| If the government wanted to pump USD$1T into the economy, is investment into a stack of sheds full of rapidly depreciating computers the most effective use of USD$1T? Some example contrasting options: - Worldwide investment in 300mm wafer fab equipment is projected to be USD$107-138B per year through to 2028.[1] USD$1T buys 100% of the global production of 300mm wafer fab equipment for about 7 years. - European Union countries are projected to spend approximately USD$250B on electricity generation and transmission infrastructure projects in 2025.[2] The US is projected to spend a similar amount.[3] China is projected to spend approximately USD$460B.[4] USD$1T buys 4 years worth of European Union or US expenditure on electricity generation and transmission infrastructure, or 2 years worth of Chinese expenditure. - Worldwide biopharmaceutical R&D was estimated to amount to USD$276B in 2021.[5] More conservative estimates include USD$102 in 2024.[6] Using the larger estimate, USD$1T buys 3.5 years of global biopharmaceutical R&D. [1] https://www.trendforce.com/news/2025/10/14/news-2nm-race-dri... [2] https://www.iea.org/reports/world-energy-investment-2025/eur... [3] https://www.iea.org/reports/world-energy-investment-2025/uni... [4] https://www.iea.org/reports/world-energy-investment-2025/chi... [5] doi:10.1038/d41573-024-00131-2 (https://www.analysisgroup.com/globalassets/insights/publishi...) [6] https://www.iqvia.com/blogs/2025/06/global-trends-in-r-and-d... |
First, the $1T number seems to be completely pulled out of thin air. There were recent speculations that OpenAI is working on an IPO at the level of $1T, but obviously this has nothing to do with US loan guarantees of $1T. The talk about US loan guarantees originated with the WSJ interview of the OpenAI CFO, Sarah Friar. She was very clear that she can't give any details, because nothing is concrete yet. She did not mention any numbers, at all. The $1T is pure speculation, based on nothing.
Second, Sarah Friar mentioned that the US Government is very receptive to the entire sector, and OpenAI is always consulted. But OpenAI is not lobbying the government for loan guarantees for themselves only. If there are any such loan guarantees (which is very likely), they'll be for AI investments in the US in general, so, for example, Anthropic and Google will also benefit (and probably Nvidia, AMD, Intel, Oracle, Amazon, Meta, etc).
Third, Sarah Friar used the word "backstop", instead of loan guarantees. This might sound like a synonym, but I don't think it is. She emphasized that the financing will be organized by the private industry; the government backstop is there to facilitate that. How could such a backstop look, so it is most efficient for both the taxpayers and for the investors (most of which are also taxpayers)? It could be some first loss tranches in some pools of loans, for example the first 10%. This way, the industry can raise $10BN, while the government is on the hook for only $1BN. Even more likely, the government will provide second loss backstop (the technical term is a "mezzanine tranch"). For example, they'll absorb the losses between 5% and 15% in a loan portfolio. The first loss will be absorbed by some speculative investors, who will be compensated by a very generous yield.
I also expect that new loans will only be raised if old loans perform well. Overall, I don't expect the US government to be on the hook for $1T at any given time, and I would be surprised if they'll be on the hook for much more than $1 BN at any given time.