At the All‑In Summit 2026, Jensen Huang, CEO of NVIDIA, publicly pushed back against the rampant AI doomsday theory. He argued that some predictions about civilizational extinction risks posed by AI amount to sensationalism without scientific basis. Donald Trump joined the event via a live phone call and stated publicly that claims of AI going rogue and taking over the world are “a hoax altogether.”

A phone call at the summit laid bare the fierce divide between Silicon Valley’s two competing visions for AI.

At the All‑In Summit 2026, Jensen Huang, CEO of NVIDIA, publicly pushed back against the rampant AI doomsday discourse. He argued that some predictions of civilization-ending risks posed by AI amount to sensationalism lacking scientific evidence.

A dramatic interlude mid-session elevated this industry debate straight into policy territory: Donald Trump dialed into the summit live and stated publicly that claims of AI going rogue and taking over the world are “a hoax altogether.” He framed data centers as the oil of the next two to three decades and voiced full-throated support for the full-speed buildout of AI infrastructure.

On one side, safety advocates warn that Recursive Self-Improvement (RSI) could trigger an intelligence explosion, calling for slower research and global oversight. On the other, accelerationists represented by Huang and Trump insist technology remains controllable and oppose holding back industrial progress with doomsday narratives.

Silicon Valley’s debate over AI’s fate has now moved fully into the public arena.

Jensen Huang: Many AI panic predictions have already been disproven by reality

Huang targeted a widely circulated recent blog post by Dario Amodei, CEO of Anthropic. Endorsed by multiple leading AI labs, the paper even quantified the probability of AI triggering “civilization-level extinction,” sending shockwaves through Silicon Valley.

Huang did not dismiss the entire piece. He said sections covering internal safety and lab risk whistleblowing deserved serious attention, yet its apocalyptic forecasts for the future were entirely unsupported by science.

“Written by researchers, yet not scientific conclusions. This is deeply irresponsible.”

He cited a string of widely circulated AI panic forecasts from past years, all of which failed to materialize:

Predictions that radiologists would be completely replaced by AI within five years; in reality, industry demand for radiologists has risen further.

Forecasts that AI would generate 90% of all code within six months and wipe out half of entry-level jobs — none came true.

GPT‑2 and Llama3 were once declared “too dangerous for public release.” Claims that half of white-collar jobs would soon vanish also never materialized.

In his view, such panic stems mostly from humanity’s innate fear of unknown technological boundaries, rather than rigorous scientific deduction. Crucially, safety does not mean hitting the brakes on AI.

Safety and technological progress are not an either-or choice. The U.S. can lead in innovation while managing risks. Regulation should focus on tangible, existing problems instead of shackling development for hypothetical doomsday scenarios that have not yet occurred.

RSI is no rogue dark magic; it is merely controllable engineering

One core keyword at the heart of this debate is RSI, Recursive Self-Improvement.

The core concern of safety advocates: once AI gains the ability to iteratively design and optimize the next generation of AI, it will enter an unconstrained intelligence explosion loop and ultimately slip out of human control. RSI has been framed as dark magic opening Pandora’s box.

Huang offered a wholly different engineering perspective: RSI is not some mysterious hazardous technology. It is a mature collection of techniques including in-context learning, reflection mechanisms, reinforcement learning and synthetic data generation, built to continuously boost AI problem-solving capabilities.

With techniques such as LoRA (Low-Rank Adaptation), capability enhancements can be achieved without modifying the base foundation model weights. After accumulating sufficient experience, teams then iterate and update the base model. The entire workflow stays firmly under engineering governance.

“You may run RSI experiments internally at your company, but before releasing products publicly, you must conduct evaluation and testing to rule out performance degradation.”

Huang stressed that the term RSI is being weaponized to deliberately stoke fears of runaway technological spirals. As labs evolve from research institutions into engineering organizations, risks can be contained within manageable bounds by refining testing, sandboxing and monitoring protocols.

Trump’s surprise call: AI will not rule the world; data centers are the oil of a new era

While Huang unpacked industry risks and RSI technology, the venue received an unexpected call. Trump joined the venue’s loudspeaker system and addressed thousands of attendees, openly siding with the accelerationist camp.

