$1.5 trillion. That’s what Jensen Huang expects AI spending to hit in 2026, according to a recap from Parkev Tatevosian, CFA. Hold that next to Nvidia’s $33 billion in trailing 12-month revenue and you get a ratio that should make anyone pause. The company selling the shovels is pulling in roughly 2% of the gold rush’s projected annual budget, and its CEO is on television telling everyone the turmoil will pass.
I review AI tools for a living. I don’t model semiconductor demand, I don’t price options, and I have no opinion worth printing on whether Nvidia stock is cheap. What I do have is a folder full of tools I’ve tested, a running list of ones that quietly died, and a healthy suspicion of anyone who tells me the future is fine. So let me take Huang’s comments from the seat I actually sit in.
What he actually said
Three things, roughly. On CNBC’s Squawk Box, Huang said AI has gone through a new inflection point. Fortune describes him as the biggest counterweight to AI doomerism, the camp that treats AI as an existential threat to humanity, and notes he stresses that safety has to be solved for the potential to show up. And in a separate clip, he warned that what’s coming in 2026 could be transformative and also harmful, urging a careful, practical approach.
Read those together and the “all of this is going to pass” framing gets more interesting than the headline suggests. He isn’t saying nothing is happening. He’s saying the panic is temporary and the capability curve is not.
Why a tool reviewer cares about a chip CEO’s mood
Because the spending number is the water the whole ecosystem swims in. Every agent framework, every wrapper, every “AI-powered” dashboard in my review queue exists downstream of compute that somebody is paying for. When the projected spend is measured in trillions, you get what we have now: hundreds of tools shipping weekly, most of them thin, a handful genuinely useful.
That abundance is not a gift. It’s a filtering problem, and it lands on you, the person choosing what to put in your workflow. Here’s what the current conditions reliably produce:
- Tools priced below cost. Generous free tiers funded by somebody’s runway, not by unit economics. Great while it lasts. Plan for the pricing page to change.
- Feature announcements that outrun the product. If a tool’s changelog moves faster than its docs, you’re testing an idea, not using software.
- Consolidation you didn’t vote for. The tool you standardized on gets acquired, and your integration becomes someone else’s roadmap item.
- Capability jumps that invalidate your setup. Huang’s 2026 warning cuts both ways. If model capability takes another step, a lot of scaffolding built to work around today’s limits becomes dead weight.
None of that is a reason to sit out. It’s a reason to pick differently.
How I’d buy in a market that’s told to stay calm
My honest take: treat every AI tool as a rental, not a purchase. That single mental shift solves most of the problems above.
Favor tools that let your data out. Export matters more than features. If I can’t get my prompts, evals, logs, and configs out in a plain format, I don’t care how good the interface is. Lock-in is the real cost, and it’s the one that doesn’t show up on the pricing page.
Prefer thin integration layers over deep ones. The more of your process lives inside one vendor’s product, the more a capability jump or an acquisition hurts. I’d rather wire three decent tools together with code I control than hand the whole workflow to one platform that’s eighteen months old.
Judge on what works today, not the roadmap. I’ve stopped giving credit for announced features. A tool either does the job in my test this week or it doesn’t. The 2026 pitch is not a product.
And take the safety point seriously, because Huang does. His argument isn’t that risk is fake, it’s that risk is solvable and worth solving. For a tool buyer, that translates into unglamorous questions: what does this thing do with my data, what permissions does the agent actually hold, can I see what it did after the fact. Tools that can’t answer those don’t belong near anything that matters.
The useful part of the message
I think Huang is probably right that the market noise passes and wrong to expect anyone downstream to feel reassured by that. A trillion-dollar spending forecast and a warning about potentially harmful capability arriving the same year are not a contradiction. They’re a description of the conditions.
For those of us picking tools, the practical read is simple. Stay flexible, own your data, buy for this quarter, and keep the exit cheap. The turbulence may well pass. Your vendor might not.
🕒 Published: