“Silicon Valley operates on the premise that compute demand is unlimited, yet the recent slowdown in token spending growth suggests that this appetite may exist only when compute is effectively free,” he wrote in one of his regular missives. Slok’s immediate concern is that even modest price increases aimed at reflecting computing power’s true costs could prompt customers to cut usage or switch to cheaper open-source alternatives. This would leave the five biggest hyperscalers pouring ever more capital into infrastructure even as their returns decline.
Underlying this reluctance to pay is a more troubling possibility: that the promised AI productivity boom is arriving far more slowly than projected or, in many cases, failing to materialise at all. If that continues, Slok warns, the global AI growth story risks becoming a margin-draining liability.
If that is the case, it may be time to question whether the wider computing revolution, which we have come to regard as an unqualified force for good, was really worth it economically either.
Such scepticism is hardly new. Robert Solow, the late, great economist, quipped in 1987 that “you can see the computer age everywhere but in the productivity statistics”. At the time, computers were becoming ubiquitous in offices and businesses, yet their spread was failing to produce the surge in productivity that many economists predicted.
What is striking is how stubbornly the question persists. Productivity gains eventually emerged, notably in the late 1990s, but the longer-term growth record of advanced economies remains disappointing, particularly given the mountain of debt accumulated along the way.
Contrast that with the post-war period, when the US and other advanced economies enjoyed remarkable, sustained productivity growth. Admittedly, that growth had begun to slow before computers came into their own. But the question remains: why, for all their promise, did computers do so little to restore it to its former heights?
It is an uncomfortable question, not least because it flies in the face of everyday experience. To anyone who uses one, the suggestion that computers haven’t made us more productive seems absurd.
But what is good for the individual is not necessarily good for society.
It is becoming clear that information technology tends to redistribute the existing economic pie, rather than make it bigger. This is because much of its apparent economic dynamism stems from network effects that encourage socialisation and disintermediation, as well as a shift away from material production towards asset-light models. This, in turn, tends to favour financialisation over genuine wealth creation.
At the same time, while reliance on open protocols, universal access and network effects encourages explosive adoption that seems to benefit everyone, it also has a downside: the gains can be offset by the harms they enable.
This is where research by Geoffrey West, a complexity scientist, into cities offers insight.
One of his key findings is that, as cities grow, both the benefits and the costs of economic activity increase faster than their populations, a phenomenon known as superlinear growth. Information technology amplifies this effect by connecting ever more people.
Crucially, such growth says nothing about the quality or value of that activity. It tells us that more is happening, not whether society is better off. More connections generate more opportunities; that is true. But, as it turns out, they also generate more competition, transactions and threats, including from monopolisation and wealth concentration. According to West, the overall outcome is a general acceleration in the pace of life, without necessarily a corresponding increase in collective prosperity.
An innovation treadmill?
If superlinear growth is left unchecked, its mathematics points towards a finite-time singularity, threatening Malthusian-style collapse. West argues that innovation can stave off this fate by resetting the growth trajectory, but only temporarily. Each breakthrough buys less time than the last, forcing society to innovate ever faster merely to keep the system going.
In other words, humanity finds itself on an innovation treadmill, running ever faster just to stay in the game.
The implication is that every legitimate business made more productive by information technology may find its gains eroded by competitors, predators or criminals made equally productive by it. The outcome can resemble a perpetual arms race between good and bad actors.
Other costs are harder to quantify, let alone capture in productivity statistics. Take the benefits of cheap communication and universal connectivity.
Not so long ago, international telephone calls were prohibitively expensive, while moving abroad meant losing contact with the culture, media and communities one left behind. The internet swept those barriers away, seemingly to everyone’s benefit. We entered, as the media theorist Marshall McLuhan called it, the age of the global village.
Yet there may have been an overlooked social cost. By allowing us to remain immersed in the cultures of our home countries, digital connectivity arguably weakened incentives to assimilate into host societies. Translation technology further compounded the effect by reducing the need to learn the local language.
The same logic applies to AI. Many of us can sense this instinctively: for every advantage AI gives us in working faster, it seems also to make scammers, fraudsters and criminals more effective at exploiting us. The result is a growing breakdown in social trust, the costs of which may be far harder to quantify in monetary terms than any productivity gains.
The consequence is a paradox: societies are more connected than ever technologically, but less cohesive culturally, with communities living alongside one another yet increasingly socially apart.
Businesses, for their part, may be waking up to the same problem: the difference between the benefits technology promises and the value it delivers is narrowing.
Many may even remember the BlackBerry effect: companies that first equipped employees with the devices enjoyed a productivity advantage, but only until everyone else caught up. What began as a competitive edge became a cost of doing business soon enough.
AI threatens to shorten that period of competitive advantage dramatically.
This is where Slok’s observations become pertinent again. Companies will pay handsomely for AI compute if it delivers sustainably higher margins. But if competitors can acquire the same capabilities immediately, where is the lasting return? Hence, perhaps, their growing reluctance to pay the full economic cost of AI services.
Nothing is set in stone. As Slok himself concedes, the AI productivity miracle may yet arrive. But it seems foolish to ignore the consequences if it does not.
Even more foolish would be to overlook the growing appeal of “friction-maxxing”: a celebration of moats, tariffs, proprietary systems and doing things the hard way just because you can. Think of it as the intellectual equivalent of going to the gym. I hear it is already a big thing with Gen Z.
