The Information's TITV Google Battles Nvidia with TPUs and OpenAI With Pretraining, Automating Wall Street | Nov 25, 2025
39 snips
Nov 25, 2025 
This discussion features Erin Wu, a technology reporter focused on Google’s TPU strategy, who explains how Google aims to compete with NVIDIA by pitching TPU chips to major players like Meta. Stephanie Palazzolo, an AI reporter, reveals concerns around Google's pretraining advantages with Gemini 3. Finance reporter Miles Kruppa discusses rising debt linked to Oracle’s data centers, while Ron Bencer analyzes the IPO outlook for 2026. Lastly, Chaz Englander, CEO of Model ML, shares insights on automating financial workflows with AI and their recent funding success.
AI Snips
Chapters
Transcript
Episode notes
Pre‑Training Enables Stronger Generalization
- Pre‑training gives models broader generalization power that helps them create novel outputs beyond training data.
- That strength explains Sam Altman's concern after Google's Gemini 3 advances in pre‑training.
Data And Process Drove Google's Pre‑Training Wins
- Google's pre‑training gains likely stem from high‑quality data and disciplined experimental processes.
- Key hires and acquisitions, like bringing in researchers from Character AI, also accelerated improvements.
Huge Project Finance For Oracle Data Centers
- Developers raised at least $65 billion in project finance this year to build data centers for Oracle.
- That surge concentrates risk if syndication fails or if Oracle must shoulder more borrowing itself.