Industry
The Day Models Become Boring
A reliable signal announces that a layer is commoditizing: the debate about it becomes boring. Models are getting there.
There is a very reliable signal, in the history of enterprise technologies, to spot the moment a layer commoditizes: it is the moment the debate about it becomes boring. People stop comparing vendors with passion. Conferences stop devoting their flagship sessions to it. Comparisons stop occupying the specialized press. Not because the technology is less important, on the contrary it is now everywhere, but because the gaps between the options no longer justify the intellectual effort of ranking them.
One can verify this signal on past waves, and it does not deceive. Databases ceased to fascinate in the mid-nineties, not because they had failed, but because they had become a base one no longer discussed. Servers met the same fate a decade later, then cloud services in turn. Each time, the silence that wins the debate is not a sign of decline, it is a sign of maturity: the technology is so well installed that it has ceased to be a subject.
Why silence is a sign of maturity
This signal deserves a pause, because it is counterintuitive. One spontaneously believes that an important technology arouses intense debate, and that a debate that dies out announces a decline. It is the opposite: intense debate is the sign of a still-young market, where the gaps between options are large and the bets open; the debate that dies out is the sign of a mature market, where the options are equivalent and the interesting question has moved elsewhere. Boredom is not the end of the story, it is the displacement of its center.
You recognize a commoditizing technology not when it fails, but when it ceases to be a subject of debate.
This displacement of the debate’s center always follows the same path. As long as a technology is young, the question that fascinates is “which to choose”: one compares, tests, bets, because the choice has heavy and lasting consequences. Then the options converge, the gaps narrow, and the question “which to choose” loses its interest, not because it no longer has an answer, but because all the answers are roughly equivalent. At that moment, attention shifts to the next question, the one that had no urgency yet: “how to articulate these bricks that have become interchangeable.” The debate does not die, it rises a notch.
The early signs are already here
Applied to language models, this signal is lighting up before our eyes. Model demonstrations, which packed rooms eighteen months ago, are starting to resemble each other. Benchmarks arouse polite interest rather than passionate debate. Serious organizations stop asking which model is best and start asking which ones to combine, in which architecture, under which governance. The intellectual center of gravity of the market is shifting, slowly but clearly, from the model to what is built above the model.
One must specify what commoditizing means here, for the word can mislead. A technology that commoditizes does not lose its importance, it loses its differentiation. Electricity is commoditized, and no one will contest its importance; one simply no longer chooses one’s supplier by the quality of its electrons. Models will follow this path: they will remain indispensable, but the choice between them will cease to be a strategic act to become a trade-off of cost and availability. This shift, from differentiation to commodity, is exactly what is beginning.
To commoditize is not to lose importance. It is to cease making the difference.
This distinction between importance and differentiation is the heart of the misunderstanding about the commoditization of models. Those who refuse to believe models commoditize argue, rightly, that they are more and more important, more and more capable, more and more central. All this is true, and has nothing to do with the question. A technology can gain in importance and lose in differentiation at the same time: it is even the normal trajectory. The more models become powerful and omnipresent, the more they resemble an infrastructure, and the less the choice between them makes the difference. Importance and commoditization do not oppose each other, they advance together.
When this commoditization is complete, the strategic conversations in AI committees will have radically changed in nature. They will no longer bear on the choice of a model vendor, but on the architecture of the upper layer: who has one, who does not, what quality, what sovereignty, what durability. The model will have become an ordinary purchasing decision, an invoice line negotiated without devoting a committee to it, like hosting or telephony. Strategy, for its part, will have entirely moved one notch above.
This transition has a major, and slightly cruel, strategic implication. Organizations that structure their AI thinking today around the choice of a model are preparing a strategy that will be obsolete at the precise moment it starts to work. Conversely, those that structure their thinking around the upper layer, semantic, orchestration, governance, build an asset that will gain value exactly as the model layer loses its own. The calendar rewards those who look at where value is going, not where it still is.
To build your AI strategy on a model today is to bet on what will be worth least tomorrow.
One must measure how useful this signal is, precisely because it is available before the commoditization is complete. One does not need to wait for models to have become commodities to know it; it suffices to observe the debate that, already, is running out of breath. The signal precedes the phenomenon, which leaves those who can read it a window to act before the others. Those who wait for the definitive proof, that of the fully commoditized market, will act at the same time as everyone else, that is, too late. Those who read the early signal, the boredom setting in, have a lead the signal itself offers them.
One might think it will suffice, when the time comes, to switch one’s strategy from the model to the upper layer, and that there is therefore no urgency. It is a calendar error. The upper layer is not bought ready to use the day one needs it: it accumulates in usage, over eighteen months or two years, as it learns the matters, the methodology, the permissions of the organization. Whoever waits for models to be fully commoditized to start building their layer will, at that moment, be two years behind whoever started while the debate on models was still in full swing.
It is this inversion that MAX anticipated from its conception. Not by betting against models, which remain necessary and good, but by deliberately placing its investment in the layer whose value will grow as that of the one below decreases. The moment the debate on models becomes boring will be the moment the layer above becomes decisive, and the organizations that will have already built theirs will operate with a lead the others will take years to close.
The boredom that will win the debate on models will be the signal that the real game has begun above.
