Trending
I’m one of the many analysts and investors who’s been bullishly banging the table on Nvidia (NVDA) for years. We’ve been long the stock since September 2020 in my paid Disruption Investor advisory.
But unlike a lot of other folks, I never soured on its prospects and turned bearish when it was fashionable… like much of 2026 until yesterday when the stock popped almost 9% and gained more than $440 billion in market cap after its fiscal Q2 earnings call.
Not because I’m special. Trust me, I’m not.
But because I’ve been following the money, the business, and the simple math. Not some “guru,” “influencer,” or talking head.
In fact, I’ve been saying Nvidia would become the first $10 trillion market cap company—and that it could get there before 2030, and possibly as early as the end of 2028—here in Grow or Die since November 2025: NVIDIA $10T.
I said it again in January 2026: What’s next for Nvidia? Explained the mechanics in February: How Nvidia becomes the first $10 trillion company.
And just this week, the day before the company released its blowout earnings report, I published: Don’t sleep on Nvidia.
That recommendation looks smart now. But it’s really not. It’s just a function of paying attention to the right things.
What I want to do for you today is explain those things so we can start to paint a picture of where Nvidia goes from here.
First, a brief recap of Nvidia’s incredible results…
Nvidia grew revenue 106% year-over-year from $46.7 billion to $96.2 billion in fiscal Q2 2027 (ended July 26, 2026).
That’s unprecedented. No company in history has ever organically doubled quarterly revenue from such a large base.
The closest I could find was Exxon Mobil in Q2 2021. It increased revenue by about 108% year-over-year that quarter. But that was a rebound from the COVID oil-price collapse. And the year-earlier quarter’s revenue was only about $32 billion because crude had crashed.
It’s practically a law of business physics that a company the size of Nvidia can’t double revenue as fast as it just did. Because the bigger the base, the harder it is to grow on top of it.
Nvidia’s “Data Center” revenue, which is the AI business, grew 117% year-over-year in Q2 to $89 billion. The company guided for total revenue of $108 billion this quarter, which would be its first $100 billion quarter. And it guided for 70% revenue growth in fiscal 2028, which is basically calendar year 2027.
As amazing as these numbers are, they don’t help us see where the company is going beyond about a year from now. What does help us start to paint the picture of where Nvidia could be several years down the road is four specific takeaways from the quarterly call (and some of the company’s other recent commentary).
Four important takeaways from the quarterly call…
#1 The customer list has broadened considerably
For two years, the most popular argument against Nvidia was basically: “Its whole business depends on four giant customers, all of which are building their own chips to become less reliant on Nvidia.”
It’s true that these four “hyperscalers”—Amazon, Google, Microsoft, and Meta—have been driving much of the AI infrastructure buildout. They’ll continue to do so for at least a few more years. And all four are indeed building their own AI chips these days.
But as the hyperscalers have increased their AI spend, so have lots of other companies. Governments around the world have too for their “sovereign AI” initiatives.
And we now have a lot more color on that broadening AI spend…
In Q1 Nvidia started separating its AI revenue into two buckets. Bucket one is the hyperscalers. Bucket two is everybody else, which Nvidia calls ACIE (short for AI Clouds, Industrial, and Enterprise). So ACIE covers the neoclouds, governments, hospitals, banks, carmakers, drug companies, AI startups, etc.
In Q2 (the quarter Nvidia just reported):
Bucket one (hyperscalers) generated revenue of $48.7 billion, up 13% sequentially and 101.2% year-over-year.
Bucket two (ACIE) generated revenue of $40.3 billion, up 25% from the previous quarter and 138.5% from the year-ago quarter.
The “everybody else” bucket is nearly half of Nvidia’s AI business, and it grew about twice as fast last quarter.
Beth Kindig, lead tech analyst at I/O Fund—still the sharpest Nvidia analyst out there in my view—ran the numbers and she figures bucket two could pass bucket one in as soon as 2 quarters, or 5 to 6 quarters using a more conservative model.
Meanwhile, most of the customers in the faster growing ACIE bucket will never even think about designing their own chips.
