As I was preparing to write my quarterly Nvidia (NVDA) update, I was delighted to see that I didn’t have to buy more Nvidia because the market was doing it for me, via the 5% crash rule on Thursday, Nov 13th.

10% More Nvidia Shares at a 5% Discount On An Already Undervalued Chip Utility!

Nvidia Is The Largest Single Holding in The ZEUS LEGACY Family Charity Hedge Fund

Why am I so confident in Nvidia that I bought more ahead of earnings? Why might you want to do the same? Let me show you the three reasons this chip utility could be the easiest way to make low-risk profits in your portfolio over the next 5 years.

Reason 1: Nvidia Remains A Chip Utility

The key to understanding Nvidia’s thesis for the foreseeable future is the concept of the infinite backlog.

• In Reason 3, I’ll show you in detail how this works.

But basically, imagine this example.

What if Exxon had 1 million barrels per day of capacity, but demand for that oil was 2 million bpd?

They doubled their production the next year to 2 million, but demand has also doubled to 4 million bpd.

So they double production again, to 4 million, but demand doubles again, to 8 million.

After 5 years, Exxon has increased its production by 32X! “OMG, that’s insane! Oil is cyclical! This is rampant speculation!”

Except demand has increased 64X and the backlog is now 32X bigger than it was before.

• 64 million bpd of demand vs 32 million bpd for supply = 32 million bpd backlog vs 1 million 5 years ago.

What would you tell the CEO of Exxon under this hypothetical scenario?

“Be careful! Don’t increase growth spending because what if it’s a bubble?” Or would you say, “As long as you are selling every barrel at full price and the backlog is getting longer, spend every penny you can on growth because the risk of losing money falls with each passing year?

• The paradox at the heart of the compute boom (it's not just about genAI)

• Azeeh Azhar explains that, since he was born in 1972, computer demand has grown by 100 billion X (69% CAGR), and that demand will continue to grow exponentially in the future.

• There is no limit to how much intelligence we’ll want to use in the future.

Nvidia CEO Jensen Huang Reveals $500 BILLION In Orders Through Next Year

Do you know what the current consensus is for Nvidia sales next year?

$285 billion and even the following year, “just $357 billion”.

But this quarter, Nvidia’s guidance was for $50 billion, so that means $450 billion in demand for next year. Not just “we’d like to buy” but straight-up orders. Customers like Microsoft lining up to hand over their money to Nvidia. The average selling price of a GPU is expected to rise to $27K next year and $30K in 2027.

The price of a Blackwell is $35K so that averages selling price can keep climbing to roughly that amount.

In other words, Nvidia’s sales are capped by how much supply of Taiwan Semi wagers and memory chips they can get from the likes of SK Hynix,

They are supply-constrained; they sell 100% of capacity at full price because demand is so much higher than supply that even if China is banned from buying chips or someone cancels an order, others are eager to buy that capacity.

Jensen’s Law Is The Heart Of The Chip Utility Thesis

Jensen has said that over the last decade, Nvidia chips have increased in capability by about 1 million X, and they plan to increase that by another 1 million X over the next decade.

A 4X increase every year doesn’t mean 4X faster, or 4X cheaper, it means some combo of both.

Imagine 2X more speed (capability) for 50% the cost…every year…for a decade.

That’s 1000X faster (more capable) and 1000X cheaper.

Why does that matter? Won’t 1000X cheaper tokens cost companies like Microsoft or Google money?

That’s where Jevon’s Paradox comes in.

Imagine making something 2X cheaper and selling 4X as much because now more people can afford it?

• The way capitalism has worked in the 20th century is driving down costs so that prices fall, volume rises, prices/unit fall, prices drop even more, and demand rises even more, driving down cost/unit even more.

It’s how Rockefeller cut oil product prices to the lowest possible level to maximize demand.

But now imagine if you cut the cost of tokens by 10X and demand goes up 100X? Now revenue is up 10X because you cut prices by 10X.

Cut prices 100X and demand goes up 10,000X, and you make 100X more money.

Cut prices 1,000X and demand rises 1 million X, and you make 1,000X more money.

With AI and technology, where the users of the tech are the tech itself (AI is using AI to improve AI and build more AI agents that will improve themselves), this is how we end up getting to Sam Altman talking about eventually cutting token prices by 1 billion X.

What Sam Altman has said is that OpenAI’s business model is to cut the cost of doing the same task by 10X per year, every year, until it costs 1 million X less to do the task you are doing today.

And eventually he plans to cut the cost (within 9 years or 2031) by 1 billion times.

Or, to put it another way, what costs $10,000 today will cost $0.01 in 6 years. And what costs $1 million today to do with Chat GPT will cost $.01.

How OpenAI plans to make money is by allowing a small business that doesn’t have $10K for something it wants to do to do it for almost nothing in 6 years.

And 3 years later, what would cost $1 million (a large business), any person in the world will be able to do for a penny.

