Why TSMC Is Buying into Both OpenAI and Anthropic: Three Investment Principles from ASML and ARM to AI Model Companies

On November 10, 2026, TSMC's board of directors approved several financial, dividend and major investment items. It also approved subscribing for US$500 million each in OpenAI and Anthropic (for Anthropic, common stock at the initial public offering price; for OpenAI, preferred stock; both amounts are the ceilings the board approved). In addition, the board approved setting up an investment company of no more than US$500 million, similar in nature to what is generally called corporate venture capital (CVC), to make strategic investments in the semiconductor industry and its related ecosystem.
Anthropic's initial public offering was scheduled to be priced on November 18, 2026 and to list on the Nasdaq in the United States on November 19 under the ticker ANT, with an estimated price range of US$70 to US$78 per share.
In addition, Kevin Zhang, TSMC's Senior Vice President and Deputy Co-Chief Operating Officer, said in an interview with Nikkei Asia on November 13, 2026 that TSMC set up an emerging business fund of about US$200 million back in 2021 and that its returns have been excellent. The new fund has been expanded to US$500 million, with the goal of nurturing startups that could become the next Nvidia or Broadcom.
He said the venture fund will be used to support chip startups and to widen the scope of investment. The new fund may invest in suppliers and other upstream partners, with the aim of extending its investments beyond direct customers. Zhang added that, if necessary, TSMC does not rule out further expanding the size of the fund.
The key points of Zhang's interview made me notice that TSMC's investment strategy is now running on two tracks. Besides investing in trillion-dollar-scale major customers such as OpenAI and Anthropic, TSMC is also setting up a CVC that invests in the semiconductor industry and its related ecosystem, more like Intel Capital or Qualcomm Ventures, whose targets are smaller companies with potential and influence in the industry. Here I would like to share my thoughts.
TSMC's Competition with Samsung Extends from Chip Design Companies to AI Model Companies
Let me start with why TSMC is subscribing for US$500 million each in OpenAI and Anthropic.
To put the conclusion first: these two companies have long been end users of TSMC's chips. But in 2026 Anthropic confirmed that it will develop its own inference chips, and foreign media reported that it was in talks with Samsung about manufacturing them. TSMC's announcement of an investment at this moment is meant not only to strengthen the relationship but also to compete for the next major customer.
Look more closely at how OpenAI and Anthropic became TSMC's indirect customers. OpenAI developed its own AI chip with Broadcom. The code name that circulated earlier was "Titan"; when the chip was officially unveiled on June 24, 2026, it was named "Jalapeño." It uses TSMC's 3-nanometer process and was expected to enter volume production in the second half of 2026 and be deployed by the end of the year. OpenAI has said its second-generation chip is already in development.
However, under the custom chip model, the company that places wafer orders with TSMC is usually Broadcom, which provides the design services, not OpenAI. So OpenAI is more like a "customer's customer" of TSMC than a direct customer on TSMC's books. As for the Nvidia GPUs that OpenAI uses in large numbers, they are also all made by TSMC.
As for Anthropic, the computing power it currently uses comes from three kinds of chips, all made by TSMC: Google's TPUs, designed by Broadcom and MediaTek; AWS's Trainium; and Nvidia's GPUs.
However, in August 2026 Anthropic confirmed that it is building its own chip design team to develop custom chips specifically for inference, designed together with its future language models. According to a June 2026 report by The Information, Anthropic was in talks with Samsung about having Samsung manufacture the chip. Anthropic did not confirm this, nor did it announce a design partner or a launch date.
That is why the possibility of Anthropic placing orders with Samsung makes TSMC's investment more interesting. If the Samsung report is true, TSMC's investment in Anthropic is not only about binding an existing customer's customer more closely; it also aims to win a future customer.
The production partner for Anthropic's in-house chip has not been decided. By becoming a shareholder, TSMC holds one more card in this race. It can be said that the competition between TSMC and Samsung has extended from "winning chip design companies" to "winning AI model companies."
TSMC Invested in ASML for EUV and Exited Once the Goal Was Met
Having covered OpenAI and Anthropic, if we look back, TSMC has made quite a few strategic investments. Besides buying stakes to win customer orders, as with OpenAI and Anthropic, another purpose has been to help partners in the industry's upstream and downstream ecosystem and supply chain, and this help is in fact also helping TSMC itself. The most representative cases are three companies: ASML, IMS and ARM.
Take ASML. In August 2012, ASML launched its "Customer Co-Investment Program," inviting major customers to put up money to speed up research and development of EUV and 450-millimeter (18-inch) wafer technology. TSMC invested about 838 million euros for about a 5% stake in ASML and committed another 276 million euros over five years to research and development. Intel and Samsung also took part, acquiring stakes of about 15% and 3%, respectively.
