Nvidia's throat

Nvidia's Lifeline

Can Nvidia's terrifying 75% gross margin be sustained? June Yoon, a columnist at Financial Times, analyzes that the market always focuses on the demand side, but the real factor that determines Nvidia is being overlooked. What is this factor? Will it have long-term uncertain effects on Nvidia?

The author believes that if Nvidia's gross margin fluctuates by one percentage point, that equates to a $2 billion swing in annual gross profit. Such volatility is enough to reshape Nvidia's profit forecasts and also to affect the valuation logic of the entire market.

Although Nvidia currently has a forward P/E ratio of only 21x, which is not high among tech giants—actually, it's relatively restrained—to maintain a valuation of over 20x in the long run, two extremely harsh conditions must be met simultaneously:

The first condition is that demand for AI infrastructure must stay strong year after year.

The second condition is that TSMC will not raise prices, nor will it seek higher profits from foundry operations.

The author argues that the current situation is the market focuses almost entirely on the former condition, while few assess the potential risks of the latter.

Currently, Nvidia’s most profitable products include the H200, Blackwell series, and the next-generation Rubin architecture. All of these rely on TSMC’s exclusive advanced 4nm and 3nm process technology. This precise industrial division of labor allows Nvidia to avoid more than $40 billion in annual fab construction and upgrade expenses, but as a result, it also hands the company's lifeblood to a single partner.

The author believes that in semiconductor manufacturing, capacity scarcity equals pricing power. Globally, currently, there is no alternative manufacturer capable of handling Nvidia’s massive orders at the same scale, performance, and yield rate.

This means that not only does TSMC set the absolute price for every wafer, but it also controls the allocation of advanced manufacturing capacity. What's more critical is that Nvidia must compete with rivals such as Google, AMD, and Broadcom for TSMC's scarce capacity.

The author says that in such a complete seller's market, while TSMC may be motivated to support Nvidia, objectively, TSMC has no obligation to protect Nvidia's luxury-level gross margin.

It’s undeniable that Nvidia's CUDA platform has established a strong software moat due to high developer switching costs, but this moat is built on top of physical manufacturing, and the two are not comparable.

Additionally, chip design leadership requires constant maintenance or renewed proof in every technological iteration; meanwhile, the barriers to manufacturing are much stronger. Over these years, Broadcom, Google, and AMD have challenged Nvidia, but has any company shaken TSMC's position? No.

Now, the entire chip industry is held up by two irreplaceable physical bottlenecks: ASML’s monopoly on lithography equipment, and TSMC’s grip on advanced chip manufacturing.

Regarding market concerns over geopolitical risks, even if TSMC expands capacity in places like Arizona to hedge these risks, the right to allocate capacity still lies with the same company—the only change is where it's manufactured. And as long as the supply chain sees any material disruption, advanced capacity only becomes more scarce, further enhancing TSMC’s bargaining power.

Therefore, the author believes that, in the semiconductor industry, power always follows capacity, and currently, that decision-making power is firmly in TSMC’s hands.

Jason points out that actually, the market has never forgotten TSMC’s power. Looking back at the recent AI disruption trend—TSMC dropped too at first, but the market quickly realized that if the software layer were truly disrupted, it could actually benefit semiconductors, since all demand still depends on hardware, and thus on TSMC.

As for the gross margin issue, I remain cautious about the author’s assessment.

Michael Porter wrote in Competitive Strategy that a company wants to find a foothold in the market must have either a cost advantage, product differentiation, or focus on a unique segment.

For Nvidia, its high gross margin is not based on cost advantage but on differentiation, or more precisely on the pricing power that comes from technological dominance. Therefore, costs are not the core factor affecting Nvidia—it’s pricing.

As for TSMC, who doesn't know they hold the lifeblood of the whole semiconductor industry? But TSMC is not just a company; it's the core of Taiwan’s silicon shield. If TSMC were to significantly raise prices, it would not only face extra political pressure but also disrupt the global semiconductor supply chain balance, which goes against TSMC's philosophy of steady, long-term operations.

So essentially, TSMC acts more like a master coordinator of supply, striving to balance capacity allocation across major tech companies. Even though it holds a monopoly, it is unlikely to charge monopoly prices.

Therefore, I believe that at this stage, the pressure on Nvidia still mainly comes from the demand side: the upper limit of costs that Tech Giants can bear and the speed at which in-house chip development erodes market share.

Finally, a news supplement: today, there are reports that the US is drafting a bill to restrict AI chip exports—not only targeting China this time, but all countries worldwide.

My view: just take this news as a rumor. Chip manufacturing is not something one company or one country can decide alone. Without global supply support, even the US cannot accomplish it independently. Ultimately, even if the bill goes into effect, the main players in this market are unlikely to change.