Jensen Huang wants no brakes.
The Nvidia CEO said in 2026 that AI should be built “as fast as we can,” rejecting calls to slow things down. He made a version of that argument at Dreamforce alongside Salesforce CEO Marc Benioff, again in a CBS News interview, and again at Davos in conversation with BlackRock’s Larry Fink. His position is consistent: AI is essential infrastructure, market forces are enough to keep it safe, and new regulations aren’t needed.
If you’re someone who uses AI agents to draft emails, sort your inbox, or handle customer questions, this might feel like a debate happening several floors above you. It isn’t. The speed at which these systems get built is the speed at which they show up in your tools, whether or not you asked for them.
Why the guy with the chips says go faster
Nvidia makes the hardware nearly every serious AI system runs on. Faster AI development means more demand for that hardware. That’s not a gotcha, it’s just useful context. Huang isn’t a neutral observer, and he doesn’t pretend to be one.
His actual argument is worth taking seriously on its own terms though. Calling AI “essential infrastructure” puts it in the same mental category as electricity, roads, or broadband. We didn’t pause the electrical grid until we’d solved every safety question. We built it, and safety standards developed alongside it, often in response to things going wrong.
The market-forces piece is the part where reasonable people start disagreeing. The claim is that companies shipping unsafe AI will lose customers, get sued, and get outcompeted by companies shipping safer AI. Safety becomes a feature you compete on, not a rule you’re forced to follow.
What that means for people using agents
Here’s the practical translation. If safety is a market feature rather than a regulatory floor, then you, the person picking which AI agent to use, become part of the safety mechanism. Your choices are the enforcement.
That’s a lot of responsibility to hand to someone who just wants their meeting notes summarized. But it’s the logic of the position, so it’s worth understanding what it asks of you:
- Read what an agent can actually touch. An agent with access to your email and calendar is a different risk than a chatbot in a browser tab. Permissions matter more than the marketing copy.
- Notice whether a company publishes anything about its safety testing. Some do. Some don’t. That difference is information.
- Watch how companies behave when something goes wrong. The response to a failure tells you more than any launch announcement.
- Keep a human in the loop on anything that spends money, sends messages to real people, or touches sensitive data. This one is boring and it works.
Not everyone in the industry agrees
Huang’s stance breaks with some other leaders in the field, including Anthropic’s Dario Amodei, whose position he was explicitly contrasted with at Dreamforce. That split is genuinely useful for the rest of us, because it means there’s no single “AI industry view” being presented as settled fact.
When the people building a technology publicly disagree about how fast to build it, that’s a signal you’re allowed to have an opinion too. You don’t need a machine learning PhD to notice that “move fast and let the market sort it out” and “slow down and prove it’s safe first” are different bets with different failure modes.
The move-fast bet fails if something breaks badly before anyone catches it. The slow-down bet fails if the caution is one-sided, and whoever isn’t being careful ends up ahead anyway. Neither side is being unreasonable. They’re weighing different risks.
Where I land on it
I find the infrastructure framing convincing and the market-forces framing shaky. Those two things can both be true.
AI probably is becoming infrastructure. It’s already threaded into software most people use daily, usually without a label. Trying to halt that is like trying to halt the internet in 2003.
But infrastructure is exactly the category where we normally don’t rely on market pressure alone. Building codes exist because buildings falling down is bad, and customers can’t inspect a foundation. Most AI agent users can’t inspect what’s under the hood either. They see the interface and trust it.
So the honest version of the market-forces argument requires something Huang doesn’t really address: for consumers to push companies toward safety, consumers need to be able to tell the difference. Right now, mostly, we can’t.
That’s the gap. And until it closes, “as fast as we can” is going to mean different things depending on whether you’re selling the chips or using the results.
🕒 Published: