NVIDIA closed its 2026 fiscal year with $215.9 billion in revenue (up 65% year-on-year), and with a record market capitalisation that briefly crossed $5 trillion in April 2026. NVIDIA is arguably one of the most successful companies in the history of the stock market, but beneath the headline statistics is an unusual risk profile for a company of this scale: layers of concentration risk which are not independent of each other, yet the market prices them as if they were.
Concentration risk matters because the more dependent a company is on a single customer, supplier, or geography, the fewer buffers it has between an external shock and a direct hit to earnings. For NVIDIA, that dependency runs across three dimensions simultaneously.
Customer Concentration
Firstly, NVIDIA's Q2 FY2026 10-Q filing reveals that just two unnamed direct customers accounted for 39% of total revenue in Q2 2026 and CFO Colette Kress confirmed on the Q4 FY2026 CFO commentary that the top five cloud providers and hyperscalers (large-scale cloud computing companies that operate data centres which power AI) accounted for just over 50% of Data Centre revenue. This suggests meaningful revenue concentration and a dependency on a handful of hyperscalers such as Alphabet, Amazon, Meta, and Microsoft. Whilst they may be NVIDIA’s largest customers, they are also a credible long-term threat, as they each actively develop their own chips to reduce their own dependency.
Fabrication Concentration
Secondly, every advanced chip NVIDIA designs is manufactured by TSMC, headquartered in Taiwan. TSMC holds an estimated 70% share of the total chip manufacturing market by revenue and with no near-term alternative, NVIDIA cannot dual-source its way out of this dependency. The result is that the revenue base of the world's most valuable semiconductor company depends entirely on a fab it cannot control, on an island it cannot protect.
Regulatory Concentration
The regulatory exposure is the only risk that has already been quantified in a single earnings report. In April 2025, the US government ruled that NVIDIA needed a licence to export its H20 chip to China and the sales halted immediately. In a single quarter, NVIDIA took a $4.5 billion charge for inventory it could no longer shift and orders it could no longer fulfil, and flagged a further $8 billion in lost revenue for the following quarter.
One government decision, issued overnight, erased the equivalent of an entire quarter's worth of China revenue. NVIDIA's CFO estimates China's data centre computing market will grow to nearly $50 billion, and with the company essentially locked out of the market, the government’s decision is a revenue ceiling with no current resolution.
Counterargument: The Bull Case for NVIDIA
The bull case for NVIDIA should not be dismissed for three main reasons. CUDA, NVIDIA's proprietary programming language, has been the foundation of AI development for over a decade, rewriting software pipelines and redeploying infrastructure to run on alternative silicon carries costs that make switching economically irrational for most customers in the near term. Blackwell demand reinforces this as it has consistently outpaced supply since launch, with NVIDIA guiding $78 billion in revenue for Q1 FY2027, reflecting order visibility that few businesses at this scale can claim. With approximately $60 billion in free cash flow generated in fiscal 2026, NVIDIA has the financial capacity to absorb shocks, such as regulatory ones, that would be fatal to smaller firms.
The bull case is compelling, but a strong moat does not make correlated risks disappear, it means the company is better placed to survive them than most. The question is whether the market has correctly priced the scenario where customer concentration, fabrication geography, and regulatory exposure deteriorate simultaneously and the answer to that is less obvious.
Why These Risks are Correlated
The risks, customer concentration, fabrication geography, and regulatory reach, all share the trigger of a deterioration in US-China-Taiwan relations. A standard model might assign a 10% probability of a Taiwan Strait crisis and discount it accordingly. But as this single event would have simultaneous consequences of disrupting NVIDIA’s only advanced fabrication plant, accelerating US restrictions on Chinese products, and causing hyperscaler customers to delay infrastructure projects due to the macroeconomic uncertainty, the realised downside is actually 100% of 3 risks at once, not 10% of one risk.
The risks are correlated, and hence standard portfolio risk frameworks (which would price risks independently) misprice the risk. NVIDIA’s 10-K acknowledges Taiwan dependency as a risk factor, yet the stock’s 12-month average P/E ratio was 40.3, compared to the S&P 500 average of roughly 21x, implying investors are not applying a large enough discount for uncertainty. If even one of these risks materialises, the others follow automatically, and a forward multiple that doesn't reflect that correlation will not survive the first quarter it becomes relevant.
Conclusion
The bull case for NVIDIA is real, with a dominant software ecosystem, high Blackwell demand and sufficient cash flow, but a strong moat and a mispriced tail risk are not mutually exclusive. NVIDIA's 12-month average P/E of 40.3x is more than double the S&P 500 average, implying the market is treating its growth as durable and its risks as manageable and independent. But the analysis here suggests otherwise, with customer concentration, fabrication geography, and regulatory reach requiring only one adverse event to detonate simultaneously. Until that correlation is reflected in the discount rate applied to NVIDIA's earnings, the forward multiple is not just optimistic but structurally incomplete.