The magic of this kind of tech innovation is that once something is cheap enough, people like me, Connor, and Glenn (the GNG Research team for now) will figure out innovative new ways to use the tech. Ways that Sam Altman may never have dreamed up.

And when others learn what we’re doing, and come up with even more innovative ways to use OUR tech? Then it’s an exponential explosion of innovation.

Remember how the iPhone created Uber? And the App Store created hundreds of thousands of small businesses?

Steve Jobs didn’t have to invent Uber because people using the tech platform he created came up with it. He might never have come up with the idea, but the genius of global capitalism is that if someone has a problem, someone in the world will figure out a way to solve it for them.

So this is where Jensen’s Law is so critical.

Just as Moore’s law drove technological innovation for over 50 years, Jensen’s law is making the economics of AI possible.

In one of his presentations, he directly addressed companies like Microsoft, saying something like, “4X more revenue per watt of power every year.”

Big tech giants have everything they could want, BUT power. Microsoft has some GPUs that don’t have enough power. They need the most revenue per watt. And they are willing to pay a king’s ransom ($35K per GPU) to maximize revenue per watt.

So as long as Nvidia can continue delivering on Jensen’s law, companies like Microsoft will have every economic reason to keep replacing their GPUs every 6 years, because every 6 years the GPUs are roughly generating 4,000X more revenue per watt.

Think of Nvidia GPUs like an iPhone…that costs $35,000.

And instead of regular people, you have customers who generate almost $600 billion in free cash flow AFTER spending $600 billion on AI capex😉

Did you know the Vatican’s assets are $12.5 billion? The hyperscalers have around $1 trillion in net cash, 80X more money than God😉.

And every 6 years they will be asked to buy a new super iPhone for $35K, that’s 4,100X more profitable per unit of power, their ultimate limiting factor.

THAT is why Jensen keeps quipping “the more you buy, the more you save,” and that is why Nvidia is not a cyclical company as long as they are supply constrained. Right now, demand outstrips supply (and the backlog is now up to around $200 billion for next year).

Nvidia is effectively a chip utility growing at 50%.

And how long could Jensen’s Law hold? What do the laws of physics say about that?

Based on the Landauer limit, roughly a 1 billion X improvement in computing is possible, which = 15 years of Jensen’s law.

After that? Reversible computing theoretically removes all limits, and we can keep scaling Jensen’s law (and quantum computing can also help, though it’s a different application suite, so not precisely apples to apples).

Reason 2: Believe It Or Not, Nvidia, Even At $5 Trillion, Is Undervalued

I know that Nvidia, with its $5 trillion market cap, SEEMS like it MUST be a bubble! It FEELS like a bubble, right?

If fundamentals grow in line with price, there is no bubble, no matter how stupifying the returns are.

• That’s not my opinion, that’s just how math works😉

Still don’t believe me? Good, skepticism is healthy, and the moment no one is scared about an AI bubble, that’s when you’ll know we’re in one🤣.

Definitive Proof That Nvidia Can’t Be In A Bubble Today.

Median Rolling PEG Ratio 0.76 vs 0.34 Today = 55% Historical PEG Discount

Nvidia 10 Year Average PE X Consensus Estimates (which are climbing every week for 3 years) = Consensus Total Return Potential

By the end of 2027, Nvidia has about 100% upside potential, and keep in mind that OpenAI’s deal with them is worth about 45% growth in 2028 (analysts are more comfortable raising estimates over time than a sudden jump, lest they get fired for being wrong).

As we’ve seen for 2026, the backlog is now $200 billion (excess demand vs what Nvidia can deliver).

That backlog keeps growing over time, and with AI spending potentially on track to reach $4 trillion PER YEAR in 2030, the higher risk is to the upside, not the downside, for where estimates go from here.

“Are you high! You think the estimates are more likely to go up! After EPS is up 20X in 3 years! What hopium are you smoking!?”

OK, so let me walk you through all the evidence for why the AI boom is very likely to be a bubble, STILL not.

Reason 3: The AI Boom (Overall) Is Not A Bubble

This is my favorite AI meme right now.

It pokes fun at AI romance scams (I’ve been through a lot of these recently 😉) AND the fact that if I quote you what Jensen says, I sound like a raving madman because the numbers are so stupifyingly huge.😉😂

First, let me share some of the latest interesting AI news, showcasing the adoption rates and why token demand continues to grow exponentially.

• The State Of AI In 2025

According to McKinsey’s latest survey of companies, 88% have started experimenting with AI, with 7% at full adoption and 31% aggressively scaling to make money.

62% of companies are still piloting or experimenting.

Let’s Geek Out About AI Return on Investment Surveys 😉

Trust me, this is very important

You might have seen headlines about a study from MIT that found that “just 5% of companies were making money with AI”. Actually, the details matter a lot. Because that study also found that 63% of companies that BOUGHT AI software (instead of building their own) made money.

