What can be done with 7 trillion dollars? There was brief speculation about the possibilities when, in February, the Wall Street Journal wrote about Sam Altman’s 7 trillion plan to reshape the global semiconductor industry. He wants to raise money from sovereign wealth funds and venture capitalists to end the global shortage of GPUs required to train AI models and solve Open AI's growth constraints. Although everyone on X has moved onto fierce debates about the new current thing, I’m still stuck on the same question.

The Wall Street Journal stated that Altman's ambitious plan is more expensive than the combined market capitalization of Apple and Microsoft (which comes to around $6 trillion.)  An account on Reddit posits that, with that amount, you could "pay everyone's taxes ($4.44T), repay student debt ($1.74T), pay for California's African American reparations ($800B), and keep $2B." The overall bemusement at this number is best captured in the Axios newsletter that went out a day after the WSJ piece: "Such numbers are preposterous. The fact that they're being talked about with anything approaching a straight face is indicative of the degree to which the broader AI discourse has become unmoored from reality."

As someone who never has had to interface with anything more than 5-digit amounts in my day-to-day life (which perhaps is also true for the Axios newsletter writer), before I got to fantasizing about what I might do with $7 trillion dollars, I simply wanted to understand: What does $7 trillion dollars even mean? Is it a sum of money that you can make sense of? Or is it just a hyperobject to be dangled in front of investors who want to get in on OpenAI? (That is, an entity “of such vast temporal and spatial dimensions that [it] defeat[s] traditional ideas about what a thing is in the first place.”)

The best place to start trying to make sense of this number is precedents. Most of the discourse around the $7 trillion plan fails to mention the one other place where words like "trillion" and "plan" are thrown around with abandon — government bills and mega projects. There are two contemporary analogies in statecraft to Altman's semiconductor infrastructure fever dream: The American Infrastructure Bill of 2021 and China's Belt and Road initiative.

The former, otherwise known as the Infrastructure Investment and Jobs Act (IIJA), became law in 2021. The bill, in the mouthful words of the White House, "will rebuild America's roads, bridges, and rails, expand access to clean drinking water, ensure every American has access to high-speed internet, tackle the climate crisis, advance environmental justice, and invest in communities that have too often been left behind." The bill was passed with bipartisan support and its total spending is estimated to be close to $1 trillion. This is roughly 5% of the GDP of the United States ($23T) and 25% of the country’s tax revenue ($4.44T), which is the primary source of revenue for the Federal Government.

China’s Belt and Road initiative, on the other hand, is more fluid in scope and timeline. The initiative, aimed to increase Chinese influence in Asia by building large infrastructure programs across the continent, has spent $1 trillion since it was announced ten years ago with much fanfare. The strict terms issued by Chinese banks that have driven several countries into debt has led analysts to label the initiative “troubled” and a “debt trap.”

Perhaps Altman's plan rhymes more with large government mandates such as the Infrastructure Bill or Belt and Road initiative, where small amounts are directed towards several hundred projects across a period of 5-10 years, rather than a mega corp along the lines of Apple or Microsoft. But, while the Infrastructure bill is worth 25% of America’s tax revenue, Altman’s project would be worth 15 times the current revenue of the global semiconductor industry ($527 billion) —  and 7x the projected revenue of the semiconductor industry in 2030 ($1 trillion).

Sam Altman may very well be able to execute the biggest moonshot in history. After all, a trillion-dollar company was unheard of only a while ago, and now there are a handful of them. If successful, it would take the template of a supercharged special economic zone and rival the largest economies in terms of capital and GDP. An example of what this might look like can be found in the special economic zones that were set up in China in the 1980s. These five zones (Shenzhen, Xiamen, Shangtou, Zhuhai, and Hainan) set up under Deng Xiaoping were essentially market economies that had different rules and regulations than the rest of the country. They were forays into a more “open door” economic policy and provided the template for Chinese success in manufacturing in the last three decades. A similar experiment, aimed at creating a surplus of chip manufacturing, could lead to new business and economic models that operate outside a nation-state apparatus.

In her book Extrastatecraft, Keller Easterling investigates how such “zones” that span continents function. In a paragraph echoing Marc Andreessen's "Software Eats the World" she writes, "In Notre-Dame de Paris, Victor Hugo famously observed that ‘architecture [like that of the cathedral] was developed in proportion with human thought; it became a giant with a thousand heads and a thousand arms, and fixed all this floating symbolism in an eternal, visible, palpable form.’"

The novel goes on to say that the printing press usurped architecture as the vessel of cultural imagination and stole its supernatural power. Hugo prophesied: "This will kill that. The book will kill the edifice." Just like software ate the world and the book killed architecture, Altman's multi-trillion dollar, globally-distributed Silicon zone will make AI compute infinite and abundant — dealing the final blow to edifices of human thought and finishing what the printing press started.

Infrastructure projects have a long lifetime value. For example, I found a $14M bridge that is being reconstructed over the San Antonio River as part of the aforementioned Infrastructure bill. This project is on a freight route connected to Port of Laredo, the second biggest trade port of entry in the country, ahead of Port of Los Angeles. Based on some rough calculations, a 35-minute detour around the bridge would cause a loss of $30M per year for freight trucking operations. Unless there are any exponential shifts in how freight is transported, the bridge would end up saving several hundred million dollars for freight trucking operations in just a few years. Semiconductor projects on the other hand move in 4-5 year planning cycles, meaning that projects with large capital expenditure could become obsolete quickly as the technology changes. This is a challenge that Sam Altman’s plan will have to reckon with.

Intel learned about the unique challenges of semiconductor manufacturing the hard way. Back in the mid-2000s, when Steve Jobs approached Intel for a chip to put on the iPhone, Intel turned him down, and Steve Jobs went with Qualcomm and ARM. Intel rode on accumulated momentum for a few years until competitors started beating them to the next generation of chips. Now, Intel is just one of the several legacy companies relying on government regulation and investment, such as the CHIPS Act, to maintain their competitive edge.

People who have been deep in semiconductor development understand these challenges. Jim Keller, CEO of Tenstorrent and co-author of the x-86-64 instruction set, quote tweeted Sam Altman with "I can do it for $1 trillion." According to Keller and Nvidia CEO Jensen Huang, rather than building more processors, the industry should be focused on making the existing processors more sophisticated.

"Start by eliminating the margin stacking," writes Keller, referring to the cost or profit margin added by each participant in the supply chain to deliver a product to the end user. "There are [2–3] layers. Then, make chips way faster so the hardware matches the software. That is harder but doable."

Sam Altman is quite shrewd when it comes to funding models, most famously evidenced by OpenAI being a non-profit that has a valuation of $86 billion. (Though it should be noted that OpenAI has raised roughly 100 times as much money from investors as it ever did from donors.). Stratechery's Ben Thompson says the original motivation for OpenAI being a non-profit was to attract talented developers to a mission-oriented company rather than a for-profit one. You could argue that the $7 trillion plan also has an obscure long-term motivation — one intended to wrest control of the market away from semiconductor manufacturers and towards companies like OpenAI. The challenges and capital involved here, though, are on a scale that has never been attempted, and in an industry where vast amounts of capital expenditure are necessary — but not enough — for success.