TSMC’s board of directors approved an investment plan of no more than $1 billion for two of the top model companies.
Specific split: up to US$500 million to subscribe for Anthropic common shares at the IPO issue price; up to US$500 million to invest in OpenAI’s preferred shares.
$1 billion is not a lot of money for TSMC. At the same board meeting, it also approved $40.93 billion in capital expenditures for advanced processes, advanced packaging, factory construction, and research and development through 2027.
But the signal significance of this action is greater than the amount. TSMC's previous role was very simple: it built chips for whoever designed them, without taking sides or touching downstream companies. Now it is directly a shareholder of the model company.
This is nothing new in the AI industry. NVIDIA invested in OpenAI, and OpenAI bought NVIDIA's cards; Google and Amazon invested in Anthropic, and Anthropic rented their clouds. Investors, suppliers, and customers circulate back and forth in the same relationship network.
The problem is here too. When the supplier is also a shareholder, it becomes increasingly difficult to tell how much of the order is a real demand and how much is "supported by one's own people". Analysts are already complaining about this.
For Anthropic, having TSMC enter the market to subscribe during the IPO is at least a good endorsement. For TSMC, in addition to selling shovels, it also wants to share some of the profits from the gold mine itself.