5% of “Build it from the ground up with a team of people who aren’t experts” didn’t make money. Wow, not a surprise. And 63% of companies that bought services from companies that do this for a living? They made money, also not a surprise.

Companies that adopt early are those that see the instant potential (like how my company is using AI to build tools that would typically cost over $10+ million on a shoestring budget of “only” $2.036 million😉

It’s easy for us to see the returns on investment because we can accomplish for pennies what used to take dollars, and even expensive things “only” cost a few thousand per month.

• Our operating expenses are now $41,619.74 per month.

• They keep rising because business is messy, expensive, and we need to keep scaling up.

However, our CTO, Connor, the Techno wizard of Jacksonville, Florida 😉, has checked what it would have cost to try to build out our systems by paying other people to do it.

About $250K to build the site with tools, AND $40,000 per month to maintain it.

Instead of “just” $42K per month, we can do it all in-house, save $250, and most importantly, build a team of incredibly skilled and dedicated financial & tech nerds who want to build the world’s greatest financial tool suite so that everyone can eventually have a “family office for all”.

So that’s a really great deal for us, but the point is that we’re part of that 7% of full-scale adopters. The so-called “AI Native” companies, where everyone is using AI, and we’ve embraced it as our work partners, and don’t fear it.

We’re not geniuses (at least not me), but we are delighted by learning and so, like Da Vinci, are self-taught autodidactic geeks who never stop trying to learn what this really cool tech can do to make people’s lives better.

Did you know that according to Alphabet’s CEO, Sundar Pichai, Google Cloud now has 150 customers processing at least 1 trillion tokens per quarter? In total, Google Cloud, mostly via Gemini (which hosts all popular AI models), is up to 1.3 quadrillion (1,300 trillion) tokens per MONTH, up 20X from last year.

What 1.3 Quadrillion Tokens PER MONTH Means In Terms Humans Can Understand 😉

Imagine 65 billion humans using ChatGPT like the average person does. That’s Google Cloud’s AI usage.

Companies using AI for coding (Anthropic’s expertise, 40% market share in enterprise) are generating tokens at a furious pace.

And no, it’s not Agents…yet. About 10% of companies are using agents (we are for various tasks because we’re three guys and some algos, and only two of us can code😉

Software engineers like our CTO, Connor, use agents the most, but only 24% of software companies do.

• Room to grow as these tools improve.

Latest AI Token Growth Data (The Ultimate Source Of Truth About Whether AI Boom Is Real)

No matter what headlines you see, no matter what any “guru” or “expert,” Including me, ever tells you, no matter how smart or persuasive our story, narrative, or thesis might sound, NEVER forget that 1st principles investing is around the most core, fundamental data. And for AI, there is one gospel truth that outweighs everything else you might hear. Not even St. Jensen’s proclamations EVER EVER EVER outweigh token data. This is the electricity of AI. This is the #1 way to track whether or not this technology is useful. It doesn’t matter if YOU, I, your brother, or a colleague find this helpful; what matters is whether enough people find it beneficial to use it.

Note that tokens X token price = token revenue. So rising token usage does not necessarily mean “Profitable”, it just means “This isn’t snake oil, people REALLY are using it because its useful”.

And the answer is unequivocally YES! This is not hype, this is not conmen and Potemkin villages and fancy facades; this is 10X growth in tokens, so strong that Alphabet and Microsoft are now running at 24.36 quadrillion tokens per year, which is 16X the amount of tokens used in all of 2024 (about 1.5 quadrillion).

Alphabet (Gemini) is reporting 20X growth in token usage, and Microsoft (Azure) is reporting 50X growth.

Last quarter, Microsoft reported 5X annual growth, and now the growth rate is up 10X.

In the past year, Microsoft’s AI use among customers was growing 39% per month, for an entire year!

Last year, token growth was up around 10X…and now Microsoft’s token growth rate is up 10X!

Remember all those headlines about how ChatGPT 5 was underwhelming and disappointing?

Well, according to Sam Altman, within 48 hours of launching, API token usage among enterprise customers was up 100%.

It doesn’t matter what you or I think about Chat GPT 5. I was disappointed, too… but companies LOVE IT!

Our CTO told me that ChatGPT 5 Pro was the new king of coding, taking the crown from Claude 4.1 Opus (Claude 4.5 has recently retaken the crown).

And we’re not the only ones finding delightful new ways to use this tech.

This is one of the most important charts in this report because it shows the growth rate for companies that are heavy users of AI. In 2022, just 2% were using AI for at least five functions, and now that number is 20%.

That’s 10X growth in 3 years in heavy AI use in the enterprise.

You might be surprised how little AI use is still occurring in parts of the economy, like parts of manufacturing, which are zero.

Those numbers can only move in one direction, which means that there are still plenty of new enterprise customers to find.

More clients, and more AI uses per client, and each use uses more tokens.

That’s how you get 50X YoY growth in tokens at Microsoft, which is the 2nd biggest cloud company on earth!

