Andreessen Horowitz reports GPU rental prices rose across every Nvidia chip generation in 2026
GPUs are aging like fine wine, with rental rates up across the board since January 1st Approximate change YTD: - H200: +160% - B200: +120% - RTX PRO 6000: +65% - H100: +35% Even the 6-year-old A100 rents for more than it did at the start of the year Charts of the Week:
Context
Rising spot and short term rental rates across both Blackwell and six year old Ampere GPUs contradict assumptions of rapid hardware depreciation and point to sustained compute scarcity driven by inference and agent workloads.
Data center GPUs have historically been modeled with three to six year depreciation schedules as newer architectures deliver higher compute density per watt. Throughout 2026, compute exchange startup Ornn has tracked transacted on demand GPU rental prices across cloud providers to benchmark real time compute supply and demand.
Deep dive
In its October 23 Charts of the Week column by Moses Sternstein, Andreessen Horowitz published data from Ornn Data showing that on demand GPU rental rates have risen across every tracked Nvidia generation between January 1, 2026 and October 19, 2026. Using a trailing seven day average indexed to 100 at the start of the year, Ornn's index shows Nvidia H200 rental prices up approximately 160 percent year to date, B200 rates up roughly 120 percent, RTX PRO 6000 Workstation rates up 65 percent, and H100 SXM rates up 35 percent.
Even the six year old Nvidia A100 SXM4, first released in 2020, ended mid October renting for roughly 5 percent more than it did on January 1 after spiking in June. In the accompanying newsletter analysis, a16z noted that comparing historical rental prices against a standard six year straight line depreciation curve yields a 9 percent to 143 percent pricing surplus across the five models, with the sharpest upward inflection beginning in April 2026 alongside surging token volume on platforms like OpenRouter.
Ornn Exchange separately reported that its B200 index recently reached roughly 8.54 dollars per GPU hour while H100 rentals hovered near 2.81 dollars per hour. Because Ornn's index measures spot and short term transacted rates rather than multi year fixed price hyperscaler commitments, the surge primarily reflects acute scarcity for flexible capacity as enterprises and AI labs scale inference and agentic workloads.
What to expect next
If older Hopper and Ampere fleets continue commanding rental rates above their original six year depreciation curves, cloud operators and GPU debt lenders will hold stronger collateral values when existing three year contracts roll off, though data center power, transformer delivery timelines, and high voltage switchgear bottlenecks could keep flexible spot capacity tight across major cloud regions well into 2027.