“Robots will not rule the world, and AI will not take over humanity. The whole thing is a hoax.”

He lauded the strategic value of data centers, calling them “the oil for the next 20–25 years, with influence far exceeding the internet.” Data center construction can lift prosperity across U.S. states and local communities, and voices opposing data center buildout are eroding America’s competitiveness.

He added that AI development still requires prudence, yet that should not be used as justification to stall industrial advancement.

Huang echoed him on stage: whoever wins AI wins global competition. Narratives that hinder innovation must not prevail. Every industry and region across America should share AI’s dividends.

A telling contextual detail: opinion polls at the time showed public sentiment leaning toward greater concern over AI risks and restrictions on AI development. Speaking out publicly against doomsday narratives amid this climate, Huang said, required courage.

NVIDIA’s thesis: Superintelligence has already arrived in select domains

When asked whether humanity has reached the AGI moment, Huang answered affirmatively — yet he redefined the benchmark for “superintelligence.”

It does not need universal, all-purpose capability. Outperforming humans within narrow vertical fields constitutes narrow superintelligence. For autonomous driving, the system only needs to operate vehicles safely, not make omelets. When crash rates drop to one-tenth of human drivers, it is already superintelligent in the domain of driving. We have likewise reached superintelligent benchmarks in protein synthesis and virtual drug screening.

Guided by this assessment, NVIDIA’s strategic logic is clear: move up the stack toward end-user applications while rooting itself in foundational infrastructure.

NVIDIA’s approach is not to capture all profits, but to build foundational tools and foster a flourishing ecosystem.

Foundational technologies including cuDNN and Megatron Core have been developed and opened industry-wide to power AI frameworks and large model training. Following its acquisition of Hugging Face, NVIDIA is also doubling down on the open-source ecosystem.

Huang drew an analogy: closed-source models are like bottled water, delivering premium experience; open-source models are free water sources. Neither can exist without the other.

Over the past six months, $400 billion in AI venture capital has poured into startups, 80% of which built their businesses atop open-source models. Open source enables startups, researchers and students alike to participate in AI innovation. America’s victory in AI is not a victory for a handful of tech giants, but for society at large.

Beyond that, NVIDIA is building cutting-edge open models for vertical sectors: Alpamayo for autonomous driving and Proteina Complexa for protein synthesis. These models are built in response to genuine client demand where enterprises lack in-house R&D capacity, rather than to deliberately disrupt competitors. On the capital side, NVIDIA partnered with Blackstone and Goldman Sachs via the Cloverleaf project to finance AI compute infrastructure, building a broad network of regional cloud partners across Australia and Southeast Asia.

A clash of two narratives: panic or pragmatic risk mitigation

This summit conversation is fundamentally a clash between Silicon Valley’s two ideological camps.

The underlying anxiety of safety advocates: Recursive Self-Improvement carries risks of intelligence explosion, and humanity may lose control over advanced AI. They advocate cross-border regulation and slower research, prioritizing safety above all else.

Accelerationists argue: risks that have already materialized should be addressed through engineering, audits and testing. Unproven doomsday prophecies should not suppress a genuine industrial revolution. Risk governance does not equal frozen innovation. Third-party audits and assessments can be introduced, yet regulation should target real-world incidents rather than imagined civilization extinction.

Huang does not deny the importance of safety. He recognizes the value of lab whistleblowers and supports third-party evaluation and audit frameworks.

The point of divergence is whether dramatic doomsday rhetoric should force the whole industry to hit pause.

“Even if risk narratives held true, we ought to focus on solving problems instead of frightening ordinary people.”

AI will reshape work, yet it automates individual tasks rather than eliminating entire occupations. Meanwhile, AI infrastructure will generate massive new jobs in construction, energy, software and more.

The technological wave marches onward. The real challenge is not guarding against a nebulous AI apocalypse, but governing tangible bias, misuse and lab accidents emerging as technology lands in practice — so technological dividends can be shared as widely as possible.