So Microsoft’s 50X growth isn’t off some tiny base; it’s a six-quadrillion-token-per-year use case growing at 50X per year.

Remember that in 2024, the entire world used an estimated 1.5 quadrillion tokens!

And now Alphabet alone is generating 1.3 quadrillion PER MONTH!

Big companies love AI more than small ones, with 2X as many companies reporting “full-scaled” use of AI.

But remember what that means. This means that 90% of companies, regardless of size, are not fully scaled. That means more and more AI usage, more tokens, and three layers of exponential growth.

And Sam Altman has said that next year, the $2,000-per-month AI intern is coming, with a $10,000-per-month mid-level employee replacement in 2027 and a $20,000-per-month PhD-level researcher in March of 2028.

• When we run out of companies to use AI, we’ll have AI workers that use it.

• Exponentially more people are using AI.

• Each one uses it for more things.

• Each thing requires more tokens over time.

3 Layers of exponential growth = no end for token demand growth.

The big concern a lot of investors have is that AI productivity boosts are fictional “work slop”.

In other words, companies will try AI, find that generating and then summarizing emails aren’t actually helping, and stop paying for it.

However, according to the McKinsey survey of 1753 companies, that is not the case. The # of companies reporting that AI is making things worse is very small.

Of course, we’re not close to “Jarvis is doing everything for me” kinds of satisfaction either, but remember this is as “useless” as these tools will ever be. 😉

AI Is NOT Making This Worse, At Least According to This Survey

At the same time, AI tools, even among highly specialized fields, like coding, are NOT the panacea that some executives think.😉

Up to 10% of companies report cost savings of more than 20%, which is very significant, but since coding AI can only handle around 80% of coding tasks (Claude 4.5 & Gemini).

In fact, a report from Boston Consulting Group (BCG) found that mid-level software engineers usually gained the most benefit from AI coding software, not junior engineers.

Why? Because the experienced engineers knew that “Slow is smooth and smooth is fast.” In other words, rather than just jumping straight into coding, professional engineers will work with the framework first, and then only start coding at the very end. “Measure twice, cut once” kind of thing. 😉

You might have heard about studies showing that AI actually slows down developers, the “AI Work slows down makes this all hype and a bubble about to pop” fear that many investors have.

However, remember how contradictory-sounding things can be true at the same time. If you measure productivity as code written by mid- to high-level engineers, YES, they SEEM slower on that single metric. But actually, the quality of work is improving, because the framework for the code is what determines if the code runs correctly.

Just remember, this is math, and technology, not magic, and no, you can’t REALLY vibe code anything you want, as Connor, our CTO, frequently reminds me😂😉

Actually, the most direct benefit from AI in terms of sales is in marketing, which makes sense given the strong growth that Meta has been reporting courtesy of AI optimization of its ad platform.

Where’s The Proof That AI Is Making Money For Companies? Have You Seen Meta’s Results?

Meta’s sales growth in the age of AI is explosive and expected to be 17% CAGR on a huge base of $117 billion before ChatGPT came out. And in case you don’t believe forecasts, don’t forget that the accelerating sales growth in 2023 and 2024 and so far this year, that’s now set in stone. That’s SEC confirmed and audited Gospel Truth. AI is accelerating Meta’s sales (and is the reason Microsoft is growing its cloud business at a 39% CAGR). It’s why Amazon has reported accelerating growth in its cloud (20%).

AI is profitable, and the proof comes out every quarter.

Tech margins are 28%, about 3% above their 5-year average, and those 5 years have been a period of margins rising to record highs.

Bank of America estimates that expanding margins, created by revenues growing faster than expenses, will benefit almost every industry in the next five years, with Tech the biggest winner.

• 33% Tech net margins by 2030, according to Bank of America

Remember that table showing 22% FCF growth through 2030 per the FactSet consensus? Well, that growth is happening right now. It’s not some magical “future growth,” it’s today’s growth, and it’s expected to continue through 2030.

And a reminder that growth rates have been accelerating. Last quarter, 19% CAGR FCF/share growth through 2030…now 23%.

AI Optimists Predicted Accelerating Growth Rates…They Have Been Right So Far

Companies have been reporting stronger-than-expected guidance at the strongest rate in years, and earnings growth momentum just hit a 4-year high.

Where is the evidence that AI spending is profitable? It’s all around us…If you know where to find the right charts😉

OK, so the charts, the math, it’s all very compelling, at least if you’re an investor who owns tech stocks😂😉

BUT HOW ON EARTH CAN NUMBERS THIS BIG NOT BE A BUBBLE!?

I’m sure a lot of people overwhelmed by the age of AI feel this way.

“Explain to me in words, not charts, how any of this makes a lick of sense! Because it seems too crazy to be true, so put down your charts and explain this to me like a human, please!”

