The appropriate depreciation rate for NVIDIA GPUs has consistently ranked as one of the biggest unknowns of the ongoing AI build out, with proponents continuing to argue in favor of a longer shelf life, while detractors contend that a 3-year useful life estimate is most appropriate.

The latest hourly rental rate figures for NVIDIA GPUs, however, appear to bolster the point of view espoused by the proponents of longer useful life estimates, which bodes well for the income statements of AI hyperscalers.

According to Silicon Data, NVIDIA H100 GPUs were commanding an hourly rental rate just shy of $2 in January 2026. Now, however, those rates are decidedly north of the $2 threshold, and approaching the $3 mark.

Meanwhile, NVIDIA's B200 GPUs were going for an hourly rate just below the $5 threshold in January, but are now hovering between $5.50 and $5.80.

Of course, one theory behind this aberrant price appreciation relates to the fact that AI models themselves are progressively becoming much more efficient to run on the same compute footprint, which allows older NVIDIA GPUs to better retain their value amid a Jevons paradox-type paradigm, which posits that as barriers around a new technology ease, its uptake increases proportionally.

Finally, this development means that hyperscalers like Microsoft can comfortably increase the useful life estimate for their stash of NVIDIA GPUs, which should reduce the quantum of GPU cost expensed each year via the depreciation line-item on the income statement, giving an additional boost to their net income.

Follow Wccftech on Google to get more of our news coverage in your feeds.