a super army of millions of AI Agent scientists.

This army will self-sustain, automating the design of the next-gen, even more extraordinary GPU architectures, and could even rewrite CUDA code itself.

This “standing on its own shoulders” dimensional leap will bring NVIDIA unprecedented monopoly profits. Because the supply chain of the physical world is tightly controlled, most competitors will be left out in the cold.

In the end, NVIDIA will not only monopolize the majority of the world’s available “AI labor force”, but also reap the richest monopoly gains as humanity enters the silicon-based civilization era!

In the tech circle, AGI means AI performs as well as or better than humans in most economically valuable tasks.

In recent years, tech giants have fiercely competed to “touch the threshold of AGI” first.

However, in Wednesday’s earnings call, when asked about the feverish pursuit of AGI by companies like OpenAI, Jensen Huang calmly said,

This isn’t the first time he’s said this. Back on Lex Fridman’s podcast in March, he expressed similar views bluntly.

This time, not only did he declare reaching this key milestone, but he also took it a step further.

Jensen Huang pointed out unreservedly,continuing to debate “what is AGI” or “what standard counts as AGI” is now completely meaningless.

Why?

Because the tech world has no universal consensus on “intelligence” itself, let alone on how to measure AGI.

It’s like everyone is sprinting towards a finish line, but nobody knows what it looks like. In that case, arguing about abstract definitions is just wasting time.

What truly excites Jensen Huang—and what truly warrants global caution—is the fundamental qualitative leap in AI’s capabilities.

According to the latest data, NVIDIA’s Avo architecture has already achieved a perfect 100% in the long-horizon autonomous agent benchmark ARC-AGI-3!

This perfect score carries weight. The ARC test, proposed by François Chollet (the father of Keras), is recognized as the “toughest IQ test” in AI.

It targets large models’ Achilles’ heel: “pattern memorization”. It challenges AI withnever-before-seen, with no historical data for reference logical puzzles, requiring the AI to use only a handful of examples and exhibit human-like abstract reasoning ability.

In the past, even OpenAI and Google's strongest models repeatedly stumbled here—cracking 50% accuracy was already extremely difficult.

However, NVIDIA’s Avo architecture scored a perfect 100%, and did so under pure zero-shot prompting—for all 183 levels across 25 public environments, without explicit rules, clearing them all through independent reasoning!

This not only makes NVIDIA the first company in the world to achieve a perfect score on this test but also marks that AI has finally jumped out of the “pattern matching” dead end, acquiring the general cognitive ability to solve unknown and complex problems.

AI is saying farewell to the era of “you ask, AI answers” passivity.

Today’s AI is anautonomous agent with a cutting-edge universal architecture.

Upon receiving a task, it can independently break it down, execute steps autonomously, and even reflect and learn new skills after completion, realizing recursion-based self-improvement.

It not only works but also summarizes experience, getting smarter and smarter as it goes. This is Jensen Huang’s vision of true “AGI deployed”.

If benchmark results are not convincing enough, how credible is NVIDIA’s real-world “AGI chip-making” story?

At Computex and GTC Taipei 2026, Jensen Huang and Cadence jointly launched the ChipStack AI Super Agent—officially claimed to have reached Level-5 autonomous capability.

This system orchestrates workflows with Codex and Nemotron as engines, calls Cadence Xcelium for RTL simulation and Jasper for formal verification—all tasks run in the NVIDIA OpenShell sandbox.

The official data is exciting: typical validation closure cycles shrink from about five weeks to less than a day, and RTL verification cycles are accelerated by over 40x.

The massive internal validation system at NVIDIA—thousands of engineers, tens of billions of compute-hours annually, millions of test cases—will all be carried by this agent system.

If “achieving AGI” is a technological shock, then NVIDIA’s financials are a capital market shock—blowing past expectations.

Key figures this quarter:

• Revenue: $96.2 billion (expected $92.2 billion, +106% YoY)

• Data center: $89 billion (expected $85.8 billion, +117% YoY), driven by hyperscale customers ($48.7 billion) and enterprise AI ($40.3 billion)

• Net profit: $59.7 billion (incredible 62% profit margin, up 6% YoY)

• Gross margin: 75% (expected 75%, up 250bps or 2.5% YoY), gross profit reached $72.1 billion

• Adjusted EPS: $2.22 (expected $2.10, +120% YoY)

Based on 91 days, that’s about $1.06 billion in revenue every single day—including weekends.