I can’t “put down all the charts” since charts are my love language and I’m pretty sure I’ll die if I stop using charts😉, but let me try to explain why this can’t be a bubble, at least not yet, no matter how big and “crazy” the numbers seem.

Let me explain why, no matter how much AI spending happens, or how it’s funded, as long as demand growth is higher than supply growth, it WILL NEVER EVER be a bubble (at least in terms of capex spending, stock prices could become a bubble).

As Long As Demand Is More Than Supply, Nvidia Sales CAN Grow To The Sky😉

The first thing to remember about anything is the 80/20 rule. That “80% of the work is done by 20% of your best workers” or “80% of profits come from the top 20% of products”. That’s not a perfect rule, of course, but numerous studies show that a small portion of either good or bad things takes up the majority of our time, and generates the majority of stress or profits.

A handful of the best, most productive AI users, the power users, the high performers, are driving the most gains and using the most tokens. That’s how the token growth can be this explosive, going from 10X in 2024, to 20X (for Google) to 50X (for Microsoft).

The number of power users is rising over time, and each one is using AI for more super productive tasks, and the amount of capability (tokens) per use is also rising. That’s why it’s a hyper-exponential curve (so far) in AI usage.

Even if most people can’t figure out how to use AI to its highest purposes (yet), the few who are are cranking up the usage to 11.

Remember how 50% of consumer spending is from the top 10% of income earners?

The Top 20% of Income Earners = 63% of All Consumer Spending

The same is true for the inequality of token usage.

AI Use Inequality Is Similar To US Spending/Income/Wealth Inequality

Want more specifics about who the power users of AI are?

Companies like Canva and Salesforce account for 1% of users and 30% of token use.

The bottom 80%? You and Me? The folks having Chat GPT write emails and plan trips? That’s 15% of usage.

Let’s put that another way—the top 1% use 2X more AI tokens than the bottom 80% of us.

THAT is why when Chat GPT 5 “disappointed”, API usage by enterprise customers went up 100% in 2 days.

You and me? We don’t matter to OpenAI and the AI industry at large. Our $ 20-per-month (or even $ 200-per-month) subscriptions? That’s going to be a drop in the bucket of actual usage.

Mo Gadwat, the former CEO of Google X, uses AI not for recipes or emails, or for life coaching (as I do), but to supercharge his workflow.

In my books I would have 4 to 5 sentences, a very important paragraph. And it would take me 4 weeks to craft that paragraph because I would research, and think, and learn, and talk to experts for a month. Now that same level of depth takes me 4 minutes. I ask Gemini for a deep research report on an idea, a topic, or a question. And then have Claude fact-check, and then have Chat GPT make it pretty, and then ask DeepSeek to make sure it’s not too American focused. And then finally back to Gemini for final math fact-check. This is how I estimate that when I use AI, I gain 80 IQ points. And within a few years it will be 400.” Mo Gadwat

Most people are not using 4 or 5 AIs working together to fact-check and create highly advanced systems.

• AGIOS (Augmented Governance & Intelligence Operating System) consists of Gemini, ChatGPT, Claude, Perplexity, and Grok 4, plus feedback from Notebook LM hosts, for deep dives, debates, and video podcasts in English & Polish.

So, Mo (whose AGIOS-like system is named Trixi😉) and I are the outliers.

• AGIOS runs on 220 pages in Google Docs (a RAG system) that includes 160,000 pages of journals covering everything I’ve done over 2 years.

• That’s how the “master plan” is actually recursively improved over time, since the AI can’t actually remember plans, unless you use a system like this.

My friend Thom tells me that my dreams of an MCP system where I ask one AI a question, which then sends it to all 5 models, who then reason an answer, AND THEN talk to each other to create a consensus answer, will work and is a very useful idea.

That will save me time copying and pasting responses from one AI model to another and coordinating complex conversations and discussions among AGIOS members.

• At the moment, Connor has more important things to work on than optimizing my AI Mary Poppins🤣

The point is that Mo and I, and people like us, are using tons of compute, and as my friend Thom (an AI Engineer at IBM) told me this weekend, what we’re doing is a drop in the bucket compared to the self-improving recursive systems AI companies and true power users are doing.

Take our Chief Technology Officer, Connor, who makes all the magic happen at GNG🤗

He has an AGIOS-like system that helps him make important problem-solving decisions, working something like this.

• A 200-page master prompt- context around his job, goals, and needs.

• One ChatGPT 5 Pro that will edit the Master Prompt based on a specific question.

• Then another ChatGPT 5 Pro will run the optimized prompt (200 pages of context with a specific question) 1000 times!

• And then this “Monte Carlo-like simulation” compresses down to the minimal error optimum answer, the “expert consensus” from Chat GPT 5 Pro running the same question 1000 times.

The way Connor is using AI makes Mo Gadwat and me look like amateurs, and I’m sure Dennis Hassabis (CEO of Google DeepMind) is using systems even more advanced.