Jane Street and Hudson River Trading are running Nvidia systems to power their trading platforms. Drug giants like Eli Lilly and Roche (and soon Bristol Myers Squibb) are using Nvidia systems to accelerate drug research & development. Japan’s national AI company is building one. South Korea’s LG and Hyundai are building them. 35 new Nvidia-powered systems have recently been unveiled across Europe for sovereign AI initiatives. And new cloud companies have popped up in Armenia, Africa, Taiwan, India, Australia, and Malaysia, all powered by Nvidia.
None of these customers have AI chip design teams. And none of them want one. A drug company doesn’t want to spend years and a fortune designing a custom processor. It wants a machine that shows up on a truck and works.
What’s more, the hyperscalers have shown no signs of pulling back yet.
Nvidia CFO Colette Kress said it clearly on the call: “With cloud industry backlog now greater than $2 trillion, CapEx by the top 5 hyperscalers is expected to reach nearly $800 billion in 2026 and $1.3 trillion in 2027.” (Oracle is the 5th hyperscaler she referred to.)
And Amazon—which does design its own AI chips and has basically been Exhibit A for the whole “hyperscalers will defect” thesis—committed to deploying another 2 million Nvidia GPUs from now through the summer of 2028. Amazon is also adopting Nvidia’s full physical AI stack (Omniverse, Cosmos, Isaac, and Jetson) to power its fleet of warehouse robots.
So the giants are still buying from Nvidia, a new class of customer has showed up behind them, and both are compounding at once.
#2 Why AI chip market share is an increasingly meaningless metric
Some analysts claim Nvidia’s share of AI chips will shrink from about 90% today to more like 70% over the next couple years as the hyperscalers deploy more of their own silicon.
I think that’s probably about right. I also think it matters much less than most folks assume.
I used to frame my AI chip market share comments around demand. Basically, who cares what market share is if demand is so strong the company sells out of every chip it can make.
But something a bit more subtle is going on that’s worth pointing out. It has to do with revenue per gigawatt. We know the company thinks it matters because CFO Kress brought it up in her prepared remarks and CEO Jensen Huang brought it up again in the Q&A portion of the call.
The gist is that during Nvidia’s Hopper generation—the chips that powered the initial ChatGPT boom—the company captured about $18 billion in revenue per gigawatt. That increased to about $25 billion per gigawatt with the current generation, Grace Blackwell. With Vera Rubin, the generation that just started shipping, revenue per gigawatt is about $40 billion.
This matters so much because power is the scarcest resource now. You can find land. You can pour concrete. You can’t conjure a gigawatt out of thin air. That takes turbines, substations, transmission lines, and years of permitting.
So what actually drives Nvidia’s sales isn’t its AI chip market share. It’s how much revenue the company can squeeze out of every gigawatt of deployment.
Nvidia’s revenue per gigawatt continues to climb because it keeps adding parts to the box. With Hopper it basically sold GPUs and one flavor of networking. With Vera Rubin, it’s selling GPUs, CPUs, its NVLink internal networking, your choice of two external networking systems, and a new speed-focused inference chip from its Groq partnership.
It’s sort of the difference between selling someone an engine and selling them the engine, transmission, chassis, wiring harness, cooling system, and the software that runs the dashboard.
My point is that if Nvidia’s share of AI chips falls from 90% to 70%, but its revenue per gigawatt continues to grow fast, and the number of gigawatts being built continues to grow fast, Nvidia continues to grow fast.
This concept of revenue per gigawatt is basically the other side of the coin represented by the next takeaway I want to mention…
#3 The metric that matters most now: tokens per watt
As you may know, tokens are like AI’s units of intelligence. The standard rule of thumb for English is that 1 million tokens equals about 750,000 words.
For years, AI chips were judged on raw computation power (FLOPS). How many operations the chip could carry out per second.
Today, a much more useful measurement is tokens per watt. How much useful AI output you can get out of each unit of electricity.
