Picture the scene at some fund manager’s desk in early 2026. Two tabs open. One shows Nvidia’s fiscal 2026 numbers: $215.94 billion in revenue, up 65.47% from $130.50 billion the year before. Earnings of $120.07 billion. The other tab shows a headline about Alphabet selling AI accelerator chips to outside customers. The manager stares at both, then sells Nvidia.
That’s the whole story right now. Not a collapse, not a bubble popping, just a stock that keeps running at the same ceiling and bouncing off it because one competitor finally started shipping something people want to buy.
What actually changed
Nothing about Nvidia’s business deteriorated. The company posted revenue growth that would be considered a typo at most firms its size. Earnings of $120 billion is more profit than most companies generate in gross sales. Analysts still like the long-term picture, and they’re not being sentimental about it — the numbers support them.
What changed is that Google’s chip effort stopped being a slide in a keynote and started becoming a product line with financial and performance data behind it. That distinction matters more than any single quarter of Nvidia results. For years the story was that Google built TPUs for Google. Internal silicon. A cost-optimization play for one enormous customer that happened to be its own parent. Nobody priced that as a threat because it wasn’t one — it was a customer choosing to make instead of buy, and that customer was still buying plenty.
Selling those chips to other people is a different business entirely. It creates a second vendor in a market that has functionally had one, and markets with two credible vendors price very differently from markets with one.
Why the stock reaction is rational even though the fundamentals are fine
This is the part that confuses people who only look at earnings. Nvidia’s current valuation doesn’t rest on 2026 revenue. It rests on the assumption that Nvidia captures most of the AI compute buildout for years, at margins nobody can undercut. That assumption is doing the heavy lifting, not the $215.94 billion.
The moment a second supplier appears with real performance data, the market has to ask a question it has been comfortably ignoring:
- Does Nvidia still set the price, or does it now negotiate?
- Do the largest buyers keep single-sourcing, or do they split orders to create competitive tension?
- How much of Nvidia’s moat is silicon and how much is CUDA and developer habit?
None of those questions have clean answers yet. But you don’t need answers to move a stock. You just need the questions to exist. A stock priced for certainty reprices the instant certainty becomes probability.
What this means if you build with these tools
Here’s where I break from the finance coverage, because I don’t especially care about the share price. I care about what developers and teams actually pay to run models, and on that front this is good news that will arrive slowly.
Competition in accelerators shows up at the application layer eventually, but it takes a while and it takes an ugly form first. The first phase is fragmentation. More hardware options means more runtime targets, more quantization quirks, more “works on one stack, mysteriously 30% slower on the other” debugging sessions. If you’ve been building on one vendor’s software stack because it was the only serious option, you’re about to inherit portability work you didn’t ask for.
The second phase is the good one: pricing pressure on inference. That’s where most teams actually spend money, and that’s where a credible second supplier does the most useful damage. Nobody running an agent in production is going to notice which chip served the tokens. They will notice the bill.
My honest read is that the software lock-in is stronger than most people assume and weaker than Nvidia would like. Years of tooling, kernels, and tribal knowledge don’t evaporate because an alternative exists on a spec sheet. But the biggest buyers — the ones with in-house infrastructure teams and enough volume to justify a migration — are exactly the customers who can absorb that switching cost. And they’re the customers who account for the revenue concentration that makes Nvidia’s growth look the way it does.
The unglamorous conclusion
Nvidia is a company growing 65% a year whose stock is stuck because investors are recalculating how long the good part lasts. Google didn’t break anything. It introduced doubt into a story that was priced without any.
For anyone shipping AI products, the practical takeaway is modest and useful: stop designing your infrastructure around the assumption of one hardware vendor. Not because Nvidia is in trouble — the numbers say otherwise — but because optionality is about to be worth something, and the teams who built for it will be the ones who get paid when prices move.
🕒 Published: