Looking at NVIDIA now through Charlie Munger's latticework of mental models

If you only look at NVIDIA from a one-dimensional perspective, the conclusion is very simple: explosive performance, a strong moat, and the leading player in the great AI era.
But if you look at it through Charlie Munger's multiple mental models, the truly interesting thing about NVIDIA is not "whether it's a good company"—that's already quite clear—but rather: what kind of company is it, how long can this strength last, and where will risks emerge.

First, look at the most superficial "numerical dimension." As of April 22, 2026, NVDA's stock price is approximately $199.88, with a market cap of about $4.89 trillion and a static P/E ratio of around 49 times. In the last fiscal year, FY2026, NVIDIA's annual revenue was $215.9 billion, a year-on-year increase of 65%; of this, data center revenue accounted for the absolute majority at $193.7 billion. Looking at these numbers alone, NVIDIA is certainly still one of the world's strongest AI beneficiaries. ()

But Munger would remind you not to just look at the numbers, but to look at the structure behind them.
Today's NVIDIA is no longer a company that simply "sells GPUs." What it repeatedly emphasizes in its annual report is a full-stack platform capability encompassing chips, systems, network interconnects, CUDA, CUDA-X, various SDKs, and industry software. In other words, what NVIDIA really wants to sell is not just computing power, but the infrastructure standard for "AI factories." This change is crucial because it determines whether the company will make money from hardware or from the platform in the future.

This leads to the second dimension: the moat is not a single point of performance, but a system-level lock-in.
Many people still habitually understand NVIDIA as "strong in chips." That's certainly not wrong, but it's no longer enough. Looking deeper, its strength lies in binding together chips, interconnects, software ecosystems, developer habits, cloud vendor adaptations, and customer workflows. The annual report even states that over half of its engineers are working on software. At GTC 2026, NVIDIA further defined Dynamo as the "distributed operating system" for AI factories and introduced the Vera Rubin roadmap, continuing to push itself toward being an "AI infrastructure platform." This means that what will truly determine NVIDIA's profit margins and valuation center in the future may not just be how fast a particular generation of chips is, but how deeply the entire ecosystem depends on it.

The third dimension is incentives and counter-incentives.
Munger rarely stops at "is the company strong"; he always asks further: will others' incentives end up weakening it? For NVIDIA, this question is very real. The stronger it gets, the more its customers depend on it; but the more customers depend on it, the more they will try to reduce that dependence. The FY2026 annual report shows that one direct customer accounted for 22% of NVIDIA's total revenue, and another for 14%, indicating a high concentration. For cloud giants, continuing to purchase from NVIDIA is the short-term optimal solution, but in the long run, they will inevitably push for self-developed chips, custom ASICs, and platform diversification. This is also why NVIDIA, while reaping the benefits of AI capital expenditure, is also striving to make NVLink, the software layer, and the system layer more open and platform-like—it needs to guard not just against competition, but against customers' impulse to "de-depend."

The fourth dimension is second-order consequences.
This is also Munger's favorite question: the first-order logic is fine, and then what?
NVIDIA's first-order logic is simple: AI demand explodes, and it's the biggest shovel seller. But second-order logic immediately follows: will customer capital expenditure slow down? Will value be redistributed during the inference phase? Will self-developed chips erode its pricing power? Will geopolitics change market boundaries? Among these issues, the most realistic is export controls. In April 2025, the U.S. government required licenses for H20 exports to China, leading NVIDIA to record a $4.5 billion charge in FY2026 Q1 due to H20 inventory and purchase commitments; the company has also explicitly stated that under current rules, it is difficult to provide a competitive product for the Chinese data center market that can both gain U.S. approval and be accepted by China. In other words, what NVIDIA now faces is no longer just commercial competition, but a combination of commercial, technological, and geopolitical risks.

The fifth dimension is market psychology and expectation management.
Munger has always emphasized that a stock is not just a business, but a container for expectations. NVIDIA's biggest problem today is not that people underestimate it, but precisely that people are too quick to equate a "great company" with a "great stock." But the reality is, no matter how strong a company is, as long as market expectations are raised high enough, its stock price will still experience volatility. NVIDIA's current valuation is no longer priced as an ordinary semiconductor cyclical stock, but as an "AI-era infrastructure platform." Therefore, the truly important question is no longer "is NVIDIA's performance still good," but "can it remain strong enough to justify such expectations."

So, from the perspective of Munger's multiple mental models, NVIDIA can be summarized roughly as follows:

It is not a stock that relies solely on good-looking financial reports, but a company that intertwines technology, ecosystem, incentives, geopolitics, capital expenditure cycles, and market sentiment.

This is also why looking at NVIDIA cannot just focus on P/E, revenue growth, or next-quarter guidance.
What should really be watched are these things:

First, can it continue to solidify its "hardware advantage" into a "platform advantage"?
Second, as customers become larger and stronger, will they increasingly want to break free from it?
Third, as AI moves from training to inference and application, will value distribution change?
Fourth, will export controls and global supply chain restructuring cause it to lose some previously profitable markets?
Fifth, under such high expectations, will any growth slowdown be amplified by the market into a valuation repricing?

From this perspective, NVIDIA is still a top-tier company, there's little doubt about that; but it is also no longer in the stage where "you can easily make money just by getting the direction right."
The real thing to watch going forward is not "whether AI will develop," but: can NVIDIA continue to upgrade itself from the "strongest shovel seller" to the "hardest-to-bypass infrastructure rule-maker in the AI world"?

This is probably the core to watch when looking at NVIDIA through Munger's method.