Companies are hoarding AI compute because of FOMO — and they're sitting on most of it
THE SO WHAT
If Cast AI is right that 95% of GPU capacity is idle, AI capex has already overshot real use cases—your board is paying for fear, not throughput. Treat GPUs like a P&L line item, not a trophy: right-size clusters, sell excess capacity, or you’re subsidizing your more disciplined competitors.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIBond, a new social media platform, wants to use AI to help you kick your doomscrolling habit
AI as a behavioral nudge to get you off-platform flips the engagement model—retention becomes trust, not time-on-site. If Bond shows any traction, incumbents will either copy the “out-of-app” coaching or double down on stickiness; product teams need to decide now which side of that line they’re on.
Applied AIOpenAI enables cost-per-click ads inside ChatGPT, setting bids at between $3 and $5 per click, in addition to CPMs (Digiday)
ChatGPT is now a performance ad surface with $3–$5 CPCs—assistant UX is becoming a commerce funnel, not just a Q&A layer. If you own a brand or marketplace, you now have to budget for conversational ad spend and fight to keep your own assistants from becoming someone else’s paid acquisition channel.
Applied AI'If AI were to take anyone's job, wouldn't it take Elon Musk's job? The richest guy on Earth?' — Take-Two CEO says AI won't get rid of jobs, but allow artists to do 'higher quality work' instead
When a major game publisher frames AI as “higher quality work” for artists, it’s setting the narrative for aggressive tooling rollouts without headcount promises. Creative teams should assume scope and asset volume will expand to fill the automation gap—negotiate for control over pipelines, not job guarantees.
Applied AIFlat-rate AI plans are broken. Blame AI agents
Agentic workflows turned $20/month “all you can eat” AI into a loss leader—usage is now bounded by what agents can do, not what humans can type. If you’re building on third-party models, expect metered, per-agent, or per-workflow pricing and design your unit economics around automation minutes, not seats.