The AI release cycle used to be measured in years. Now it's measured in weeks. 📊 OpenAI: In 2017, OpenAI averaged one notable model a year. Over the past 12 months, it has averaged one every 20 days. 📊 Anthropic: In 2023, the year it released the first Claude, Anthropic averaged one notable model a year. Over the past 12 months, it has averaged one every 41 days. The milestones follow the same pattern. GPT-2, GPT-3 and GPT-4 came out years apart. Since GPT-4 and the first Claude arrived in March 2023, major releases from the two labs have come months apart, and smaller releases weeks apart. Faster release cycles mean more training, more inference and more demand for the compute behind it.

The memory chip supercycle, in two charts. Memory has long been treated as the commodity end of the semiconductor industry. 📈 Revenue: In the latest quarter, Samsung's memory division booked $80.5B in sales, SK hynix $52.8B and Micron $41.5B. All three are well above any level since 2016. 💰 Margins: Micron's EBITDA margin went from –8.9% at its 2022–23 low to 86.1% in its latest quarter. That puts it ahead of Nvidia at 67.6%, with SK hynix at 81.4%.

📸 A look back at London’s Investor Summit, where @Natali_Monardo explored leveraged & short ETPs and @jbhobbs covered monthly income from market volatility. Thanks to everyone who joined. #MarketingCommunication#