Jensen Huang: The rise of TSMC has never been achieved by working behind closed doors.

In the hardcore geek circles of Silicon Valley, Lex Fridman's talk show has always been a magical existence. It can make tech titans who usually only wear leather jackets let their guards down and talk about extremely grand, even somewhat philosophical topics.

Recently, Jensen Huang, CEO of NVIDIA, had a conversation with Lex Fridman in his car. During this dialogue, Jensen Huang not only talked about the endgame of AI, but also casually dissected what is arguably the greatest "wealth creation myth" in the history of 20th century technology: how did Taiwan region build its current unshakable dominant position in the chip world?

At a time when semiconductors have become the focus of global games, people are used to attributing TSMC's success to the "engineer dividend", "crazy overtime" or "obsessive dedication to single-point technology". But from Jensen Huang's perspective, the underlying logic of all this is actually an extreme game of interlocking "open strategy" and "borrowing chickens to lay eggs" that lasted for half a century.

Breakthroughs in core technologies never come behind closed doors through sheer hard work. The course of history is often driven by those who know how to dance on the shoulders of giants.

Philips' "Century Marriage" and Kwoh-Ting Li's Foresight

Jensen Huang pointed out an extremely obscure but crucial historical origin in the conversation.

TSMC, which is now at the height of its power, and ASML, the Dutch overlord of lithography machines, can both be traced back to Philips of the Netherlands at their root. Founded in 1984, ASML is the product of a joint venture between Philips and Dutch semiconductor equipment manufacturer ASMI, and the first product it planned to commercialize was exactly the PAS 2000 wafer stepper that Philips had developed for many years. The birth of TSMC is also inseparable from Philips.

Let's go back to the mid-to-late 1980s. At that time, Taiwan region was in the stage of transforming from OEM manufacturing to technology-intensive industries. Kwoh-Ting Li clearly realized that to gain a foothold in the future world, the most core underlying technologies must be introduced. In 1986, Morris Chang was invited by Kwoh-Ting Li to return to Taiwan, and then he invited Philips and other enterprises to take shares under the framework of government funding. In February 1987, TSMC was officially established.

The shareholding structure of TSMC when it was founded in 1987 speaks for this path: the "National Development Fund of the Executive Yuan" of Taiwan region at that time held 48.3% of the shares, Philips held 27.5% as the second largest shareholder, and private shares totaled 24.2%. Philips brought not only capital, but also in-depth participation in technology, patent licensing and the executive team. TSMC's starting point was not created out of thin air by several geniuses in the laboratory, but a classic case of completing the upgrading of the industrial system under the leadership of the government, exchanging market access and industrial policies for technology spillovers, and deeply binding global giants.

IBM's "Open Standards" and Taiwan's Strategy of "Borrowing Chickens to Lay Eggs"

If the introduction of Philips laid the foundation, what really made Taiwan region's technology industry chain explode was a gap opened by IBM in the early 1980s.

After IBM launched the PC in 1981, it made public a large amount of technical materials in addition to the BIOS in 1982, forming the "open standard" of PCs, which enabled standardized components from different manufacturers to be interchangeable. The key distinction here is: what IBM did was not "open source" in the software sense, but the disclosure of hardware architecture documents, and the BIOS was not released in the form of open source code. But it is precisely this "reserved openness" that posted the map of the compatible machine ecosystem on the wall.

This is the huge opportunity that Taiwan's engineers have been waiting for. Due to the lack of ability to build global brands and operating systems, Taiwan's tech practitioners did not make whole machines to compete head-on with Dell and HP, but got into the gaps of the industrial chain: designing chipsets, manufacturing motherboards, and doing OEM. They handed over the designed chips to the fledgling TSMC and UMC for production, and secured a position on the base of the PC industrial chain with manufacturing efficiency and cost control.

The young Michael Dell back then had an extremely simple and crude entrepreneurial model: purchasing motherboards and components in bulk from Taiwan, shipping them back to the United States for assembly and sales. The structural fact behind this is valid: Taiwan's open standard dividend was not obtained by "developing everything on its own", but by embedding in the global division of labor and maximizing manufacturing efficiency and cost control.

Three-stage rocket for half a century: Openness, Alliance and Positioning

After listening to Jensen Huang's analysis, the rise of Taiwan region's chip industry chain has a relatively clear path:

The first stage is to leverage the technology spillover of giants. Through the policy layout during Kwoh-Ting Li's period and the in-depth participation of Philips, it obtained the admission ticket for semiconductor manufacturing. The second stage is to embrace open industry standards. With IBM's open architecture, it entered the PC components and OEM ecosystem to complete capital accumulation and iteration of engineering capabilities. The third stage is to go all in on the OEM model. The "pure OEM" model promoted by Morris Chang is essentially an extreme openness: it does not build its own brand, does not compete with customers, and only serves as the underlying infrastructure for the whole industry.

