Microsoft (MSFT) earnings

Huge red flags are emerging

Starting in fiscal 2027, Microsoft is extending the estimated useful life of its data centres and office buildings from 15 years to 25 years. The result? Lower annual depreciation expense and, therefore, higher reported earnings.

But it gets more interesting.

That longer useful-life assumption also changes how some leases are classified under accounting rules. Microsoft’s future data-centre leases will now be treated as operating leases rather than finance leases. Finance leases are included in reported capital expenditure. Operating leases are not.

So, suddenly, Microsoft’s headline 2026 CapEx guidance falls from roughly $190 billion to around $175 billion, even though the underlying spending plans have not changed.

In other words, Microsoft is still pouring money into AI infrastructure at a blistering pace. It is just becoming less visible in the income statement and less visible in the cash flow statement, where operating lease payments are not included in standard free cash flow calculations in the same way as capital expenditure.

This is presentational. Nothing has changed under the hood.

The economic reality remains a massive, front-loaded build-out of AI capacity with uncertain returns. And a significant chunk of that spending is on compute hardware that has a useful life of around six years, not 25.

So the questions investors should be asking are simple:

Why this?

Why now?

Over the last four quarters, Microsoft has spent around $97 billion on infrastructure and equipment to generate roughly $37 billion of annual recurring revenue from AI services.

And the spending is accelerating.

Capital expenditures and finance leases for the fiscal fourth quarter reached $41 billion (up 69% year-over-year).

Microsoft has said it expects CapEx spending to be up again in calendar 2026, which means it will be running well ahead of prior analyst expectations.

At the same time, CapEx intensity is rising as Microsoft brings more data-centre capacity online.

Management also noted that around $25 billion of the increase is due to higher component costs, including GPUs, memory and other materials, rather than simply buying more units.

To me, this raises a much bigger question.

Is AI infrastructure spending simply getting out of control?

If so, is Microsoft now taking increasingly creative accounting and presentation measures to soften the optics?

Because changing the accounting treatment does not change the economics.

It doesn’t make the infrastructure cheaper; it doesn’t make the AI returns more certain; and it certainly does not change the fact that Microsoft is committing enormous amounts of capital today for revenues and returns that may or may not materialise tomorrow.

Investors beware!