Nvidia's GTC conference starts Monday, and the stakes couldn't be higher. The market is three weeks into a losing streak, oil is near $100, and Jensen Huang is about to walk on stage carrying the weight of the entire AI investment thesis on his shoulders.

Here's my honest take: the stock doesn't just need a good presentation. It needs Jensen to make a credible case that AI capex is the last thing hyperscalers cut when margins get squeezed. And right now, margins are getting squeezed.

Amazon, Microsoft, and Google have collectively committed to spending hundreds of billions on AI infrastructure over the next few years. But those commitments were made when oil was $70, inflation was cooling, and the market was in full bull mode. The environment has changed. The question is whether the AI buildout survives the macro headwinds or gets delayed.

What I'm watching for at GTC: concrete Blackwell Ultra timelines with actual delivery dates, not roadmap slides. Investors need to know when the next-gen chips are shipping and how much demand is locked in. Vague promises won't cut it anymore.

I also want to hear new enterprise or sovereign AI customer announcements. The hyperscalers are the biggest buyers, but if Nvidia can show expanding demand from governments, financial institutions, or new industries, that de-risks the thesis. Right now, too much of the growth story is concentrated in a handful of customers.

The China exposure is another live issue nobody's pricing in cleanly. The H200 export restrictions are still being litigated, and if the U.S. tightens controls further, that's a material revenue headwind. Jensen needs to address this head-on, not dodge it.

Here's what would send the stock lower: a repeat of last year's conference energy without new substance. The market in February 2026 was willing to give Nvidia the benefit of the doubt on hype. The market in March 2026 is not in the same mood. Investors want numbers, timelines, and proof that the AI capex cycle is real and sustainable.

One thing I keep thinking about: Nvidia's current valuation is pricing in flawless execution and sustained growth. There's no room for error. If Jensen delivers a conference that's just "pretty good," the stock could sell off hard because the bar is set so high.

For what it's worth, I ran through the valuation math. At current levels, the stock is pricing in the assumption that AI infrastructure spending grows for years without a pause. That might be right, but it's not a sure thing. And in a market that's already on edge from geopolitical risk and inflation fears, any hint of weakness gets punished.

The bull case is simple: AI is the most transformative technology shift since the internet, and Nvidia has a near-monopoly on the picks and shovels. If you believe that thesis, GTC is just a speed bump. The long-term trajectory is intact.

The bear case is also simple: the stock has run too far, too fast, and the macro environment is deteriorating. Even if AI is transformational, valuations matter. And right now, Nvidia is priced for perfection.

So what am I doing? I'm watching from the sidelines with cash ready. If Jensen delivers and the stock sells off anyway because the market is in risk-off mode, that's a buying opportunity. If he underwhelms and the stock tanks, I'm staying away until the valuation resets.

Bottom line: GTC isn't just a product launch. It's a referendum on the AI investment thesis. And in a market that's already nervous, Nvidia doesn't get a pass. Jensen needs to deliver, or the AI rally is in serious trouble.