The bulls think Nvidia is a mythical creature. The bears think it’s Cisco Systems 2.0. 

I’ve owned ~190 Nvidia shares (cost basis ~$119) since 2024, but I’m not riding the Nvidia unicorn amidst bubble fearswith enough hubris to think I’ll survive unscathed in five years. Both bulls and bears have legitimate arguments and concerns. 

But overall, Nvidia is far better positioned than Cisco ever was, even though bubble risks are real. It’s the closest parallel we have: both companies sold essential infrastructure during generational technology buildouts, both dominated their markets, and both faced questions about valuation and durability.

So let’s compare them as we assess the bull and bear cases for Nvidia today. I’ll also use my legal background to analyze export controls — a regulatory dynamic that creates both opportunity and risk, and gives Nvidia advantages Cisco never had. Then we’ll examine what valuation Nvidia needs to achieve to justify its current price, and I’ll share my decision framework for what I’m doing with my position. 

Take a run with the Nvidia bulls

Like Warren Buffett, I’m a fan of competitive moats. Or as Peter Thiel advises in Zero To One, “Monopoly is the condition of every successful business.” Whether that’s great for society and competition is a separate question, but when I’m investing, I’m looking for a massive advantage. Nvidia has it. 

(And no, I don’t care that Peter Thiel recently sold it or that Michael Burry continues to predict the next bubble bursting as he’s done nearly every year since he was right in 2008).

You might think I’m going to praise Nvidia’s GPU design and quality. Or their first mover status to adapt it to AI, just as Cisco did with fiber for the internet. 

No. From my perspective, Nvidia’s greatest competitive advantage is CUDA, the software ecosystem that surrounds Nvidia’s hardware offerings. 

According to Nvidia’s latest earnings reports, “over 4.5 million developers now use CUDA, up from 1.8 million in 2020.” That means more community support, more shared code, and more third-party tools and applications. 

You know what Cisco didn’t have in 2000? Applications. They called it “dark fiber: for a reason — nobody was using the infrastructure they installed. 

Nvidia has the opposite problem. Customers can’t get enough of what Nvidia is selling and developers are actually using not only the powerful GPUs, but the CUDA ecosystem that surrounds them. This makes CUDA not only powerful, but incredibly sticky. Developers and researchers who use it will rarely look elsewhere — why lose access to the world’s most mature and optimized GPU platform?

Cisco wasn’t selling software back in the 2000s. And Nvidia’s closest competitors today (AMD, Intel, etc.) have struggled to build similarly mature software ecosystems around their hardware offerings. 

Nvidia’s true moat is owning the full AI stack: the GPUs themselves, the CUDA software layer, the networking infrastructure (InfiniBand/Ethernet), and now the orchestration tools. An enterprise customer can buy Nvidia and have everything they need. AMD and Intel, by contrast, sell components but rely on third-party or open-source tools to complete the developer experience.

But none of this matters if enterprise customers can’t generate revenue from it, right? According to a McKinsey survey, nearly nine out of ten respondents said their organizations are regularly using AI, although the pace of progress is uneven. Most organizations are still navigating the transition from experimentation to scaled deployment, so yes, ROI is uncertain. But here’s what matters: companies are putting real money behind AI — $37 billion in 2025, up from $11.5 billion in 2024. 

That’s 3x growth. And it’s not all R&D. At least 10 AI products are generating over $1 billion in ARR and 50 AI products are generating over $100 million in ARR (led by the model APIs powering applications – Anthropic, OpenAI, and Google). These returns are increasingly distributed across departmental solutions in coding, sales, customer support, HR, and verticals like healthcare and legal. 

Enterprise adoption has been slow, but the pace of investment and the rate of returns have increased significantly year-over-year. As smaller enterprises and consumers continue to experiment and understand AI’s potential and use cases, adoption will only continue to grow.

One of my favorite recent use cases is the collaboration between Nvidia and the industrial giant, Caterpillar (which I also have a position in). Nvidia is integrating physical AI into Caterpillar’s heavy machinery and manufacturing, using hardware like Jetson Thor and software to bring real-time intelligence, autonomy, and digital twins to construction, mining, and factories. This is a great example of real world AI use cases that didn’t exist when Cisco was laying dark fiber back in the 1990s. 

One guy who has been with Nvidia since the 1990s is its founder and CEO, Jensen Huang. I love companies with strong founders and Jensen fits that mold. An added bonus is longevity and quality of the C-suite around him — great management is crucial for a great company and Nvidia has it. 

But we cannot only assess Nvidia and our AI future with rose-colored glasses. Especially when you need to detect any bears approaching. 

Take shelter with the Nvidia bears

Here’s why those bears might be right. Nvidia’s current valuation of 50x P/E prices assumes perfection. Seriously. 

They assume Nvidia maintains its competitive moat. That enterprise adoption skyrockets. That existing customers keep spending exorbitantly. That geopolitics and export controls don’t undermine its incredible business (more on that next). 

If just one of these factors backfire, Nvidia is not the same company at 50x P/E prices. So current levels are already pricing in significant growth. 

But it’s still nowhere near the Cisco 200x P/E prices. That was insane, but Nvidia could still be overvalued. 

Most bears will point to the peak of hyperscaler capital expenditure if AI return on investment fails to materialize. If you read through the most recent earnings for some of these firms like Amazon, Alphabet, Meta, and Microsoft, they all say the same things: 

• they generally expect capital expenditures to increase, particularly in support of AI products and services; 
• they rely on a “limited group of suppliers for semiconductor products, including products related to artificial intelligence infrastructure such as graphics processing units” (in other words, they’re dependent on Nvidia with few alternatives, from Amazon’s 10-Q, September 2025); and 
• their progress on AI models and products position them to capitalize on new revenue opportunities in the years to come.

It’s unclear what their current revenues are related directly to AI. They don’t report that granularly. But Microsoft has amended its mission statement to highlight AI: 

“Microsoft is a technology company committed to making digital technology and artificial intelligence (“AI”) available broadly and doing so responsibly, with a mission to empower every person and every organization on the planet to achieve more. We create platforms and tools, powered by AI, that deliver innovative solutions that meet the evolving needs of our customers.”

If that’s not a commitment to spend exorbitantly into the near future on AI, I don’t know what is. 

Bears also point to custom chip designs as a major threat. Google’s TPU, Amazon’s Trainium, Microsoft’s Maia — all designed to reduce dependence on Nvidia. If any of these succeed at scale, Nvidia’s pricing power and market share could erode quickly. The hyperscalers have the talent, resources, and motivation to attempt to build alternatives.

But even if these firms could create their own custom chips to rival Nvidia, they would have to contend with the sticky CUDA community and address switching costs. Ask mobile device manufacturers how successful they were in trying to do this with Apple. 

One thing Apple really annoys everyone with is a new phone every year. And oftentimes new USB cords and power cables. Absurd! 

That’s exactly how some of Nvidia’s customers and bears feel about Nvidia’s chips, most of which only have a 5-year design life at best. Bears often argue that at least Cisco’s fiber could stay in the ground for many years. In Nvidia’s case, those expensive GPUs need to be swapped out regularly. 

This recurring revenue is obviously a positive for Nvidia, a negative for its customers, and an opportunity for a competitor. The issue? No competitor has materialized — or appears even close to materializing — to pressure Nvidia’s margins or product design life. With that said, Nvidia is improving in this department with some of its older designs like the H100 still in production 5 years later. 

But even if none of these bear cases materialize, the bears aren’t done. They regularly point to history. How infrastructure buildouts almost always overshoot and then have to correct themselves. I even cited some of these examples in a previous essay, from Britain’s Industrial Revolution to America’s railroad and automobile booms. 

Yes, many upstarts failed. Hundreds of regional rail companies collapsed. America used to have 88 car manufacturers in the 1920s. Only a handful like GM and Ford survived. 

Is Nvidia the GM or Ford of the AI revolution? Anyone giving you a definitive answer to that question isn’t being intellectually honest. 

But there is one more additional area that bears regularly cite as a risk that I want to acknowledge but also make the case that it’s a secret regulatory moat — export controls. 

Why export controls are a secret regulatory moat

When I was a young lawyer fresh out of law school I was thrown onto a commodities trading floor and told, “Good luck.” I was basically the cop trying to police traders, salespeople, schedulers, and other operations staff who had many more years of experience and moxie. 

It’s where I first learned the term, “regulatory arbitrage.” Where you could potentially benefit from starting business in one jurisdiction and ending it in another. Oftentimes, it was just code for “here’s how I’m going to try to avoid this regulatory requirement.” 

So my answer when pressed was usually a simple, “No.” 

But there is a big exception here. When your home government makes a deal with you to bypass certain regulations provided you give them 25% of your revenues from that flow, then not only do you have a regulatory opportunity, you have a regulatory moat. 

The U.S. government did just that with Nvidia and export controls. They gave them an opening to sell certain chips (H200s) in China provided the Commerce Department gets a 25% cut. I’ve criticized this form of state capitalism, but hey, for the next 3 years at least these are the rules of the game unless Congress rediscovers its powers. 

So for now, not only has the U.S. government granted Nvidia an exception. They’ve crowned them as a winner. They have skin in the game to support Nvidia selling in China. That’s a geopolitical risk turned into a competitive advantage. 

The bears would be right to highlight that China can also block sales of chips into its country. And they’ve done just that. But while Chinese firms like Huawei are making progress on domestic chip development, they remain several generations behind Nvidia’s cutting-edge designs. For the advanced AI workloads that matter most — training frontier models and running large-scale inference — there’s no Chinese alternative that comes close. 

Yet. 

Export controls will remain a significant area of uncertainty for Nvidia as the firm finds itself caught in the middle of an escalating geopolitical battle. China and the U.S. could easily change their policies overnight. But one truth remains — both China and the U.S. need Nvidia to compete in AI. At least for the foreseeable future. So if anything, I’m leaning more towards this serving as a regulatory moat and an advantage for Nvidia with uncertainty and risk clearly present. 

Being real on Nvidia’s valuation

What does Nvidia need to earn to justify its current price? The answer is more extraordinary earnings because at current prices, Nvidia can’t afford to have a down quarter. 

Nvidia’s current revenue is ~$130B and its current earnings are ~$75B. To justify current share prices, Nvidia needs to reach ~$198B earnings by 2028, roughly 2.6x their current earnings. This assumes a 40x P/E. 

Here are my projected scenarios: 

Bull Case (25%):

• AI market grows to $600B by 2028
• Nvidia maintains 70% share = $420B revenue
• Margins stay at 60% = $252B gross profit, ~$200B net income
• Stock deserves 40x = $8T market cap (stock doubles from here)

Base Case (50%):

• AI market grows to $400B by 2028
• Nvidia maintains 60% share = $240B revenue
• Margins compress to 55% = $132B gross profit, ~$100B net income
• Stock deserves 35x = $3.5T market cap (stock down ~15% from here)

Bear Case (25%):

• AI market grows to $250B by 2028 (slower than expected)
• Nvidia share falls to 50% = $125B revenue
• Margins compress to 50% = $62.5B gross profit, ~$45B net income
• Stock deserves 30x = $1.35T market cap (stock down 65% from here)

The base case still demands a lot from Nvidia, but given current market dynamics, I think it’s the most probable outcome by 2028.

My Nvidia decision and future framework

For now, I’m holding my position. I’ll write more should that change, but I’m confident in the base case, even if it demands near perfection. 

Here’s what I’m watching closely:

• Hyperscaler capital expenditure growth rate (if it decelerates, buckle up) 
• Nvidia’s gross margins (if they compress below 58%, competition is having an effect) 
• Market share in data center GPUs (if drops below 80%, watch out) 
• China revenue trends (big source of revenue and growth potential for Nvidia so continued restrictions by China could be a big problem) 
• AMD and custom chip progress (custom chips are particularly hard and we don’t see current meaningful production workloads running on alternatives — switching costs are real and CUDA’s ecosystem will compound over time)

Should any two or more of these metrics deteriorate for Nvidia, I will likely sell. I may also do the same if Nvidia’s stock rises above $200/share at current levels (that’s projecting a $10 trillion market cap, which is absurd). 

And should we ever have a Liberation Day event where Nvidia’s price drops significantly (maybe around $90/80 or lower), I’m buying as much as I can (absent significantly different circumstances). 

And who knows? I might be like Buffett and find another cigar butt worth puffing on that’s a better opportunity than Nvidia. In that case, I’m out. 

But one comparison I’m definitely out on is the one between Nvidia and Cisco. This isn’t 1999. The dynamics are very global and very different. But with that said, many of the risks are still the same. 

Which is why if you’re going to run with the bulls, always listen to the bears.