This path proves that in the field of hard technology, knowing how to "connect" and "embed" into the global network may be more critical than simply pursuing "full-chain independence".

The Great Computing Power Game: Who is Replicating the "Magic Operation" of the Past?

The pointer of history has come to the AI era. The whole world is falling into an extremely fierce fight for computing power. What actions of today's giants carry the kind of foresight of "borrowing chickens to lay eggs" and "open alliance" back then, and what prices have they paid respectively?

The alliance between Microsoft and OpenAI is still the entry to understand "borrowing chickens to lay eggs", but its evolution itself is a more brutal lesson. Microsoft brings computing power, capital and distribution channels, and uses the "chicken" of large models to revitalize its cloud and search business; OpenAI in turn completes technological expansion. By 2026, this relationship has moved from "deep binding" to "redefining boundaries": Microsoft still holds about 27% of OpenAI's shares, and OpenAI has promised to purchase 250 billion US dollars of Azure services, but Microsoft no longer enjoys exclusive authorization, and OpenAI can turn to other cloud vendors. This shows that "alliance and symbiosis" is never a static marriage, but a set of contracts that are constantly renegotiated as the ecological niche changes.

Meta's Llama trajectory provides another footnote. The open weights of Llama was once a textbook-style "IBM moment", which gathered a huge number of developers with a very low threshold, and the number of downloads once reached about 1.2 billion. But by 2026, Meta has basically stopped the subsequent development of Llama and turned to the closed-source model Muse Spark, no longer opening weight downloads. This does not mean that "openness has failed", but shows that open source is not the end point, but a tool to fight for ecological niche in a specific competitive stage. When "lowering the threshold" no longer serves the business focus, and when the marginal cost of maintaining open source leadership exceeds the temptation of closed-source monetization, the strategic shift will happen. The more accurate lesson is: the vitality of openness depends on whether it can continuously create irreplaceable niche value for the initiator; once this condition disappears, "openness" will be re-priced.

In contrast, the one that most stably holds the niche of "open infrastructure" ecosystem is exactly NVIDIA itself. Jensen Huang knows very well that NVIDIA's deepest moat is not the GPU chip, but the huge and open software ecosystem of CUDA. As of 2025, the NVIDIA Developer Program has gathered about 6 million developers, the CUDA ecosystem covers more than 400 libraries and 3700 GPU-accelerated applications, and the cumulative number of downloads has exceeded 53 million. Developers use NVIDIA's computing power ecosystem to incubate applications, and NVIDIA earns profits from the underlying infrastructure. This open platform strategy is the "pure OEM" model in the AI era: it does not compete for the victory of applications in specific times and places, but becomes the layer that no one can bypass.

The "hard infrastructure" on the hardware side is also strengthening this pattern. TSMC's CoWoS advanced packaging is the essential process for current high-performance AI chips. NVIDIA has booked 800,000 to 850,000 pieces of CoWoS wafer capacity of TSMC in 2026, taking up more than half of the annual quota exclusively. TSMC and SK Hynix are ushering in a new round of expansion due to the skyrocketing demand for hardware brought by AI computing power, and they are still the underlying nodes that those giant enterprises cannot bypass.

Conclusion: The Cruelty and Romance of Technological History

History will not repeat itself simply, but it always rhymes.

Michael Dell, who was looking for opportunities in the Taiwan supply chain back then, could not have predicted that forty years later, the computing power network built based on these manufacturing capabilities would drive the greatest AI revolution of mankind. Kwoh-Ting Li, who introduced Philips back then, could not have fully predicted that this island would become a giant that holds the throat of the global semiconductor industry.

In this era when computing power is national strength, instead of blindly pursuing the closed-door manufacturing of "the whole industrial chain", it is better to understand the strategy pointed out by Jensen Huang: open up the ecosystem, develop through alliance and symbiosis, find the right ecological niche and stick to it. But the shift of Llama also reminds us that this strategy requires something calmer than "believing in openness": a clear calculation of one's own niche at each stage. Technological breakthrough is never an isolated island, it is a network. Whoever can become the most efficient and irreplaceable node in this network can grasp the pricing power of the next era - and "openness" or "closed source" is just a pair of replaceable shoes on the way to that node.

This article is from the WeChat official account "Tech Business", author: Tech Business, published with authorization from 36Kr.