The Lead
Two days into GTC, the most interesting story in AI isn't about a chip. It's about two narratives running at full speed in opposite directions.
Jensen Huang is projecting $1 trillion in orders. Hyperscalers are committing $650 billion in AI capex this year. Thirty-nine thousand people in San Jose are building the future.
And this morning, Bloomberg dropped a major feature asking whether the whole thing is a bubble about to pop, Bill Gurley went on CNBC to say an AI reset is inevitable, and the tech layoff counter ticked past 55,000 jobs in 74 days.
Nobody is wrong, exactly. And if you're running a company or building something right now, understanding how both of these things can be true simultaneously is the difference between making a smart bet and getting caught flat-footed.
The Deep Dive: The Bubble Question Deserves a Serious Answer (And I Have One)
Bill Gurley is not someone you wave off. This is the Benchmark partner who called out Uber's dysfunction before the 2017 boardroom coup. Monday, he sat on CNBC and said it plainly: "One day we're going to have an AI reset, because waves create bubbles, because interlopers come in." He called the burn rates at OpenAI and Anthropic "a scary way to run a company." He referenced Carlota Perez, the economist who mapped how technology revolutions actually play out, and dropped the line that I think is the most important thing anyone has said about AI markets all year: "Bubbles only exist when the actual wave is real."
That's the part people keep missing. Gurley isn't saying AI is fake. He's saying the technology is so genuinely transformative that it has attracted the exact kind of speculative excess that real waves always attract. When you combine $700 billion in annual hyperscaler spending, AI startups raising at 300% valuation premiums over 2023 levels, and a capex-to-revenue ratio that now exceeds the dot-com era, you get the conditions for a correction. Not a collapse, but a repricing.
Bloomberg's piece this morning arrives at the same conclusion from a different angle. Three years in, the gap between AI spending and AI revenue is still enormous. Companies are financing data centers through commercial mortgage-backed securities and structured debt instruments, which, if you're old enough to remember 2007, should at least make you raise an eyebrow. The circular funding between hyperscalers, chip companies, and AI labs has Wall Street asking questions that don't have comfortable answers yet.
Meanwhile, the layoff wave isn't slowing down. It's accelerating. Meta is reportedly weighing cuts of up to 20% of its workforce, roughly 15,000 people, to offset AI spending that's ballooned to $115-135 billion this year. Atlassian just cut 1,600 people, five months after CEO Mike Cannon-Brookes said publicly that the company would hire more engineers, not fewer. Block axed 4,000 in February. Amazon eliminated 16,000 in January. Over 55,000 tech jobs gone in 74 days, with AI as the headline justification.
Here's where I want to be direct, because this is a topic where I think most coverage is getting it wrong in both directions.
The layoff story is not what it appears to be. OpenAI's Sam Altman called the trend "AI-washing" in February, noting that fewer than 1% of layoffs are genuinely caused by AI replacing workers. Gurley said the same thing: CEOs are blaming AI rather than owning up to being bloated. Look at Atlassian. They're posting 25%+ cloud revenue growth, 40%+ growth in remaining performance obligations, and 5 million active users on their Rovo AI assistant. Those aren't crisis numbers. The stock went up after they announced the cuts. That tells you everything you need to know about who these restructurings are actually for.
But the people saying "AI has nothing to do with it" are also wrong. AI is changing the skills mix companies need. It is making certain roles less necessary while creating demand for others. And the speed of that shift is faster than the HR playbooks can handle. What I object to is the dishonesty of the framing. If you're cutting 900 R&D roles to improve your financial profile for investors, say that. Don't wrap it in an "AI-first" press release five months after telling a podcast audience you'd be hiring more engineers. The AI-washing isn't just bad PR. It's corroding trust in a technology transition that actually needs people to believe in it.
So is there a bubble? Yes. I think Gurley is right. Some form of correction is coming. Too much money chasing too few proven business models with too little revenue to justify the valuations. The spending is outrunning the adoption. But I also think, as Gurley himself says, the underlying wave is profoundly real. Which means the correction won't kill AI. It'll kill the companies that confused buying AI infrastructure with actually using it.
Also Worth Knowing
- Jensen hosts an open frontier models panel today at GTC. At 12:30 PM Pacific, Huang moderates a discussion with leaders from A16Z, AI2, Cursor, Thinking Machines Lab, Black Forest Labs, and Reflection AI on where open models stand against the closed frontier. NVIDIA has reportedly committed up to $26 billion to open model development. For anyone building agent systems, the quality and availability of open models is the single biggest variable in whether you can run persistent workflows at manageable cost. I've been running into this firsthand building multi-agent systems, and I'll have more to say about the open vs. closed tradeoffs in a future column.
- The SaaS selloff is overcooked. Salesforce and ServiceNow have each shed roughly 25% of their value since January. The broader software ETF (IGV) is down about 20% for the year. The narrative driving the selloff is that AI agents will automate enterprise workflows so cheaply that existing SaaS tools become irrelevant. I don't buy it. The equity repricing may be justified on valuation grounds alone, but the idea that Salesforce or ServiceNow are going the way of the dinosaur is a fantasy held by people who have never tried to rip out and replace enterprise software that's wired into every department of a Fortune 500 company. These platforms aren't just products. They're ecosystems with decades of workflow logic baked in. AI agents won't replace them. AI agents will run on top of them. Gurley's advice is to figure out your buy price now, and I think he's right about that. Some of these names are going to look very cheap in hindsight.
The Builder's Take
I've spent three days writing about GTC , and the story I keep coming back to isn't about chips. It's about a widening gap between what's being built and what's being used.
NVIDIA showed us the full stack: Vera Rubin for infrastructure, Groq 3 for decode speed, NemoClaw for governance, OpenShell for runtime. The hardware is real. The cost reductions are coming. The agent infrastructure is materializing. And yet, the Cognizant study I cited Sunday found that 63% of enterprises still report moderate-to-large gaps between their AI ambitions and their actual capabilities. The infrastructure is racing ahead of adoption. That's what creates bubbles, and it's also what creates enormous opportunities for the people who close the gap.
Here's my position, and I feel more confident about it than I did 72 hours ago: when the correction comes, it won't punish the companies that did the work. It will punish the ones that bought the narrative. The ones that hired the consultants, purchased the platform licenses, slapped "AI-powered" on an investor deck, and never once sat down to decompose a single workflow into tasks an agent could reliably execute. I know this because I build agent systems. The difference between a demo and production is entirely in the decomposition, the integration, the boring operational work of wiring intelligence into how your business actually runs.
The infrastructure is being built for you. The cost curve is bending in your favor. Jensen envisions every engineer getting an annual token budget the way they get a laptop budget. I think he's right. The question is whether you're doing the work to spend that budget effectively when it arrives.
If you are, a market correction isn't a threat. It's a clearing event that takes out the tourists and leaves more room for the builders. And I don't know about you, but I'll take that trade every single time.
Keep building,
— JW