The Q3 guidance is head-turning: projected revenue at $108 billion!

This will be NVIDIA’s first quarter ever breaking the $100 billion barrier, with average daily income surging to nearly $1.2 billion!

A year ago, cutting-edge AI labs still relied on burning venture capital; today—as per SemiAnalysis founder Dylan Patel’s calculations—AI has shifted from a cash-burning black hole to a terrifying money-printing machine—

Every megawatt of base compute power costs $10–15 million, yet can be resold for $50 million or even over $100 million in end revenue.

When NVIDIA and its major clients can reliably “turn $10 into $100,” they possess absolute “unlimited firepower” in the market.

Following in the steps of Amazon, Apple, and Alphabet, NVIDIA will join the “single-quarter $100 billion revenue” club

But the real bombshell from the call came from CFO Colette Kress. She gave full-year FY2028 guidance a year in advance:Projecting revenue growth around 70%—unprecedented in NVIDIA’s history.

Wall Street had previously expected only 44%.

What does this mean? Based on the current near-$400 billion consensus for this fiscal year, FY2028 total revenue could approach a jaw-dropping$673 billion.

This would put NVIDIA ahead of Apple and Microsoft, making it second only to Amazon in US technology revenue.

Jensen even added a hint of unrestrained pride:

With a roughly 65% net margin, NVIDIA’s net profit in FY2028 could approach $450 billion.

Its annual profit might exceed the GDP of most countries. In short, even Wall Street’s imagination can hardly keep up with Jensen Huang’s pace of raking in cash.

Since the definition of AGI is now meaningless, what is actually the core transformation in the tech world?

Jensen Huang offered a bloody-capitalist answer:

What does it mean to “generate profit-earning Tokens”?

In the world of large AI models, a Token is the basic unit of text processing.

Previously, these Tokens were just lab data. Now, as AI Agents deeply integrate with actual business, Tokens become palpable commercial value.

When AI automatically generates flawless back-end code for you, the efficiency boost it brings is a profit Token;

When AI instantly gathers global market data, crafts investment strategies, and executes trades in seconds—real money earned is the ultimate profit Token!

In the past, AI was a money-burning black hole. Now, things have changed.

Jensen Huang explained the logic clearly: More compute = more Tokens = inevitably more profit.

This is why the key metrics to gauge who’s winning the next phase of the AI war are now “Tokens per dollar” and “Tokens per watt.”

That’s also why Silicon Valley giants don’t care what AGI is called anymore—they’re all in, going wild. As Huang put it: “This is the stage we’re at, and it’s why everyone is doubling down.”

Whoever controls compute power owns the money printer of the new era. By this logic, AI is already reshaping production relations and is the strongest driver of the global economy.

NVIDIA's mass-produced and on-schedule Vera Rubin architecture will push revenue opportunities for every GW of data center from $1.8 billion (with Hopper) to $4 billion—these lightning-fast iterations are turning the company into the world’s most efficient “digital economy money printing engine.”

Last year, only a handful of top labs were scrambling for infrastructure; today, generative AI is in its golden age: new AI Lab national teams, physical robot AI, the Internet of Everything—all are battling madly for compute.

Given the powerful autonomy and self-evolution demonstrated by current AI Agents, Jensen’s next prediction is:

Consider what this means: A human-to-digital employee ratio of 1:10, or even 1:100!

As Dylan Patel deduced in ledger calculations: a ten-trillion-level compute expansion will trigger ecosystem-wide credit demand exceeding $5 trillion.

In the next two or three years, all real-world economic operations—including rising market rates and the collapse in value of traditional stocks due to soaring discount rates—will be a direct result of this AI compute monopoly and expansion.

The August 2026 earnings call will go down in history.

Jensen Huang once again gave AI a new lease of life, but more importantly, laid bare the ultimate card in AI commercialization: compute is power, Token is wealth!

In this era of “40,000 humans commanding 4 million AIs”, panic is pointless.

The future rule of survival is simple: Either become the one who controls AI, or become the one replaced by AI. 

Editor: David