Quick Answer
In 1846, Britain poured roughly 7% of its national income into railways, proportionally about three times what the U.S. spends on AI infrastructure today. The technology delivered everything it promised, and a generation of investors still lost their shirts. What sorted the winners from the wreckage wasn't conviction about the technology; it was whether ROI was measured or asserted. The man who fixed that problem gave his name to the S&P 500.
In 1846, Britain was in the grip of Railway Mania. Parliament passed 272 separate railway acts in a single year, authorizing thousands of miles of new track. Everyone was in: the Brontë sisters (yes, the novelists) held railway shares, and Emily managed the family position. Charles Darwin invested. The most celebrated man in the country was George Hudson, the “Railway King,” who controlled nearly a third of the network.
By 1849, the railway crash wiped out a generation of middle-class savings. Hudson was exposed and ruined. But Britain lay crisscrossed by railroads — the track got built. Within a decade, Britain had the densest rail network on Earth, one that would carry its commerce for the next century.
How can both be true?
How did the railway, a proven system for cheaper freight, faster travel, national markets, lose so many investors so much money?
The conventional story calls it a speculative frenzy, too much money chasing a shiny new technology. But if it was a frenzy, it wasn’t because of unsound tech, it was because of unsound measurement.
There were no standard accounts. Promoters published traffic projections, not traffic. And Hudson’s signature move was paying handsome dividends out of newly raised capital rather than earnings (a model Charles Ponzi would later copy). It wasn’t even clearly illegal, because no reporting standard existed that would have revealed it. Investors saw the dividend and read it as ROI.
Report
Finance needs to prove AI’s return: CloudZero report
260 senior finance leaders (more than half CFOs) told us why the speed of seeing AI spend, not the size of it, separates who pulls ahead on AI from who gets burned.
The man who built the meter for capital markets
At the same time, the same story played itself out in America, with its own vocabulary: watered stock, paper railroads, dividends conjured from fresh share issues.
Henry Varnum Poor, editor of the American Railroad Journal, went after the measurement vacuum. Starting in 1849, he demanded that railroads publish standardized figures: capitalization, debt, earnings, operating costs. Many refused. He published anyway, flagging the ones that wouldn’t disclose. Poor called it the investor’s “right to know.”
By 1868, Poor’s Manual of Railroads was the annual reference no serious railroad investor went without. A road that reported real, comparable earnings could raise capital cheaply. A road that offered projections paid a penalty, if it got funded at all. Measurement sorted the buildout instead of killing it.
In 1941, Poor’s firm merged with Standard Statistics to form Standard & Poor’s, which launched the S&P 500 index in 1957. The benchmark of every investment in the world is named after the man who standardized ROI reporting for the last infrastructure buildout of this scale.
The lesson isn’t “measure your costs.” Carnegie already taught us that one. The lesson is one layer up: Markets can’t price what they can’t compare. And the party who defines the reporting standard ends up owning the benchmark.
AI is in its pre-Poor phase
Hyperscalers are spending roughly $700B a year on CapEx. AI revenue is still well under $100B. As I’ve written before, the buildout needs something like $1T in annual revenue to pencil.
I don’t think AI is a bubble. Railway Mania is why the bubble question is the wrong binary: the technology can win while enormous amounts of the capital funding it earns nothing.
The AI market has already started its sorting. Last month, Alphabet beat on revenue and watched its stock sink anyway after raising CapEx guidance to $195–$205 billion, the second hike in three months, with no ROI line attached. TSMC posted a record $40 billion quarter and fell 4%. Faith stopped working. Meanwhile the reward side of the trade sits mostly vacant, because almost no company reports AI ROI in a form the market can compare. That vacancy is the opportunity.
Look at how companies report AI today, and tell me which era it resembles:
| Railway buildout | AI buildout | |
|---|---|---|
| The claim | Traffic projections in a prospectus | Usage charts in a board deck |
| The comforting non-answer | Dividends paid from capital | “The ROI will be so obvious you won’t need to measure it” |
| The number that sorted winners | Standardized earnings, published annually | Revenue net of AI spend per employee |
| Who produced the numbers | Railroads with standardized accounts | Your company, if it builds the machinery |
| Who owned the standard | Standard & Poor’s | Being decided right now |
Seats provisioned, active users, token spend, chat volume: These are traffic projections. They describe enthusiasm, not return. Only 22% of finance executives can tie AI spend to business outcomes today. The other 78% are buying railway shares on the strength of the dividend.
The re-rating comes before the requirement
Poor’s Manual didn’t wait for regulation. Disclosure became a competitive weapon decades before it became a requirement, because railroads that reported comparable numbers got cheaper capital sooner.
The same asymmetry is opening in AI. A company that reports revenue net of AI spend per employee — and the movers underneath it, like sales velocity per AE and retention per CS dollar — is making a margin and labor-leverage claim the market can price into its multiple. A company showing an adoption chart is making a claim about enthusiasm. Those two will not be valued the same for long.
Note the two different seats in the table, because they’re easy to conflate. The railroads’ seat, produce numbers worth indexing, is the one your company occupies. Poor’s seat, define the standard and build the instrument that makes numbers comparable, is the one CloudZero is playing for. Our AI signals agent captures every AI event the moment it happens and logs it as cost. AI outcome attribution connects that spend to the value it produced. Multi-dimensional allocation traces it to the workflow that earned it. Not so you can admire a cleaner bill — so that when the market starts demanding comparable AI numbers, yours already exist.
The mania was never the risk. Unmeasured mania was. Britain kept the railways; the fortunes that survived belonged to the people reading standardized numbers. The AI era’s version of Poor’s Manual is being written right now. The only question is whether your company shows up in it with numbers, or with projections.
