On Wednesday, November 26, 2025, MIT released a report (The Iceberg Index) which concludes that artificial intelligence (AI) can already do the work of 11.7% of the U.S. workforce. That adds up to about 17 million jobs and as much as $1.2 trillion in wages each year. The report is based on a huge simulation of the entire U.S. workforce.

According to the report,

“The Index measures technical exposure—where AI capabilities and human skills overlap—not displacement outcomes. For example, financial analysts will not disappear, but AI systems may demonstrate capability across significant portions of document-processing and routine analysis work. This reshapes how roles are structured and which skills remain in demand, without necessarily reducing headcount.”

“Without necessarily reducing headcount.” In other words, all 17 million jobs are not necessarily at risk today. Or are they?

In January, OpenAI announced the Stargate Project. It’s a joint venture between OpenAI, SoftBank, Oracle, and the investment firm MGX to build AI data centers. The first Stargate data center is under construction in Abilene, Texas, with one building already operational. In October, OpenAI announced five additional data center locations and over $400 billion in investment over the next three years. The company plans to invest $500 billion in AI infrastructure by 2029. This is an unprecedented investment. Spending $133 billion per year is equal to the entire federal budget of Ireland, a country with a population of 5 million people.

People are starting to recognize just the sheer scale that will be required

OpenAI Finance Chief Sarah Friar

What kind of returns does this level of investment require? Let’s assume 20% return on investment (ROI). Twenty percent compounded over seven years gives a total return of 3.5 times the invested amount. An initial investment of $0.5 trillion suggests $1.75 trillion in returns after seven years. Assuming the data centers are sold after seven years with an optimistic 50 percent appreciation ($0.75 trillion total) leaves $1 trillion in returns required in years four through ten, or about $150 billion per year. That’s an amount close to the GDP of New Mexico, whose population is about 2 million. One possible financial scenario is shown in the table below.

Year	Investment	Active Facilities	Revenue Needed to Service Capital	Exit Value This Year	Notes
2025	133	0	0	0	Construction begins
2026	133	1 cohort active	30	0	Construction continues
2027	133	2 cohorts active	70	0	Construction ends
2028	100	3 cohorts active	150	0	Operating period begins
2029	0	3 cohorts active	150	0	Operating income
2030	0	3 cohorts active	150	0	Operating income
2031	0	3 cohorts active	150	0	Operating income
2032	0	3 cohorts active	150	250	Sell Cohort 1 (Y1 build)
2033	0	2 cohorts active	100	250	Sell Cohort 2 (Y2 build)
2034	0	1 cohort active	50	250	Sell Cohort 3 (Y3 build)

How does OpenAI/Stargate intend to turn all that compute into roughly $150 billion in revenue per year? Let’s consider a few options: subscribers, advertising, porn, and “augmenting” human jobs.

According to OpenAI, ChatGPT has 700 million weekly active users as of September 2025. To make $100 billion per year from active users would require every user to pay $140 per year. For comparison, Wikipedia received an average donation of $10 from 8 million people which supported 25 billion monthly page views in FY23-24. Only a small fraction of users are willing to pay anything. ChatGPT has an estimated 20 million subscribers paying around $20 per month, or $400 million. OpenAI earned a total of $2.7 billion in 2024.

Google earns about $225 billion per year in ads from all sources, including YouTube, search, maps, chrome and Gmail. It seems unlikely that OpenAI’s business plan is to take over half of Google’s business over three years. Porn is a $100 billion per year industry so it could provide a fraction of the needed returns, but it can’t finance the entire $500 billion investment.

What about “augmenting” jobs? Here the likely business model is to sell raw compute capacity to companies building their own synthetic labor force, including OpenAI. As noted above, MIT’s Iceberg Index simulation suggests that AI tools are sufficiently advanced that $1.2 trillion per year in human jobs could be automated today. Could a law firm reshape “how roles are structured” for example, by firing its paralegals and replacing them with synthetic labor that costs significantly less? Is the market for synthetic labor sufficient to justify the investment? Yes, it is, but does Stargate have the scale to replace all those jobs?

The table below provides one estimate. The effective human workdays per day is reduced by a factor of one hundred (1% usable output) to account for variation in AI model success rates across task types and complexity. There is no “1.0 human” across all tasks either. One human workday is defined as 1,000 useful tokens of output, roughly one page of analysis or writing. Note that the Stargate estimate assumes 100 tokens per second, rather than the larger GPU numbers. This conservative assumption accounts for uneven utilization, queueing, memory constraints, and other real-world inefficiencies in large-scale compute clusters.

Entity	Tokens/second	Tokens/ day	Human-normalized tokens/day	Synthetic labor workdays/day
Human worker	0.01	1,000	1	1
H100 GPU	~50	4 million	~4,000	~40
GB200 GPU	>150	13 million	~13,000	~130
Vera Rubin GPU	>450	40 million	~40,000	~400
Stargate (7M GPUs)	>700M	70 trillion	~70 billion	~700 million

These are all very rough estimates, but the table suggests that Stargate will have the compute necessary to replace all 17 million jobs “where AI capabilities and human skills overlap,” as identified in the Iceberg Index paper.

In September, CNBC reported that “Over $2 trillion in AI infrastructure has been planned around the world.” The article went on to quote OpenAI Finance Chief Sarah Friar who said, “People are starting to recognize just the sheer scale that will be required…” “We’re just getting going here in Abilene, Texas, but you’ll see this all around the United States and beyond.” Friar didn’t say what the required scale is necessary for, but it sure looks like OpenAI/Stargate is building synthetic labor factories to replace tens of millions of human jobs over the next few years.