Can Nvidia’s margins last?

Unlock the Editor’s Digest for free

Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.

Nvidia has become an outlier in the chip industry. Its gross profit margin of 75 per cent is closer to that of a luxury brand or software company than a traditional chipmaker.

Even the industry’s top performers have historically operated with gross margins closer to 50 per cent. That gap has become the single most important variable underpinning Nvidia’s valuation. The question is whether it can last.

Nvidia is often described as a chip designer that derives much of its value from its proprietary software and developer platform. That is accurate but incomplete. Nvidia makes its money selling physical chips that must be manufactured in fabrication plants the company does not own.

Nvidia’s most advanced AI chips rely heavily on fabrication by Taiwan Semiconductor Manufacturing Company. Its most profitable products, including the H200, Blackwell and the next-generation Rubin architecture, are made on TSMC’s advanced 4 and 3 nanometre production processes. There is currently no alternative capable of manufacturing those designs at the same scale, performance and yield. 

In chipmaking, whoever controls advanced manufacturing has the greatest leverage. TSMC decides how much it charges for each wafer and how much advanced production capacity each customer receives. Nvidia is one of its largest clients. But it is competing for the same scarce manufacturing capacity as the world’s biggest tech groups, including Google and Microsoft, which are designing their own AI chips. Those chips tend to handle specialised workloads rather than fully replace Nvidia’s systems. Even so, they rely on the same advanced production lines at TSMC. The global rush to build AI infrastructure has left the company with no shortage of demand.

This division of roles has been enormously profitable for both Nvidia and TSMC. Nvidia designs the architecture and the software. TSMC builds the chips and spends more than $40bn each year expanding and upgrading fabrication capacity. Nvidia enjoys software-like margins without bearing the cost of constructing and upgrading factories. TSMC keeps its most advanced factories running at full capacity.

For now, that balance has held. Nvidia’s valuation, however, assumes it will continue to hold. Shares of Nvidia trade at around 21 times forward earnings. For a company that has grown revenue at triple-digit rates in recent years, that appears almost restrained. 

Yet it looks reasonable only because those forward earnings assume gross margins remain near historic highs. Nvidia’s valuation rests on two conditions holding simultaneously for years. First, demand for AI infrastructure must remain strong enough for Nvidia to sustain premium pricing. Second, TSMC must not claim a larger share of the profits embedded in each chip. The first condition receives constant attention. The second is discussed far less.

As Nvidia’s earnings base is now so large, even a small decline in margins would meaningfully affect how the company is valued. At current revenue levels, every one-point move in gross margin would represent about $2bn in annual gross profit, which would be enough to move earnings forecasts.

Geography adds another layer of risk. Advanced chipmaking remains concentrated in Taiwan, so any disruption would directly affect output. TSMC’s Arizona expansion provides some diversification but the plants remain under the same company and allocation decisions. The underlying dynamic is unchanged. Any disruption would strengthen TSMC’s leverage by making advanced capacity even scarcer.

Today, the chip industry revolves around two critical bottlenecks: ASML in advanced lithography equipment and TSMC in advanced chip manufacturing. Credible alternatives are rare and the barriers to entry are vast.

Nvidia’s advantage sits above manufacturing, in chip design and in the software ecosystem built around its processors. Its Cuda platform, the software developers use to programme Nvidia GPUs, underpins much of today’s AI development, making switching costly. 

Yet that is still not the same as controlling advanced manufacturing. A lead in chip design must be renewed with each product cycle. Ownership of the industry’s core manufacturing infrastructure is far harder to displace.

Nvidia has been earning software-like margins while avoiding the capital intensive burden of manufacturing its chips. TSMC has every incentive to support Nvidia’s growth. But it has no obligation to protect Nvidia’s margins. In this industry, power follows production capacity. Today, it is held by TSMC.

[email protected]