TSMC's purpose in investing in ASML was clearly technology, not holding shares. As soon as the lock-up on its shares expired on May 1, 2015, TSMC began selling, and by October 8, 2015 it had disposed of the entire 5% stake, receiving about US$1.771 billion. For the full year 2015 it recognized a disposal gain of about NT$22.07 billion.
In hindsight, however, ASML's share price later rose several times over. Samsung held its shares until recent years before substantially reducing its stake and made more money than TSMC, so many people say TSMC "sold too early."
Yet this case also shows that TSMC's purpose in investing in ASML was to make sure EUV was developed successfully on schedule and that TSMC could get the equipment first, not to profit from the share price. Looking back, the goal TSMC set for this investment was achieved: EUV later became the key to TSMC's leadership in processes of 7 nanometers and below.
The second representative case is IMS. Most people are less familiar with this company. IMS makes multi-beam mask writers, key equipment for EUV photomasks and indispensable for EUV photomasks at 2 nanometers and below. It is also a key company in TSMC's supply chain.
In September 2023, TSMC bought about a 10% stake in IMS Nanofabrication from Intel for about US$433 million. So far I have not found any announcement or report of TSMC selling its IMS shares. Since there is no record of a sale, TSMC should still hold IMS. Moreover, IMS is not a listed company and has no public share price, so the book value of this investment cannot be estimated at present.
TSMC invested in IMS mainly for its multi-beam mask writers, key equipment that matters greatly to TSMC's core foundry business. From a strategic point of view, it would also make sense for TSMC to hold the stake for the long term.
Finally, there is ARM, the world's best-known semiconductor intellectual property (IP) company. ARM's IP is used by many smartphone customers, making it the most important technology partner of TSMC's core customers. So when ARM listed in September 2023, TSMC subscribed for nearly US$100 million of shares, a stake of about 0.2%. But in February 2024 and April 2026, TSMC sold all of its ARM shares in two rounds, with a cumulative gain of about US$230 million, or about NT$7.3 billion.
Of these two sales, at the time of the first, TSMC said the purpose was to "recover its cost." It sold 850,000 shares, getting back just about the US$100 million it had originally invested. For the second sale, which cleared out the position, TSMC said it was "an adjustment of investment targets based on its financial investment planning strategy."
The market read this explanation in three ways. The first is that ARM is shifting from IP supplier to making its own chips and could compete directly with TSMC's customers, so selling out lets TSMC stay neutral. The second is that TSMC may want to recover funds for expanding 2-nanometer, A16 and CoWoS capacity. The third is that it was simply taking profits.
In my view, TSMC's sale of ARM is most closely related to the first reading. When ARM is no longer just an IP company and launches products of its own, it will compete directly with TSMC customers such as Intel, AMD and Qualcomm. If TSMC kept its shares, there would be concerns about a conflict of interest with its customers, so TSMC chose to sell.
In the same way, this explains why TSMC is investing in both OpenAI and Anthropic this time, and in the same amount: to stay neutral among customers that compete with each other.
Moreover, in summing up these investment cases, I think the most important spirit is that TSMC's investment strategy always stays strictly focused on its core business. The purpose of every investment is strategic, to strengthen the core foundry business. This principle is very important, and I believe it is a good reference for many companies when they make CVC investments.
For a company like TSMC with a huge core business, investing is never about maximizing profit the way a venture capital (VC) firm does; it is about strengthening its own core business. So TSMC did not invest in ASML to make a lot of money; once the purpose was achieved, it sold the shares and put the money back into its foundry business.
Investments in the Core Business Are Held for the Long Term, and Sales Are Only Small Portions
At this point, I can also mention another very unusual TSMC investment: SMIC in mainland China.
In November 2009, SMIC settled with TSMC over the theft of trade secrets, paying US$200 million and giving TSMC about 8% of its shares free of charge. TSMC then sold the shares over time. In June 2019 it sold another portion, leaving only about 0.2%, and in July of the same year it sold another batch, bringing its stake down to about 0.03%.
SMIC is a competitor. Putting money into a competitor will never earn more than investing in yourself, so SMIC was never a company TSMC wanted to invest in. TSMC was forced to accept the shares unexpectedly because of the lawsuit, and for TSMC the right thing to do with such an investment was likewise to sell early and put the money back into its own core business.
So when it comes to investments related to the core business, TSMC also has its own affiliated companies. These are long-term holdings that form part of TSMC's own industrial map. TSMC usually does not sell these stakes, and even when it does, it sells only a very small portion.
