The race to build artificial intelligence infrastructure has produced some of the largest construction and equipment budgets in modern business history. Data centers, advanced chips and new power capacity are being financed at a pace usually associated with national infrastructure programs. A recent analysis that converts that spending into jobs, however, suggests the bill may be far harder to pay than the industry’s optimistic forecasts imply.
According to the analysis published at PostQuantum, the companies building AI infrastructure may need roughly $2.5 trillion a year in AI revenue to justify the expanded build-out. If AI companies collect about a dime for every dollar of value their systems create, and if one full-time job is valued at roughly $100,000 a year in pay and benefits, then generating that revenue would require the equivalent of 250 million jobs. The United States, by comparison, has about 159 million nonfarm payroll jobs in total.
How the 250 Million Jobs Figure Is Calculated
The calculation rests on a few explicit assumptions. The author estimates that about $1.5 trillion has already been spent worldwide on chips, data centers and power since ChatGPT launched in late 2022, with spending continuing at roughly $800 billion to $1 trillion a year. Because computer hardware has a short useful life while buildings, power and cooling systems last much longer, and because investors expect a return on their money, the infrastructure needs to generate close to 50 cents of revenue each year for every dollar invested.
Applied to forecasts of about $5.2 trillion in AI data center spending by 2030, that rule points to an annual revenue requirement of around $2.5 trillion. Current AI revenue is far below that level. The analysis puts revenue from AI products and services at roughly $150 billion to $250 billion a year at recent run rates, meaning revenue would need to grow many times over simply to support the current pace of construction.
The jobs conversion is deliberately simple. Employers in the United States pay workers more than $16 trillion a year in wages and benefits, and the average cost of a full-time job works out near $100,000. A job-equivalent therefore does not mean a worker is literally replaced. It can also mean existing staff become productive enough to create an extra $100,000 of value. What matters is how that value is divided between the customer and the AI provider.
Why the Split of Value Matters So Much
If AI companies somehow kept all of the value their tools created, the hurdle on what has already been built would equal about 7 million job-equivalents. At a 50-50 split, the figure rises to about 14 million, more than the total number of manufacturing jobs in the United States. At a 10 percent share, the scenario for today’s building pace reaches the headline figure of 250 million job-equivalents, or about 157 percent of all US payroll jobs.
Other analysts have reached similarly demanding conclusions using different methods. Valuation professor Aswath Damodaran has described the build-out as the largest factory in history, constructed before anyone knows exactly what it will produce. Sequoia Capital has repeatedly highlighted the gap between infrastructure spending and actual AI revenue, while Bain & Company has estimated that the industry would need about $2 trillion in annual revenue by 2030. JPMorgan has calculated that earning a modest return on the planned investment could require hundreds of billions of dollars in revenue every year, indefinitely.
Lessons From Earlier Infrastructure Booms
History offers a cautionary parallel. During the fiber-optic boom of the late 1990s, companies spent heavily laying cable that the internet genuinely needed. The technology succeeded, but many builders did not: Global Crossing, one of the most prominent, went bankrupt in 2002. Customers and later owners captured much of the value, not the firms that paid for the original construction.
AI may follow a different path, and the technology is already changing how organizations operate, a theme explored in our coverage of maintaining a smooth long-term experience with new digital tools. Prices for AI capability have also fallen sharply, which benefits users but makes it harder for providers to charge premium rates. A service that becomes a cheap commodity can transform the economy while still delivering disappointing returns to the companies that financed it.
What Would Have to Go Right
For the math to work, several things would need to happen at once. AI revenue would have to grow far faster than it has so far, providers would need to keep a meaningful share of the productivity gains they create, and demand would need to extend well beyond the United States, since the American payroll alone is not large enough at realistic revenue shares. Supporters argue that entirely new products, scientific breakthroughs and efficiency gains across every sector could eventually justify the spending.
Skeptics answer that displacement estimates remain modest so far, and that automating too much paid work would also shrink the customer base that companies sell to. Either way, the jobs arithmetic reframes the debate. The question is no longer whether AI is useful. It is whether it can generate value on a scale larger than the entire American paycheck, quickly enough to pay for the infrastructure now rising to support it.
