Pinned post
Bart de Witte
Verified account This account is verified. Learn more@OpenMedFuture
🇧🇪 lived in 🇨🇭🇦🇹🇸🇰 now 🇩🇪 Med AI Democratisation 25+ yrs Digital Health leadership, father ex-IBM ex-SAP @isaree_ai blog.hippoai.org eu/acc
2.5K Following 7.4K Followers 15K Posts
Latest posts 5
Jensen wasn’t giving away Nvidia’s edge with open models.
He was protecting it.
Classic move: commoditize the complement. Keep the scarce thing (GPUs + CUDA) expensive and locked in, make everything else free so more people need the chips.
That’s why he greenlit approx $26B for open models over five years.
A CFO running the show thinks in quarters. Guidance, margins, attach rates. That mindset is fine for earnings calls.
It’s terrible for multi-year bets that deliberately make software free so hardware demand explodes later.
Open source isn’t the risk. Short-term financial optics overriding the strategy is.
He was protecting it.
Classic move: commoditize the complement. Keep the scarce thing (GPUs + CUDA) expensive and locked in, make everything else free so more people need the chips.
That’s why he greenlit approx $26B for open models over five years.
A CFO running the show thinks in quarters. Guidance, margins, attach rates. That mindset is fine for earnings calls.
It’s terrible for multi-year bets that deliberately make software free so hardware demand explodes later.
Open source isn’t the risk. Short-term financial optics overriding the strategy is.
Well done Charlotte!
Azadi is the global best selling techno track right now. This is mental. Thank you all for your love and support 💕 go.kntxt.be/035View quoted post on X
Centralized agentic AI from OpenAI and Anthropic is starting to look a lot like AOL and CompuServe trying to build a private, monopolized internet. Did that work?
Between 2021 and 2024, the best available intelligence per gigabyte of model size rose 187-fold. That is a doubling every four to five months. Moore's law doubled every eighteen to twenty-four months. Model compression has been running three to four times faster than Moore's law. Put differently: GPT-3-level capability on the standard knowledge benchmark needed 350 gigabytes in 2021. By 2025, the same capability fit in 0.4 gigabytes. That is not an improvement of twenty percent. It is a reduction of 875 times in under five years.