OpenAI eyes ambitious plan for new wafer fabs

Altman’s $7 trillion vision for AI chip production.

OpenAI’s image tool is now free for all users, though limits remain unclear. Meanwhile, the company has secured $40 billion in funding at a $300 billion valuation.

OpenAI, known for its AI models, appears to be exploring the semiconductor manufacturing sector, raising questions about the feasibility of building wafer fabrication plants. Reports recently surfaced about CEO Sam Altman’s discussions with executives from major chip manufacturers, including TSMC and Samsung, during his trip to Asia last year. Altman proposed an ambitious $7 trillion plan to construct 36 new wafer fabs and data centres, aiming to produce AI chips funded by the United Arab Emirates. He believes these facilities would support the burgeoning demand for AI capabilities.

The investment Altman suggested is staggering, amounting to a quarter of the annual output of the US economy. However, the timeline for establishing these fabs is lengthy, as it would take several years to meet OpenAI’s growing computing power requirements. TSMC, while approached for the project, found the proposal too aggressive and risky, noting that even a few additional wafer fabs would entail significant capital and risk.

Building a wafer fab is an enormous undertaking, often costing hundreds of billions of dollars due to various factors. The expenses stem from land acquisition, facility construction, equipment procurement, and ongoing operational costs. Advanced lithography machines and other essential equipment represent substantial financial commitments, while research, maintenance, and talent training add to the complexity. Current estimates place the cost of modern fabs in the billions; for instance, Intel’s factories in Arizona are expected to cost around $15 billion each, and Samsung’s Texas fab is projected at $25 billion.

Moreover, the cost of constructing a wafer fab varies by region. In Asia, established supply chains, available talent, and supportive policies contribute to lower costs, whereas building in Europe, the US, and the Middle East can be more expensive due to the necessity of importing technology and developing a comprehensive supply chain. Overall, OpenAI’s ambitions in the semiconductor space highlight the significant challenges and investments required to succeed in this critical industry.