Infrastructure capacity deficits keep hardware demand completely inelastic to double-digit price hikes.
Because AI workloads are severely supply-constrained rather than demand-constrained, leading hardware designers can implement massive price increases to protect their margins without triggering any reduction in buyer procurement.
The same conclusion keeps arriving from across the workspace's research — 2 topics independently instantiate this theme. Filter the evidence by where it came from:
With scarce capacity auctioned at premiums and short-term deals priced at $40–50M per megawatt, supply-constrained buyers keep paying up rather than walking away.
Buyers bid above record list prices at auction and pay double per megawatt for short duration because the capacity simply is not there.
Days after Nebius's 17–21% GPU rental hike, another double-digit increase lands without demand pushback because workloads are supply-constrained, not demand-constrained.
Because the supply deficit is credible and buyers still sign for the vast majority of 2027 bit supply, suppliers can push triple-digit price hikes that demand absorbs without procuring less.
Buyers absorb a higher price for a cut-down product, confirming that supply-constrained demand lets the vendor raise prices without losing any procurement.
With supply — not demand — the binding constraint through FY2028, a 41% scarcity price hike lands without any reduction in buyer procurement.
The market happily absorbs double-digit system price hikes because computing capacity remains heavily supply-constrained.
A rental market absorbing ~20% fleet-wide hikes — even on four-year-old H100s — while capacity is still being added shows buyers pinned by capacity deficits, not by demand weakness.
Severe memory supply deficits allow Nvidia to protect its margins through double-digit price increases without dampening extreme customer demand.