Five words. That’s the entire technical specification Jensen Huang handed the industry when he told CBS News that tech companies should build AI “as fast as we can.” Five words, no roadmap, no caveats, no asterisk about which parts are ready and which parts are still duct tape.
I review AI tools for a living. I install them, I break them, I write down what happened. So when the CEO of the world’s largest chipmaker steps into a market panic and essentially says the turbulence will sort itself out, my first instinct isn’t to argue with him. My first instinct is to ask what that means for the person reading a pricing page at 11pm, trying to decide whether to commit a team to a platform.
The confidence is real, and it’s also a sales position
Huang has been the loudest counterweight to AI doomerism, as Fortune framed it, and that’s a genuinely useful role in a conversation that swings between apocalypse and utopia with nothing in the middle. He’s not hedging. At GTC 2026 he put it plainly: “AI is no longer a single breakthrough or application — it is essential infrastructure. Every company will use it. Every nation will build it.”
Read that again as a tool buyer rather than an investor. “Every company will use it” is a statement about aggregate demand over an unspecified number of years. It is not a statement that the specific agent framework you’re evaluating this quarter will still have a maintained repo next spring. Both things can be true at once. Electricity became essential infrastructure and most of the companies selling generators in 1905 still went under.
That’s not cynicism about Huang. He sells the shovels. The shovel business genuinely does fine whether or not any particular prospector strikes gold, and he’s never pretended otherwise. His comments at CES 2026 and GTC 2026 both pushed the same line: the direction is settled, the pace should increase, companies like Nvidia are the ones driving it. Directionally, I think he’s right. Practically, “this will pass” is weather commentary, and you still have to decide whether to carry an umbrella today.
What I actually do with a statement like this
Nothing, mostly. And I’d suggest the same to you. Macro confidence from a chip executive shouldn’t change a single line in your evaluation checklist. Here’s what does change it:
- Does the tool work on your real data? Not the demo dataset. Yours, with the weird column names and the three years of inconsistent formatting.
- What’s the exit cost? If you need to leave in eight months, how much of your workflow is portable and how much is a rewrite?
- Who maintains it? A funded team, a single maintainer, or a very enthusiastic Discord server.
- Does the pricing survive scale? A lot of agent tooling is cheap at ten calls a day and unaffordable at ten thousand.
None of those questions get easier or harder based on whether the market is calm. They’re the same questions in a boom and in a correction. The only thing a downturn changes is how many vendors are around to answer them.
The open-source signal is more interesting than the speech
At CES 2026, Nvidia introduced the Alpamayo family of open-source models and tools. I pay more attention to that than to any interview soundbite, because open weights and open tooling are a structural bet, not a talking point. When a company that profits from compute also gives away models, it’s telling you where it thinks the durable value sits: in the layer underneath you, not in the layer you’re building.
For reviewers and builders, that’s genuinely good news. Open releases mean you can test without a sales call, measure without a trial window, and keep a working copy if the vendor relationship goes sideways. That’s a more solid foundation for a tool stack than any executive’s reassurance about market cycles.
My honest read
Huang is probably right that the current anxiety fades. He’s been right about the arc of this thing longer than most people have been paying attention to it. He also has roughly the strongest possible incentive to say exactly what he said, and both of those facts should sit in your head at the same time without canceling each other out.
So push forward, sure. Build fast. Just build fast on things you’ve tested yourself, with an exit you’ve priced out, using whatever you can run locally when the vendor stops answering email. Optimism about the category is free. Confidence in a specific tool has to be earned the boring way, one install at a time.
đź•’ Published: