Summaries like this, in your inbox every morning.
Sign up free →In February 2025, NVIDIA stated that long-thinking reasoning AI can require 100x more compute per task than simple one-shot answers. By May 2025, the company noted these new AI agents required hundreds to thousands more tokens per task, with Microsoft seeing a five-fold increase in tokens processed in a single quarter.
Blackwell architecture deployments were earmarked for inference (the step where an AI produces an answer) rather than training—a first for a new chip generation. This reflected two compounding forces: companies shipping AI to millions of users, and each query burning far more compute due to reasoning workloads.
Despite a $8 billion revenue loss from new China export controls announced in May 2025, the underlying inference-driven demand was strong enough that the business continued growing, yet options market implied volatility had eased to the 19th percentile of its one-year range, signaling trader complacency.
No comments yet. Be the first to share your thoughts!
Log in to join the discussion





Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
5 minutes a day. The AI essentials.
200+ sources · Email / LINE / Slack