5090 sales fueled by LLMs
RTX 5090 cards are selling strongly on Amazon this week, and community reporting says much of the demand is coming from local LLM/AI use cases rather than pure gaming. That shift helps explain constrained availability and the renewed debate over value versus AMD’s 9070XT and 5070 Ti. (x.com)
Cloud GPU marketplaces are listing consumer RTX 5090 instances for on-demand rent (pricing examples show offers from ~$0.89/hr), highlighting commercial AI workloads as a monetized use case. (runpod.io) Independent LLM benchmarks recorded the RTX 5090 at about 102.7 tokens/sec on 14B models at 16K context in March 2026, demonstrating the card’s raw throughput advantage for local inference. (hardware-corner.net) Community-maintained resources and benchmark projects — including a Local LLM directory on GitHub and the LocalLLM.in best-GPU guides — list the RTX 5090 among top consumer choices for running local LLMs, reflecting sustained developer interest. (github.com) Retail availability has tightened: reporting shows the RTX 5090 “vanishing” from U.S. stores in recent weeks, and German sellers have cited memory supply issues and order limits as contributors to scarcity. (windowscentral.com) Market snapshots used in press comparisons found roughly 70 Radeon RX 9070-series cards available versus a single RTX 5090 in one retailer’s inventory, underscoring comparatively stronger short-term 9070 availability. (techspot.com) Pricing spreads and fraud incidents are surfacing: the 5090’s $1,999 MSRP has translated into typical retail pricing in the $2,500–$3,000 range in trackers, and at least one buyer paid about $3,200 on Amazon and later received a household detergent package instead of a GPU. (ofzenandcomputing.com)
