America’s technology giants have found a new way to make artificial intelligence look profitable: change the maths.
Between 2022 and 2024, Microsoft, Google, Amazon, Meta and Oracle each extended the assumed life of their servers and data-centre equipment. What used to be written off over four years will now be depreciated over five, five and a half, or even six years. No accounting rules were broken; auditors signed off. But the shift has quietly turned billions in annual costs into apparent profits.
The timing matters. These companies are in the midst of the largest investment boom in corporate history, pouring more than a trillion dollars into AI data centres, chips and cloud infrastructure. Stretching depreciation schedules softens the hit to reported earnings. Analysts estimate the change could inflate cumulative profits by around $170 billion by 2028.
The assets in question — the high-powered chips that train and run large language models — rarely remain competitive for less than two years before new generations make them uneconomical. Accounting, however, allows companies to treat them as if they will stay productive for six. The gap between those assumptions is the gap between real costs and the AI story investors want to believe.
This is not just a technical curiosity. It is a political act — one that shapes how the rewards of the AI economy are distributed.
The politics of the balance sheet
Accounting is never neutral. Every assumption — how long an asset lasts, when a cost is recognised — shapes who captures value. Slow the depreciation, and expenses shrink while profits swell, lifting share prices and executive bonuses. Tax bills may be unaffected (fiscal depreciation runs on its own schedule), but the political effect is just as real: it pulls future gains into the present, inflating today’s earnings, valuations, and rewards — even as the assets beneath them age far faster than the books admit.
This also shapes the story being told about the AI boom. On paper, the sector looks like a triumph of American innovation and industrial revival. Those headline profits underpin the political narrative that Silicon Valley is leading the next great productivity surge. But the numbers are built on assumption, not evidence. The profits exist because accountants say the machines will last six years. If they turn out to be obsolete in two, that story collapses.
Meanwhile, there is a harder constraint: power. The AI industry’s appetite for electricity is enormous. A single large data centre can consume as much energy as a small city. Goldman Sachs expects AI computing to require about 5 per cent of total US electricity by 2027. Power grids are already stretched. Some new facilities will sit idle, waiting for grid connections. But the equipment still appears as productive capital on company accounts, generating paper profits even before it delivers a single computation.
When narrative becomes asset
Accounting, in this sense, is not just a language for describing business — it is a tool for creating it. The promise of future AI productivity has been converted into profits today. Depreciation schedules have become instruments of belief. The result is an economy where the politics of value are written in the footnotes of quarterly filings.
For now, the illusion holds. Meta reported $62.5 billion in profit last year, up 59 per cent. Oracle’s earnings rose despite heavier borrowing. Microsoft and Google are still treated as engines of limitless growth. But behind those figures lies a tension between the physical world of chips and power, and the financial world of accounting time. One moves much faster than the other.
When the correction comes — as equipment ages faster than expected and earnings growth slows — the political narrative will shift too. The AI boom will look less like industrial renewal and more like another cycle of rentier capitalism, where the rewards of innovation flow to shareholders and executives, underwritten by accounting flexibility and the physical limits of energy.
The technology may yet reshape the economy. But the expected profits are not proof of productivity — they are accounting choices disguised as growth, a reminder that in Silicon Valley, narrative still trumps reality.
Leave a Reply