Contents · 3 sections+
History does not repeat itself. But it does send invoices.
In the late 1990s, a Dutch entrepreneur named Jeroen Mol articulated the prevailing theology of the age with admirable candour. Profit, he observed, was not a particularly interesting concept. His vehicle, GorillaPark, Amsterdam's most exuberant dotcom incubator, raised fifty million guilders in 2000 and deployed them with the focused intensity of a company that had confused velocity with strategy. When the NASDAQ corrected, so did the philosophy.
Twenty-five years on, a new doctrine has taken hold in technology departments across Singapore, Dubai, Amsterdam, and elsewhere, as ambitious companies spend aggressively to avoid being left behind. It does not have a name yet. But it has a unit of measurement: the token.
I.The Meter Is Running
Every prompt submitted, every codebase handed to an AI agent, every "just let it generate a first draft" decision made in an engineering stand-up represents a billing event. The beneficiaries are not the companies deploying these tools. They are the companies selling access to them, OpenAI, Anthropic, Google, Microsoft, whose revenue dashboards illuminate with each API call.
This is not an argument against artificial intelligence. It is an observation about who, in the current arrangement, is actually accumulating value.
The platform companies are not burning anything. They are, to use the technical term, profitable. The token spend flowing through their infrastructure from thousands of enterprise clients is funding their margin with a reliability that most of those clients' own businesses cannot match. Your engineers' shower thoughts are someone else's earnings per share.
The structural parallel to 1999 is not merely aesthetic. In the dotcom era, venture capital funded the burn while founders deferred the fundamental question of unit economics. Today, boards approve AI transformation budgets while leadership teams defer an identical question: what, precisely, is this producing?
The vocabulary has been updated. Eyeballs became a community. Community became an AI adoption. The metric quietly omitted in each iteration has remained consistent.
II.The Question That Hasn't Been Asked
Inside most organisations currently scaling their AI tooling, a specific conversation has not yet occurred. It will. It tends to arrive not as a strategic review but as a line item, a number on a quarterly report that prompts a characteristically polite CFO to request clarification.
"Can someone walk me through the return on the AI infrastructure spend?"
The silence that follows this question is historically where interesting decisions get made. In 2001, GorillaPark's investors had a version of this conversation. So did the boards of several hundred companies whose names now appear only in business school case studies. The details differed. The dynamic was identical.
What tends to emerge from that silence is not a retreat from technology but a reckoning with discipline. The seventeen prototypes that never reached production. The agentic workflows that automated processes no one had previously confirmed were worth automating. The vibe-coded solutions to problems that a competent engineer, given adequate time, would have solved more durably for a fraction of the cost.
III.The CFO's Moment
The executives who will define the next phase of enterprise AI adoption are not the ones approving the largest model subscriptions. They are the ones insisting on the oldest questions: What does this cost? What does it produce? When does it pay?
These are not anti-innovation questions. They are the questions that distinguish strategy from superstition.
The platform companies understand this. Their business models were built on it. They have already answered the profitability question, with their customers' capital.
History suggests the correction, when it comes, will feel sudden to those inside it and entirely predictable to those observing from the outside. The bubble does not announce its intentions. It simply continues expanding until someone with a mandate and a spreadsheet asks for the actual numbers.
In most organisations, that person already exists.
They are waiting for the right quarter.