Momentum warns hyperscalers’ 5-6 year server lives outstrip 2-3 year AI GPU economics

Momentum Global flagged a widening gap between hyperscalers’ 5-6 year useful-life assumptions for server and network gear versus 2-3 year estimated economic lives for frontier GPUs.

AI now drives about 75% of hyperscaler capex, shifting new asset vintages toward shorter-lived hardware while depreciation still reflects an older, less GPU-heavy fleet.

Across the five largest US hyperscalers, property and equipment spend totaled about $550 billion in the last 12 months versus about $160 billion of depreciation.

Nvidia’s GPU release cycle has tightened to roughly annual launches, compressing the replacement cycle and lowering the earnings power of older chips via weaker cost efficiency.

Straight-line depreciation is portrayed as masking a front-loaded economic benefit profile, raising the risk that forecast earnings growth embeds accounting uplift rather than proven returns.