What Happened
A WSJ report published today details a sharp capacity crunch across the AI industry. Anthropic has been plagued by outages and started metering token use during peak hours. Its Claude API uptime in the last 90 days was 98.95%, well below the four-nines standard enterprise software normally commits to. OpenAI scrapped its Sora video app partly to free up compute for coding and enterprise products. Token use in OpenAI's API jumped from 6 billion a minute in October to 15 billion a minute by late March. Nvidia Blackwell GPU rental prices are up 48% in two months, to $4.08 an hour. CoreWeave raised prices over 20% and is now requiring three-year contracts instead of one. Anthropic's revenue tripled from $9 billion to $30 billion in four months.
My Take
This is the first time AI providers are openly telling customers "you cannot use as much as you are paying for," and it is going to reshape expectations fast. For years the pitch was unlimited intelligence on demand. Now it looks more like airline capacity: peak hours, spot prices, rationing, and three-year lock-ins. That changes how companies should buy AI. If you are running a business on an AI vendor right now, understand that your capacity is not guaranteed and your pricing is not stable. The companies that lock in multi-year committed capacity are going to have a material advantage over the ones paying spot. Also pay attention to who benefits: CoreWeave, Nvidia, and the power companies quietly becoming AI infrastructure businesses. The software layer is getting all the headlines, but the money is moving down the stack to anyone who owns a GPU, a data center, or a megawatt of electricity.
Read Original Source