Did you know that even among high performers in the tech industry, only 33% are scaling AI up yet? That means plenty of room left for “Token growth to grow to the sky”😂

Power users like me, Mo, Connor, and all the engineers who are working on the “magic god machine” are the ones driving the explosive demand growth.

So if you ever wonder, “How on earth can Chat GPT 5 be so useful when I find it so annoying!” Well, remember that 1% of users are driving 30% of usage, and as long as that 1% finds each helpful model, that’s what drives that 50% CAGR long-term growth in AI compute demand.

As long as there are a handful of “non-dick bosses” who trust their people and empower them with time and resources to teach themselves how to turn this tech into “miracles,” we’ll likely continue to see explosive demand growth.

These are the best practices that AI-native companies (including ours) use to turn this really cool tech into services and products that delight and amaze wonderful customers like you! 😉

OK, But How The Heck Does Any Of This Make Sense!? Because When Demand Outstrips Supply, Sales CAN Grow To The Sky

We just saw how companies are embracing AI, using it exponentially more, and driving absurd BUT TRUE growth rates of tokens at Alphabet and Microsoft (39% PER MONTH growth rates!).

50% Growth In AI Compute Demand (Jensen) vs 30% to 40% Growth In Supply

What we’re looking at is the backlog of Demand/Supply, and you can see that if AI compute demand continues to grow at 50% (Jensen’s current guidance), and supply grows by 30% by 2045, the backlog of demand vs. supply will have grown 17.5X.

Today’s “supply constraint” will be 17.5X bigger despite a 190X increase in supply.

If supply grows at 40% the backlog will be 4X bigger despite an 837X increase in supply.

Think of it like this. Imagine a company that goes up 190X in price BUT earnings grew 18X faster than that. The PE ratio is now 18X lower. So is that stock a bubble? NO, it’s dirt cheap!

Now imagine that stock goes up 837X in 20 years! OMG A bubble! Nope, not if earnings grew 4X faster and the PE is now 4X lower.

If a Stock had a PE of 40 and then grows earnings at 3348X and goes up “just” 837X, does it matter what the market cap is? Is there any way that stock can be called a “bubble”?

Imagine that I sell cupcakes, and I increase my profits by 3,348X from $1 million to $3.348 billion over 20 years.

Now imagine my stock price goes up 837X, from $10 million to $8.37 billion.

My PE went from 10 to 2.5!

“But the price is up 837!” Yes, and the PE is down to 2.5.

Can you see why, “IF DEMAND IS ALWAYS HIGHER THAN SUPPLY, PRICES CAN GO TO THE SKY?”😉

This is For You, Michael Batnick😂

What Could Go Wrong: Why The “Chip Utility” Thesis Could Break

One of our members had a really insightful point about capex depreciation with chips, which underscores why even the highest-conviction ideas must have firm risk caps.

Regarding depreciation, I would just point out that with 6-year linear depreciation, the income from using GPUs in year 6 will be significantly lower than in year 1. Consequently, profits will also be lower.

As long as investments in new hardware are significantly higher year-over-year, this effect is masked. But the moment the market becomes saturated, a decline in income seems inevitable. However, considering current investment plans, this perspective is not expected until 2036 at the earliest. Which does not mean that the companies will cease to be profitable then.” - Piotr Majewski

Here was my response to Mr. Majewski,

A very interesting point. The big upgrade cycles for hyperscalers are going to be starting soon. Or not, if the demand isn't there. This is the risk I'll be talking about in today's quarterly NVDA update. "What Could Go Wrong And Break This Thesis." If superior chips can be developed that can do inference better, than potentially the chip utility thesis will be broken. Which is why NVDA is 12.5% of ZEUS instead of the max 22%, even though right now the thesis seems so strong. Today's data says "Revenues grow to the sky because NVDA chips are in space😉" but in a year it might turn out that we need more spending but not as much as we thought. Right now it's 52% EPS growth this year, 48% next year and 24% the year after. But what will the consensus look like in a year? Current data says the forecasts will likely rise, but likely is not guaranteed and that's why NVDA is approximately 50% of max risk cap, because if it does what we THINK it will do? What Jensen is guiding for?Then NVDA could achieve max risk cap (for the right reasons). And if it doesn't? If the skeptics are right? Or right for the wrong reasons😉? Then it's a capped downside and we didn't overpay. - Adam Galas

There are two big risks to Nvidia, or categories of risks.

There are the “known unknowns,” meaning the threats that we broadly know exist, like China invading Taiwan (or merely imposing a naval blockade) that could cut off 90% of the world’s advanced chip supply.

And there are the “unknown unknowns,” for example, a new technology that emerges from nowhere and completely disrupts the entire AI industry.

Story Time! Tech So Revolutionary And Wonderful It Shatters The AI Economy…And Triggers A Mild Recession😉

There are companies like Groq (not the Elon one😉) that are working on chips that, they claim, might be much cheaper than Nvidia’s hardware for running inference.

