Analysts believe that NVIDIA's 75% high gross margin supports its valuation, but the core risk lies in its heavy reliance on Taiwan Semiconductor's advanced process technology. Taiwan Semiconductor holds scarce production capacity and pricing power, representing an irreplaceable manufacturing bottleneck. If Taiwan Semiconductor were to demand a reallocation of profits, every 1% fluctuation in NVIDIA’s profit margin would impact $2 billion in profits, posing a potential threat to its valuation.
NVIDIA's ultra-high profit margins are not invincible — the most easily overlooked risk in the market lies precisely with its closest partners.
NVIDIA’s current gross margin stands at 75%, far exceeding the historical average of about 50% for traditional chip manufacturers, and more closely resembling the profitability structure of luxury brands or software companies. According to June Yoon, a columnist for the Financial Times, this profit margin is the most critical variable supporting NVIDIA's current valuation, and its sustainability largely depends on whether Taiwan Semiconductor will claim a larger share of profit distribution.
June Yoon pointed out that while the market pays significant attention to whether NVIDIA's AI demand can be sustained, there is notably insufficient discussion on the rising bargaining power of Taiwan Semiconductor. Based on the current revenue scale, every one-percentage-point change in NVIDIA’s gross margin translates to approximately $2 billion in annual gross profit, which could significantly impact earnings forecasts and, in turn, influence market valuation.
NVIDIA’s stock is currently trading at a forward price-to-earnings ratio of about 21x. June Yoon believes that while this valuation appears restrained, it hinges on two conditions: sustained strong demand for AI infrastructure and Taiwan Semiconductor not seeking to expand its profit share per chip — both must hold true simultaneously and endure for several years.
NVIDIA’s Profit Model: The Dividend of Design and Manufacturing Division
NVIDIA is often described as a chip design company that extracts value from its proprietary software and developer platforms. However, June Yoon argues that this description is incomplete. NVIDIA’s revenue comes from selling physical chips, which must be manufactured in foundries that NVIDIA does not own.
NVIDIA’s most profitable products — including the H200, Blackwell series, and the next-generation Rubin architecture — are all produced using Taiwan Semiconductor’s advanced 4nm and 3nm processes. Currently, no alternative manufacturer can produce these designs at the same scale, performance level, and yield rate.
This division of labor has created substantial returns for both parties: NVIDIA focuses on chip architecture and software ecosystems without bearing the annual capital expenditure of over $40 billion required for foundry construction and upgrades; Taiwan Semiconductor maintains full utilization of its most advanced production lines. June Yoon describes this relationship as symbiotic — but also as NVIDIA’s core vulnerability.
Taiwan Semiconductor’s True Leverage: Scarcity of Capacity Equals Bargaining Power
In the semiconductor industry, whoever controls advanced manufacturing holds the greatest sway. Taiwan Semiconductor determines the pricing of each wafer and allocates advanced capacity among its customers. NVIDIA is one of Taiwan Semiconductor’s largest clients, but it competes with global tech giants like Google and Microsoft for the same scarce capacity — these companies are also designing their own AI chips and rely on Taiwan Semiconductor’s cutting-edge production lines.
The global boom in AI infrastructure construction has left Taiwan Semiconductor struggling to meet demand. June Yoon pointed out that under this dynamic, Taiwan Semiconductor has every incentive to support NVIDIA’s growth but no obligation to protect NVIDIA’s profit margins.
Geopolitical factors further exacerbate this risk. Any disruption in the supply of advanced chip manufacturing will directly impact output. Taiwan Semiconductor's expansion plans in Arizona provide some degree of geographic diversification, but the related factories still belong to the same company, and control over capacity allocation remains unchanged. June Yoon believes that any disruption would only reinforce Taiwan Semiconductor’s bargaining position due to the heightened scarcity of advanced production capacity.
Boundaries of the CUDA Moat: Fundamental Differences Between Software Advantages and Control Over Manufacturing
NVIDIA’s competitive advantage lies not only in chip design but also in the software ecosystem built around its processors. The CUDA platform is a foundational tool for developers writing programs for NVIDIA GPUs and is deeply embedded in the core processes of today’s AI development, making migration costs prohibitively high.
However, June Yoon emphasized that this is not the same as controlling advanced manufacturing. Leadership in chip design must be continuously re-proven with each product cycle; whereas control over core manufacturing infrastructure is far more difficult to challenge.
She also noted that the entire semiconductor industry currently revolves around two key bottlenecks: ASML’s control over advanced lithography equipment and Taiwan Semiconductor’s dominance in advanced chip manufacturing. Both lack credible alternatives, with extremely high barriers to entry.
Two Prerequisites for Valuation, But the Market Focuses Only on One
June Yoon’s core assessment is that NVIDIA’s ability to sustain software-level profit margins depends on a rare industry structure — one in which Taiwan Semiconductor consistently chooses to cooperate as a partner rather than as a profit maximizer.
NVIDIA’s profit scale is already substantial; even a moderate decline in gross margin would significantly impact the company’s valuation. While the market remains highly vigilant about the sustainability of AI demand, there has been little discussion about whether or when Taiwan Semiconductor might reprice this partnership.
In her view, power in this industry follows production capacity. And today, that capacity is in the hands of Taiwan Semiconductor.
Editor/Lambor
