CoreWeave Extends Nvidia A100 Contracts Through 2029

CoreWeave has extended the commercial lifespan of Nvidia’s 2020 A100 graphics processing units by securing a leasing contract through 2029, underscoring the enduring profitability of legacy artificial intelligence hardware. During the company’s second-quarter earnings call, Chief Executive Mike Intrator revealed the agreement, noting that pricing for prior-generation SKUs currently sits at or above historical peaks. The announcement accompanies robust financial results, with CoreWeave reporting $2.58 billion in quarterly revenue, an 112 percent year-over-year increase, and a $104 billion backlog that includes over $25 billion in new commitments since July. Chief Financial Officer Nitin Agrawal added that capacity coming up for renewal represents a minimal portion of the fleet, as older-generation average selling prices remain resilient. The nine-year deployment cycle directly challenges prevailing industry skepticism regarding accelerated silicon obsolescence. Investment investor Michael Burry recently alleged that hyperscalers artificially inflate AI economics by stretching GPU useful-life assumptions to five or six years, despite Nvidia’s rapid architectural cadence. CoreWeave’s 2029 commitment, alongside previous reports of H100 units being rebooked at 95 percent of original pricing and Nvidia CFO Colette Kress defending sustained A100 utilization rates, demonstrates that older chips can outpace standard depreciation schedules when market demand and infrastructure constraints align. A primary catalyst for sustained A100 demand is a critical mismatch between legacy data center infrastructure and next-generation power requirements. Air-cooled Nvidia DGX A100 systems consume approximately 6.5 kilowatts at peak load and operate within traditional facility designs rated for roughly 20 kilowatts per rack. In contrast, Nvidia’s latest Blackwell-based GB200 and GB300 NVL72 configurations require 120 to 140 kilowatts per rack and mandate direct-to-chip liquid cooling systems. Deploying Blackwell hardware necessitates complete electrical and thermal infrastructure overhauls, preventing seamless migration into active A100 halls. Consequently, energized, air-cooled capacity lacks higher-value alternatives, making the rental of aging silicon a financially optimal solution for cloud providers. CoreWeave’s operational metrics highlight the intensity of this demand. The company contracted power availability to 3.7 gigawatts during the second quarter, rising to 4.2 gigawatts by the time of the earnings call, while only 1.5 gigawatts remain online. This substantial overcommitment indicates that customer reservations already nearly triple deliverable capacity, reinforcing a business model built on leveraging existing infrastructure bottlenecks. As Nvidia continues to advance its architecture, the strategic persistence of older GPU fleets suggests that facility limitations will continue to dictate hardware lifecycles, allowing established data center operators to maximize asset utilization and maintain premium pricing long after initial market debuts.