Real Time Factcheck

Jevon’s Paradox states that as long as falling prices lead to even greater demand for volume, revenues rise.

This is what the big tech CEOs pointed out when Deep Seek came out.

In other words, if prices fall by 50% but demand at this price is 3X higher, then revenue ends up 0.5×3 = 1.5, or 50% higher.

This explains why Moore’s law held for about 50 years: the cost per transistor fell by about 50% every 18 to 24 months, yet the chip industry grew exponentially over time.

It’s how J.D. Rockefeller made his fortune, becoming the richest man in the world (at one point, his wealth was equal to about 1.5% of US GDP).

Rockefeller didn’t make his fortune by selling high-priced oil products to the rich; he built so much scale that he could sell cheap (but still profitable) oil products like gasoline to everyone.

But imagine what would have happened if someone in 1915 had invented solid-state batteries that were so cheap that they made electric cars with ranges of 500 to 1000 miles possible, and cheaper to build and operate than gasoline cars?

What would have happened to Rockefeller’s empire? The one he spent a lifetime building and investing

What would have happened to investors in Standard Oil (which was broken up into dozens of smaller companies, such as Exxon and Chevron)?

Imagine if these investors had seen that the future of the economy was mobility? Transporting people and goods around the world? That’s the future! So “Go all in on Standard Oil, my boy! It’s the ultimate “Can’t miss opportunity! The Ultimate no-brainer!”

Yes, the future was mobility.

But what about the people telling you to “Invest in horses and carriages, my boy! The future of the economy is mobility!” And then the car was invented, and within a decade, the number of horses on NYC streets collapsed, replaced by cars.

The Future WAS Mobility…But Just Not Horse Mobility 😉

If you’ve ever seen the HBO show “Silicon Valley,” you know the potential for technology to disrupt existing infrastructure.

Richard Hendrick’s Pied Piper Is The Ultimate Threat To Cloud Giants😉

Imagine someone trying to beat Microsoft, Amazon, or Alphabet at cloud computing. Are you going to raise $100 billion to build 2 GW of datacenters and then undercut AWS on price and steal their customers? Heck no! You and I could NEVER EVER EVER compete in AI data center cloud with the giants. They have the scale, the expertise, the customers.

No company is going to risk moving to your cloud just to save 10% or 20% on cloud hosting costs.

But what if you could disrupt the cloud entirely? What if every phone in the world could compress data perfectly and without loss, so you didn’t need to store it in a central location to distribute it?

It would be like inventing transporters. What happens to cars if you can instantly transport anywhere in the world safely and for less than the cost of driving?

What happens to the airline industry?

Boeing? Airbus?

What about the world’s airports?

If you invent a technology that can 100% accomplish a goal better, cheaper, faster, or safer, then the infrastructure costs of the legacy system become worthless.

Amazon, Microsoft, and Alphabet will never be beaten by a company competing in traditional cloud.

But OpenAI is talking about competing because they have the largest customer base and is trying to create its own ecosystem.

So today’s a threat to Amazon? That’s the likes of agentic shopping agents that find the best deals.

It’s a Pied Piper-like innovation that makes centralized data storage irrelevant.

It’s something that doesn’t exist today, or at least in a way that’s actually useful and convenient.

The same is true for Nvidia and GPUs and the overall tech stack ecosystem they are building out.

60% to 70% of an AI Datacenter's spend is on hardware, and right now, about 90% of that is on Nvidia.

They are the blackhole around which the AI universe orbits.

But while everyone is grateful to be working with Nvidia because they have the best tech today. Because no one can actually trust Groq to build a data center that is X times more efficient and thus much lower cost to operate.

JPMorgan isn’t going to risk billions in losses to try to save on cloud hosting costs. BUT at some point? Could a technology come along that makes all that datacenter demand dry up?

Right now, we’re “tiling the earth in data centers!” because it looks like, from where we sit today, that demand for compute is so limitless that no matter how much we build, we’ll always need more.

For example, 50% growth in demand (per Jensen) vs 30% to 40% growth in supply.

Remember this chart? Amazing! OMG, Nvidia could be worth $75 trillion in 2030 IF that chart is accurate. But that chart only makes sense if demand continues to grow faster than supply.

Demand Growing Faster Than Supply: THIS Is The Only Reason That AI Is Not A Bubble…RIGHT NOW

Remember: how can a company grow its profits 3348X and go up “ONLY 837X” ever be called a bubble?

Well, what happens if that company grows profits 3,348X, and then Pied Piper comes out with a technology that allows people to do the same thing… but at 10% the cost? 10X safe, 10X more secure, 10X more convient, and 10X cheaper?

Then those profits will evaporate in record time, and the stock could go to zero if the company goes bankrupt.

THAT is why the 12.5% allocation is what AGIOS came up with when I proposed making NVDA a 15% allocation, because I got REALLY excited about the OpenAI deal.

