The Great AI Infrastructure Wager Meets Its Cost Reckoning

The artificial intelligence boom that has propelled technology stocks to record valuations is entering a more complicated phase, as the vast revenues flowing to cloud infrastructure providers collide with mounting evidence that corporate adopters are struggling to extract commensurate value. Aggregate spending by the largest technology companies on AI is now projected to surpass $725 billion in 2026, a figure revised upwards by $125 billion and standing 77 per cent above last year's outlay. Viewed from Moscow's financial press, the logic appears unassailable: first-quarter earnings from the American hyperscalers showed cloud computing revenues accelerating sharply, with Alphabet's Google Cloud surging 63 per cent, Microsoft's Azure notching 40 per cent growth, and Amazon Web Services climbing 28 per cent. The infrastructure providers are, for now, the unambiguous beneficiaries of the AI revolution.

Yet the experience of the companies actually deploying these tools tells a markedly different story. Microsoft, itself one of the leading purveyors of AI services, has begun scaling back its internal use of Anthropic's Claude model, citing untenable costs. Ride-hailing giant Uber exhausted its entire 2026 artificial intelligence budget in just five months, a pace of consumption that has sent shockwaves through corporate technology planning circles. The original promise, widely propagated during 2025, held that AI would simultaneously reduce operational expenditure by replacing human labour and boost productivity by an order of magnitude. Instead, as industry observers note from New Delhi, costs have risen far faster than any measurable efficiency gains, prompting a quiet but consequential reassessment across boardrooms from Bengaluru to Seattle.

This recalibration extends well beyond the consumer internet sector. In Sydney, the Australian Securities Exchange has announced a significant increase in its capital expenditure envelope, lifting its target to between A$180 million and A$200 million for the coming fiscal year, up from a previous range of A$160 million to A$180 million. The funds are earmarked for modernising critical market infrastructure after years of technical failures that eroded stakeholder confidence and drew regulator scrutiny. While the ASX programme is not exclusively AI-driven, it forms part of a broader global pattern: institutions are committing unprecedented sums to technological infrastructure at precisely the moment the cost-benefit equation of the most hyped component of that infrastructure faces serious questioning.

The divergence between infrastructure providers' soaring revenues and adopters' mounting disenchantment presents an uncomfortable question for policymakers and investors alike. Is this the productive build-out of a transformative general-purpose technology, or an echo of the fibre-optic overinvestment that preceded the dot-com collapse? Analysts in London caution that the comparison, while tempting, may obscure as much as it reveals. Unlike the turn-of-the-century bubble, today's AI spending is concentrated among a handful of deeply capitalised incumbents whose core businesses generate enormous free cash flow. The reckoning, if it comes, is more likely to manifest as a sobering repricing of expectations than a systemic rupture. What appears increasingly certain is that 2026 will be the year the AI industry is forced to demonstrate that its products can deliver value commensurate with their extraordinary cost, rather than simply promising it.