The reason: when a data center is consuming all the power it can, that’s it. You can’t add more compute to get more units of intelligence. You need to add electricity. And that could take years in today’s environment. So the only way to grow output at a site you’ve already built is to make each watt produce more intelligence.
This is also the reason customers are lining up for Nvidia’s new Vera Rubin AI systems. Nvidia says Vera Rubin delivers 30X more tokens per megawatt than Grace Blackwell Ultra. That’s best-case scenario. CoreWeave tested a Vera Rubin rack in the real world and measured about 10X the tokens per megawatt compared to the prior generation.
10X is not 30X. But it’s still staggering.
Customers couldn’t care less that Nvidia is raising prices by about 15% on its new systems (due to skyrocketing memory prices) if they’re getting 10X the units of intelligence per unit of electricity.
That’s the engine underneath everything else in this piece. It’s how Nvidia increases its revenue per gigawatt and improves its customers’ economics at the same time.
#4 Why the “circular financing” you’ve heard about doesn’t concern me much
Nvidia is investing in, guaranteeing, and backstopping the very customers that buy its products.
Critics call this circular financing and point to the dot-com era, when telecom equipment makers lent money to startups so those startups could buy their gear. When the startups died, the lenders got crushed.
While this phenomenon is real and deserves scrutiny, I’m not that concerned about it at this point.
For one, the scale is completely different. Nvidia currently has about $165 billion in these types of financing arrangements. That’s only about 1.3X Nvidia’s trailing twelve-month free cash flow. And only about half of my two-year forward free cash flow forecast.
The dot-com telecom vendors were financing customers with balance sheets and cash flows that couldn’t justify it. That’s not what’s happening here.
What’s more, what Nvidia is financing is much more valuable. If a customer folds, what’s left behind is the same Nvidia racks that run most AI models in existence, in every cloud, for every workload. And there’s a line of buyers stretching out the door. Compare that to dot-com fiber optic cable, which sat dark and unused for a decade.
So, while it’s not zero risk, “circular financing” as you hear about it in the media and what it really means for the business are two different things.
So where does the stock go from here?
In late 2025 and early 2026 I published my math for how Nvidia becomes the first $10 trillion market cap company. I forecast data center revenue (remember that’s AI revenue), of $390 billion for calendar 2027 and $550 billion for calendar 2028, and applied a price-to-sales multiple of 17 to 19.
Nvidia’s FY28 guide implies about $690 billion in total revenue for what is essentially 2027. Data center revenue is running around 93% of the total, so call it $640 billion of AI revenue.
In other words, my revenue forecast was too low. By a lot. Nvidia should clear my calendar-2028 AI revenue forecast a full year early.
Meanwhile, after this week’s move, Nvidia is worth about $5.3 trillion. That’s about 17.5X trailing twelve month (TTM) revenue. When I first made the $10 trillion call in November, the stock was trading at about 30X TTM revenue. Against this fiscal year’s expected revenue, it’s about 13X. And against next year’s guided $690 billion, under 8X.
So the stock got much cheaper as the business grew dramatically.
Now let’s look forward to fiscal 2029 (basically calendar 2028). Nvidia hasn’t provided any guidance that far out but it has said supply stays tight through the end of fiscal 2028 at least and underlying demand is still running near 100% growth.
I think a forecast for fiscal 2029 revenue growth of 40% is still conservative. That puts revenue near $970 billion.
To be worth $10 trillion against $970 billion of revenue, Nvidia would need to trade at about 10 times sales.
10, versus over 17 today, and about 30 when I first made the call.
If Nvidia holds anything close to today’s profitability, $970 billion in revenue will produce well over $500 billion in net income. A $10 trillion company earning $500 billion trades at 20 times earnings. That’s completely reasonable (conservative even) for a business still compounding 40% a year.
These aren’t precise numbers of course. It’s just a framework… and every input can move.
The point is that the math to a $10 trillion market cap is much easier today than it was just a year ago.
There will be pullbacks in the quarters ahead. Sometimes sharp ones. But in about two years I think Nvidia’s business will be able to easily justify a $10 trillion market cap.