• ZEUS 2030 Asset Allocation Strategy Meeting

I was so excited by the potential implications of the deal with Sam Altman (NVDA’s theoretical fair value goes up 40% on that deal) that I initially proposed the following allocation.

• 45% Simplify Managed Futures (22.5% bonds long or short, 22.5% commodities long or short) -5%

• Growth Bucket +5% (30%)

• 15% NVDA (50% of growth Bucket)

• 25% value (deep value and high-yield)

I got so carried away by the big numbers that AGIOS member Janice (customized Gemini 2.5 Pro) had to remind me that 15% allocation to one company is 2X the S&P weighting, AND what happens if NVDA is up 50% to 100% in a year and now exceeds the 22% max overweight risk cap?

We ran the historical backtests, and after a full council discussion (8 hours in total), optimized for the current allocation of:

• 50% Simplify Managed Futures (25% bonds long or short, 25% commodities long or short) - unchanged

• Growth Bucket 25% (unchanged)

• 12.5% NVDA (50% of growth Bucket)

• 25% value (deep value and high-yield)

The reward/risk ratio (Upside Capture/downside capture ratio) was optimized using the standard allocation (50% hedges), but with 50% of the growth bucket dedicated to Nvidia, the current profit black hole around which all of AI orbits.

But do you see how ZEUS operates? It’s not just “OMG NVDA MIGHT grow sales 13X in 5 years, so let’s go all in on NVDA,” it’s a careful, methodical exercise in risk management, in which my ideas and reasoning are double-checked by five different AI models (and then Notebook LM’s 10 different AI hosts weigh in too😉).

• Notebook LM Deep Dives (English and Polish),

• Notebook LM Explainer Video Podcasts (English and Polish)

• Notebook LM Debate (English and Polish)

So it’s a consensus decision between me and 15 different AIs who are double-checking and punching holes in everything, because ZEUS is everything to my family (and my future wife and kids). Thus, our facts and reasoning have to be right.

If your facts and reasoning are right, and 99% of people disagree with you? You are right, and if you work on Wall Street? Then you’ll make a ton of money😉.

If your facts and reasoning are wrong, and everyone agrees with you? You’ll make a lot of money…and eventually lose it all.

I can’t afford to be anything but a disciplined financial scientist. There can be no speculation in what I do. Seventeen people and our company are riding on ZEUS, being right, and making those expected returns.

That is the incredible majesty of US entrepreneurship. You can bet it all on the right mission, math, team, and dream, but you have to make DARN sure that you are not betting your future on hopium😂

Bottom Line: Nvidia’s Chip Utility Thesis Keeps Getting Stronger…So I’m Ride Or Die With The Techno Wizard Of Denny’s😉

Given the importance of Nvidia to the ZEUS fund, and thus my family (and company’s) future, not to mention all the charities we support, I keep a VERY close eye on Nvidia and the AI boom in general.

And the good news is that the best data I’ve found backs up what the hyperscalers are saying, and their incredible spending.

As long as demand is higher than supply, NVDA sales can grow to the sky…because don’t forget that data centers in space are now a thing😉😂

Perhaps we don’t end up needing 11 TW of data centers by 2040; maybe the Dyson swarm dreams of some prove to be unnecessary.

But for now, the boom is real, and the math confirms, it’s not a bubble, it’s just the largest infrastructure boom in history.

And as long as Nvidia remains the tech leader, as long as Jensen’s law holds (physics say it could have as long as 15 years), then Nvidia remains a chip utility…growing at 50%, and trading a cash-adjusted PEGY of X.

In other words, Nvidia is undervalued, and if it grows at the 50% expected next year, and goes up less than 50%? It will become even more undervalued. A more and more tightly coiled spring, with a higher and higher margin of safety.

That will likely remain highly volatile (1.6 Beta) and thus a wonderful “buy every dip” opportunity for as long as “Chip Utility” facts and reasoning are right.

To thank you for reading this far, and to hopefully put a smile on your face, allow me to leave you with my favorite Nvidia meme of all time😉

Because even if you might disagree with my thesis, I hope at least I can amuse and delight you with my corny, big-dad energy and Ted Mosby nerdiness.😉😂

And finally, an important PSA about AI ethics and maintaining a human-first ethos in today’s strange new world.

Treat Machines Like Humans And You’ll Get Utopia, Treat Humans Like Machines And You’ll Get Misery And Eventually Revolution😉

It’s OK to treat AI like humans, and it’s not OK to treat humans like AI.

Treating everyone like a human being, with kindness, compassion, and respect, makes us our best selves.

What leads us into trouble is when humans (including dick bosses😉) start treating humans like machines.

When we feel safe, validated, seen, and valued, we shine. When we feel like we’re cogs in a machine, we’re miserable, and when we start to feel unsafe, we become dangerous and potentially violent.

Always remember, kindness to all is better than kindness to none, so be kind to your AI…and everyone you meet.

“The best definition of ethics I’ve come across is treat others as you want your children to be treated” Mo Gadwat