OpenAI and other hyperscalers, the companies pouring money into large-scale computing power, are part of an AI infrastructure buildout whose data centre spending is expected to reach nearly $1.1 trillion by 2027.
Industry calculations show they will need an extraordinary rise in productivity to justify that outlay by 2030.
The analysis comes from Jessica Wachter, who positioned the question not by guessing at model adoption but by asking how fast earnings must grow to cover the planned spending.
The hunt for data
In a separate move tied to the same data hunger, the OpenAI Foundation said it would fund a proposal by the policy analyst Ruxandra Teslo to buy science assets from bankrupt biotech firms.
And it has assembled what it called "high-quality scientific datasets", drawing on regulatory filings, manufacturing plans and safety data that would otherwise be lost.
Together the two threads underline the pressure on labs to secure both computing power and exclusive, high-value training data as they scale; the maths Wachter outlines explains why investors and executives are racing on both fronts.
Investment or gamble?
How quickly hyperscalers turn that capacity into profitable, productivity-boosting services will determine whether the current buildout proves an investment or a gamble.