These companies include GUC (design services), Vanguard International Semiconductor (specialty processes), VisEra (color filters for image sensors) and XinTec (packaging); SSMC in Singapore (a joint venture with NXP); and overseas joint-venture fabs such as JASM in Kumamoto, Japan (a joint venture with Sony, Denso and Toyota) and ESMC in Germany (a joint venture with Bosch, Infineon and NXP).
Going a step further, I believe the US$500 million investment company TSMC is setting up this time will follow the same three principles in its investment strategy. But according to Kevin Zhang, its targets may be earlier-stage, newer startups.
I make this judgment based on a very simple idea. TSMC has now reached the position of global leader and has to deal with technological breakthroughs and innovation in every part of the semiconductor industry, from 3DFabric, CoWoS advanced packaging, silicon photonics and the ecosystem materials and equipment for new types of memory, to startups in specialty chemicals, electronic gases, special photoresists and test equipment. Innovation and breakthroughs in the supply chain are a very important part of this.
What is even more central for TSMC is finding future product customers. It needs to invest very early in the next IC design customers like Nvidia and AMD, but the kinds of future customers are also changing greatly: companies such as Apple, the major cloud service providers (CSPs) and AI model developers, as well as physical AI such as autonomous driving, robots and outer-space applications. With so much to push forward, TSMC's CVC will certainly have more to do.
TSMC's Three Investment Principles Offer Lessons for Many Companies
Of course, TSMC's past investments have not been without failures. In 2009, TSMC invested in LED and solar. That investment had its own background and circumstances: it was a new assignment that Morris Chang gave Rick Tsai when he replaced him as TSMC's chief executive.
By 2015, however, TSMC announced its exit from LED and solar. Of the companies it set up itself, TSMC Solar and TSMC Solid State Lighting, the lighting business was eventually sold to Epistar for NT$825 million and the solar plant was closed. Its outside investment in Motech was turned into a financial investment, and the shares were gradually sold off. When it sold the lighting business, TSMC recognized a loss of about NT$740 million; in the quarter the solar plant stopped production, TSMC recognized a loss of about NT$2.8 billion.
Finally, I would like to sum up. From many cases, I have drawn three important investment principles at TSMC. I believe these principles and practices are something many companies can learn from, especially in today's new era of AI, when so many opportunities are tempting but also come with violent swings. They offer important lessons both for corporate CVC strategy and for ordinary investors.
First, investment is for the core business, not for making money. ASML and IMS secure key equipment and technology; ARM, Anthropic and OpenAI keep TSMC close to its most important customers and the direction of technology. Financial returns are only a by-product.
Second, TSMC buys only minority stakes and does not seek control. ASML 5%, IMS 10%, ARM about 0.2%: in each case TSMC is a small shareholder, because its role is only to help the other party, and through that process it also helps its own core business. It does not aim to run the other company's operations, nor does it want to maximize its own profit. This is consistent with the principle of "not competing with customers": a stake can strengthen the relationship without interfering in management.
So if we connect this principle to the investments in Anthropic and OpenAI this time, the logic is very similar to ASML in 2012. It is not about being bullish on the share price, but about tying TSMC to its most important customers at the most critical moment for the AI industry. This also echoes the view I have been discussing in earlier articles: TSMC does not bet on any single customer winning, but it makes sure it stands beside every important customer.
Third, once the purpose is achieved, a conflict of interest arises, or diversification fails, TSMC chooses to exit and focus on its core business. Once the purpose is achieved, it does not hang on, and it takes the money back for better use. That is why TSMC sold all of its ASML shares as soon as the lock-up expired and sold out of ARM a little over two years after investing. TSMC handles these financial holdings with considerable discipline. LED and solar were a rare failure for TSMC, but since exiting in 2015, TSMC has seldom ventured into businesses outside foundry.
I believe the path TSMC has taken with these investments offers good lessons for many companies and individuals. Especially in today's highly volatile stock market, investors can also learn from it: first have a core business of your own, then think clearly about which gains you should go after and which you should not. Otherwise, with such a fine world-class company here in Taiwan, it would be a great pity not to learn something from it.
Related Articles

Samsung's HBM Packaging Expansion Will Raise Demand for TSMC's CoWoS: How Taiwan and Korea Divide the Work in Advanced Packaging

Come Up with a Way! How Former TSMC CTO Chenming Hu Extended the Life of Moore's Law and Showed TSMC How to Overtake Intel

Farewell to Jensen Huang: The Best Storyteller in AI Could Not Write His Own Ending. What Comes Next for Nvidia, the AI Industry and Taiwan's Supply Chain
