AI compute running short
The AI industry is hitting hard capacity limits: outages, rationing and sharply higher GPU prices have emerged across the market. (the-decoder.com) Hourly rent for Nvidia Blackwell GPUs is reported at $4.08 — up about 48% from $2.75 two months earlier, according to a compute‑price index. (intellectia.ai)
The business of renting artificial intelligence chips has tightened fast: prices are up, supply is constrained, and major model providers have posted capacity-related disruptions. (the-decoder.com) Artificial intelligence models run on graphics processing units, or GPUs, which are specialized chips built to handle many calculations at once. The Decoder reported that shortages are now showing up as outages, usage caps and delayed access across the market. (the-decoder.com) One price signal came from Nvidia’s new Blackwell generation. Intellectia, citing a compute-price index reported by The Wall Street Journal, said hourly rental rates for Blackwell GPUs reached $4.08, up from $2.75 about two months earlier, a jump of roughly 48%. (intellectia.ai) Nvidia has been warning investors that demand for its newest systems is arriving at extraordinary scale. In its recent earnings materials, the company said Blackwell revenue ramped faster than any product launch in its history. (investor.nvidia.com) That squeeze is reaching users through software, not just hardware markets. OpenAI’s public status page has repeatedly logged degraded performance and elevated errors tied to heavy demand, while Anthropic has used rate limits and availability controls during busy periods. (status.openai.com) (status.anthropic.com) The bottleneck is not a single chip shortage. Companies need whole clusters of GPUs, plus networking gear, power, cooling and data-center space, and each missing piece can slow deployment even when chips are available. (the-decoder.com) The timing tracks a change in how these systems are being used. The recent push into “agentic” tools — software that runs longer, multi-step tasks instead of answering one prompt at a time — increases computing time per user and keeps expensive chips occupied longer. (intellectia.ai) Cloud providers and model companies are still adding capacity, but new supply takes months to install and connect. Until more clusters are online, the market is signaling scarcity in the simplest way it can: higher rental prices and tighter access. (the-decoder.com)
