I last wrote about NVIDIA ((NVDA Quick QuoteNVDA - Free Report) ) as the Bull of the Day in early September when it seemed like estimates had surged so much that they couldn’t go any higher.

This was two weeks after the company’s August 23 quarterly report and most analysts had made their upward revisions based on strong guidance from Jensen Huang & Co.

But the full wrath (and math) of the AI tsunami that NVIDIA has given life to is still being underestimated by Wall Street analysts and investors.

And that means the sales and profit estimates for NVDA keep getting revised upwards.

For perspective, here was the update I gave on September 6...

Because look what happened with the Starship Jensen’s results and guidance: They blew the roof off with a roughly $2 billion beat on the Q2 top line and then forecasted almost a $4 billion beat for the current Q3.

This means that the crowd of spreadsheet jockeys (Wall Street analysts), who have been underestimating NVIDIA’s growth in hyper-scale accelerated computing, have spent the last two weeks punching new, much more bullish assumptions into their models that keep NVDA a Zacks #1 Rank.

Here are where their consensus revenue projections stand now (9/6) on the Zacks Detailed Estimates page...

FY’24 (ends January): $53 billion for 97% growth
FY’25 (begins in Feb): $76 billion for 43% growth

And there’s profit optimism too...

FY’24 EPS consensus just moved from $7.79 to $10.46
FY'25 EPS consensus just moved from $10.77 to $15.48 (48% annual growth after the current year’s 210%+ advance!)

Based on these revenue and profit estimates for next year, NVDA trades at 30X EPS and only 16X sales. As I described in my June video and article, NVDA should continue to trade at a premium of 20X sales as it progresses to $100 billion in annual revenue...

Nvidia DGX: Workhorse of AI Will Drive NVDA to $2 Trillion

(end of excerpt from September 6 report)

The Moonshot to $100 Billion in Sales

In that June video and article, I suggested that NVIDIA sales could climb to $100 billion annually by 2026. At the time, sales estimates for next year (FY’25 which ends in January of 2025) were just beginning to breach $50 billion after their May earnings and product announcements.

Fast forward just over 3 months and here’s where the consensus views have ascended to...

FY’24 (ends January): $54.4 billion for 101.7% growth
FY’25 (begins in Feb): $80 billion for 47% growth

At a $1.1 trillion market cap, NVDA now trades under 14X next year's sales.

And the profit optimism persists too as the premier maker of GPUs commands top margins. Here’s a look at the move higher in EPS estimates just in the past four weeks since my last public report (I update for members every week)...

FY’24 EPS consensus just moved from $10.46 to $10.74
FY'25 EPS consensus just moved from $15.48 to $16.32 (52% annual growth after the current year’s 221.5%+ advance!)

$100 Billion -- Sooner Than Even I Imagined

Keep in mind that these are consensus views. The high end estimate on next year’s topline is $116 billion.

Given the surge in demand for NVIDIA systems architecture like DGX Grace Hopper 200 supercomputers, even my optimistic expectations are getting run over quickly.

To wit, last week KeyBanc analyst John Vinh put out a note where he explained why he was raising his price target from $670 to $750.

After the firm's quarterly supply-chain channel checks, his top takeaway is that AI data center demand remained extremely strong, outstripping supply by about 20%.

In his model, that computes to datacenter revenue of $45 billion for fiscal year 2024, up over 100%.

Vinh reported that while China’s demand remained weak, it didn’t deteriorate significantly. More importantly, Vinh sees a new opportunity for Nvidia in the second half, driven by “very strong” demand for L40S, a GPU designed for datacenter workloads like generative AI and large language model inference and training.

He estimates a revenue contribution of $2.25 billion to $3 billion from this next-gen GPU.

I haven't talked much about the new L40S platform yet because we're still in a revenue phase driven by a mix shift from Ampere 100 chipsets to the higher ASP (average selling price) Grace Hopper 100 "admirals."

A DGX system which holds eight GPUs, can command a price tag between $200,000 and $300,000 depending on configuration and availability (i.e., demand outstripping supply).

But the L40S will be another game-changer for NVIDIA in datacenters because the workload capacities are hyper-industrial strength.

Breaking down why Vinh sees NVIDIA DC doing $45 billion this year, he predicts that the Microsoft/Open AI alliance will procure 450,000 Nvidia GPUs this year, translating to $11 billion in revenue, and 1.6 million GPUs next year, contributing $40 billion in revenue.

These estimates are based on ASPs for GPUs of about $25,000, which is really the low end in this demand-driven environment. The real market is closer to $40K each.

And Apple ((AAPL Quick QuoteAAPL - Free Report) ) is expected to purchase 200,000 H100 GPUs next year, representing $5 billion in revenue.

“We believe NVDA has order and backlog visibility through 1H24 at this point and have not heard of any order cuts from its major CSP customers,” Vinh said.

To wit, KeyBanc expects Nvidia's data center revenue of $101 billion for fiscal year 2025, up 124%.

CUDA the Kraken

In my September report, I reviewed the “killer app” that makes NVIDIA GPUs so special...

For the past seven years since I first learned about what Jensen was creating from GPU gaming cards, I’ve said that the key to NVIDIA innovation dominating the world of hyperscale accelerated computing revolved around three factors...

1) The capabilities of GPU stacks creating "massively parallel architectures" for harnessing big-data with modeling, automation, and simulation

2) The CUDA (Compute Unified Device Architecture) hardware + software stack that enables fast training and deployment of machine learning and deep learning models

3) The evangelism of thousand of developers who get ingrained in the platform tools and never want to leave that ecosystem

Long-time Apple evangelist Gene Munster recently said the same thing...

“CUDA has created a moat around Nvidia's chip business. It would be difficult to get developers to switch to a different platform.”

And that’s why observers like me and Dan Ives of Wedbush think that this is the "iPhone moment" for NVIDIA and its AI tools.

In conclusion, let's hear from the AI wizard himself, Jensen Huang...

"The world has something along the lines of about a trillion dollars' worth of data centers installed in the cloud and enterprise. And that trillion dollars of data centers is in the process of transitioning into accelerated computing and generative AI."

I think this translates into NVIDIA hitting $100 billion per year in revenue much sooner than my 2026 projection in June. So don’t miss your chance to buy NVDA under $450 again. I’m pretty sure you won’t see it below $400 ever again.

(end of excerpt from Sep 6)

As I write this on Friday afternoon, NVDA shares are popping back above $450 after a trip down to $410 on September 21. And I also see a new article by the Wall Street Journal’s Dan Gallagher in his Heard on the Street column titled “How Nvidia Got Huge — and Almost Invincible.”

Here’s a quote that sums up the built-in invincibility of the developer moat…

Over time, CUDA has grown to encompass 250 software libraries used by AI developers. That breadth effectively makes Nvidia the go-to platform for AI developers; Credit Suisse analysts said those libraries "provide a starting point for AI projects that aren't available on non-NVDA systems, " in a report earlier this year. During a speech at the Computex conference in May, Nvidia's Huang said CUDA was downloaded 25 million times over the last year, which was more than double the software's life-to-date downloads prior to that.

Looks like I wasn't crazy optimistic enough calling for $100 billion in sales in 2025. Let’s make that call $125 billion now.