AI
“Harris Kupperman, founder of hedge fund Praetorian Capital, wrote in a 2025 essay that because data center components break down so quickly, the AI industry would need to generate about $1 trillion in revenue across 2025 and 2026 just to cover the costs of hyperscale data center buildouts.”
Plain restatementIn an essay published in 2025, Harris Kupperman of Praetorian Capital estimated that roughly $1 trillion of revenue would be required for the AI industry to break even on the data center capital spending of 2025 and 2026, with the short useful life of data center equipment as the driver of that estimate.
This one largely checks out. Harris Kupperman, founder of Praetorian Capital, did publish an essay on his firm's website on October 5, 2025 in which he wrote that adding 2025 and 2026 together, the AI industry would need roughly $1 trillion in revenue just to break even on data center spending, and many trillions more to earn a real return. He reached that number by shortening how long he assumed AI data centers stay useful, which raised his earlier estimate of $160 billion. Two details in the post are loose. His argument is that the equipment and buildings become obsolete fast as new chips and designs arrive, not that components physically break down, and his $1 trillion appears to be an annual revenue level needed to cover two years of construction rather than total revenue earned during 2025 and 2026. It is also one investor's estimate built on stated assumptions about spending, asset life and profit margins, which he described as directional rather than precise. The separate Bain figures in the same post, $6 trillion of annual AI revenue needed by 2031 and a $4.2 trillion shortfall, match what Bain published in its annual technology report in late September 2026.
Harris Kupperman, [drifted from the evidence:] founder of [drifted from the evidence:] hedge fund Praetorian Capital, [drifted from the evidence:] wrote in a 2025 essay that [drifted from the evidence:] because data center components break down so quickly, the AI industry [drifted from the evidence:] would need to [drifted from the evidence:] generate about $1 trillion in revenue across 2025 and 2026 [drifted from the evidence:] just to cover the [drifted from the evidence:] costs of [drifted from the evidence:] hyperscale data center [drifted from the evidence:] buildouts.
[added by the neutral restatement:] In an essay published in 2025, Harris Kupperman of Praetorian Capital [added by the neutral restatement:] estimated that [added by the neutral restatement:] roughly $1 trillion of revenue would be required for the AI industry to [added by the neutral restatement:] break even on the data center capital spending of 2025 and 2026, [added by the neutral restatement:] with the [added by the neutral restatement:] short useful life of data center [added by the neutral restatement:] equipment as the driver of that estimate.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- The essay exists, is correctly attributed, and is correctly dated to 2025. "An AI Addendum" was published on the Praetorian Capital site on October 5, 2025, building on an August 20, 2025 essay by the same author.
- Harris Kupperman is the founder of Praetorian Capital, as the claim states.
- The $1 trillion figure is real and appears nearly verbatim in the essay, in the sense the claim gives it: revenue needed to reach break even on two years of buildout, with the author adding that many trillions more would be needed for an acceptable return.
- The "let alone start turning a profit" framing in the post's caption matches the essay, which distinguishes break even from an acceptable return on the spend.
- Short equipment and facility life is genuinely the driver of the figure. The number rises from a prior $160 billion estimate only because the author shortened the assumed depreciation life.
- The gap framing is supported. The essay puts then-current AI revenue at roughly $15 to $20 billion a year against the $1 trillion requirement.
- The claim says "$1 trillion in revenue across 2025 and 2026," which reads naturally as cumulative revenue earned over those two calendar years. The essay's arithmetic points to an annual revenue level required to service the combined capex of both build years, derived from annual depreciation and compared directly against an annual revenue run rate of $15 to $20 billion. The phrase "across 2025 and 2026" describes the two years of spending, not the window in which the revenue would be earned, and the post's own wording leaves that unclear while using the word "annual" correctly for the Bain figure in the same piece. The author's own sentence is itself compressed on this point, so the post inherits an ambiguity rather than inventing one.
- The stated mechanism is paraphrased into a different physical story. "Components break down so quickly" suggests hardware failure. The essay's argument is obsolescence: new GPU generations arrive every year or two and cooling, racking, power and layout designs change, so the assets lose economic value long before they stop working. The direction of the argument survives the paraphrase, but the reason does not.
- Marketing as evidence, in a mild form: the figure is presented as a flat finding. It is one investor's published estimate, built on a stack of stated assumptions about capex totals, asset life and gross margin, which the author himself describes as directional rather than precise. A reader would not learn from the claim that changing the depreciation assumption alone moved the number from $160 billion to $320 to $480 billion.
- Whether the author intended the $1 trillion as an annual run-rate requirement or as a cumulative two-year total. The surrounding math and the comparison to a $15 to $20 billion annual figure point strongly to an annual requirement, but the sentence containing the number does not use the word "annual."
- Whether Kupperman has revised the $1 trillion estimate in any later writing. A search for a 2026 update returned no specific revision, so the figure is treated as his October 2025 position.
- The precise gross margin assumption inside the $320 to $480 billion calculation. The 25 percent figure is reported by a secondary outlet and traced to the prior essay, not quoted from the addendum text that was retrieved.
The essay exists, is by the named author, and contains the number. In "An AI Addendum," dated October 5, 2025, Kupperman revises his earlier assumption about how long AI data centers last: he had previously assumed a 10-year depreciation curve, which he now describes as unrealistic given how fast the technology is advancing, and says that based on his conversations the physical data centers last three to ten years at most, with changes to cooling systems, chip and racking designs, power systems and overall layouts driving that. On the figures, the essay states that speeding the depreciation curve to the three to five year range implies his prior breakeven revenue number of $160 billion to justify 2025's capex spend is inadequate, and that the industry probably needs revenue closer to the $320 billion to $480 billion range just to break even on the capex to be spent that year. The decisive sentence for this claim reads: "Adding the two years together, and using the math from my prior post, you'd need approximately $1 trillion in revenue to hit break even, and many trillions more to earn an acceptable return on this spend." It is immediately preceded by the observation that the industry is spending over $30 billion a month, roughly $400 billion for 2025, while receiving a bit more than a billion a month back in revenue, and that this ignores the hundreds of billions of additional data centers to be built in 2026, and followed by the reminder that revenue was then running at around $15 to $20 billion. The earlier August 2025 essay supplies the underlying arithmetic: assuming the building depreciates over 30 years, chips are obsolete in three to five years, and other equipment lasts about ten years on average, giving roughly a 10-year blended curve, the 2025 data centers would suffer $40 billion of annual depreciation while generating somewhere between $15 and $20 billion of revenue. Futurism's October 2025 article on the same essay noted the margin assumption behind the conversion from depreciation to required revenue, describing it as a generous 25 percent gross margin and putting actual AI revenue at closer to $20 billion annually.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Compact share page: ai.trueseeker.com/s/cd578984efb7/Ti9kOhC_D4Pi5wE3JZjuHy30N6R