Good morning! Ever wake up and think, “You know what my holiday season really needs? AI-generated pics of a tech billionaire cosplaying a firefighter.” Sam Altman just shared an AI holiday thirst trap starring himself, complete with festive lights and suspiciously perfect abs. It was meant to create buzz around OpenAI’s latest image model, specifically built for accuracy and realism. The only problem is the image includes a December calendar… from 2024. Looks like reality is subjective this year 🍌
Also: Google Gemini 3 Flash is fast, cheap and everywhere
China’s “Manhattan Project” secretly built an EUV machine to end US chip dominance. Extreme ultraviolet lithography (EUV) is the tech behind the world’s most advanced chips. Until now, only one Western company controlled it. America banned selling EUV to China for national security reasons. China’s response? Quietly spend years reverse engineering it by hiring former engineers from ASML. Now China has a working prototype. It can’t produce usable chips yet, but experts think they’ll get there by 2028, years ahead of schedule. So much for that export ban.
Google is teaming up with Meta to break Nvidia’s grip on AI chips. Nvidia dominates because its CUDA software makes it easy for devs to build AI models using PyTorch. Google has powerful chips too, called TPUs, but switching to them meant ditching mainstream tools and learning Jax. Now Google and Meta are making TPUs compatible with PyTorch. Competition in tech is alive and well, as long as your name is Google, Meta, or Nvidia.
⚡ Raise your AI game
Go from idea to production-ready AI. Google Cloud Startup technical guide gives you everything you need to build and scale AI agents for maximum efficiency.
Learn how to:
- Build and scale AI agents to maximize your output
- Implement powerful techniques like RAG
- Set up agentic workflows for production, step-by-step
OpenAI’s polyamorous circle expands with a $10 billion offer from Amazon. The deal would push OpenAI’s valuation beyond $500 billion, as Amazon looks to land one of the largest AI customers ever for its Trainium AI chips. OpenAI already committed $38B to Amazon’s cloud, but that was mostly on Nvidia’s hardware. This new partnership would deepen ties, possibly even paving the way for AI-powered shopping on Amazon. At this rate, everyone will soon be paying OpenAI to stay relevant.
Oracle promised OpenAI the moon, then lost funding. Oracle’s planned $10 billion Michigan data center was supposed to power OpenAI’s AI ambitions. Only Blue Owl Capital, the investor behind Oracle’s previous data center projects, won’t fund it. Oracle’s debt is over $100B, the stock is down 45%+, and lease commitments have doubled in 3 months. Oracle needs a new investor quickly, or their $300B commitment to OpenAI might become tech’s largest unpaid bill.
An obscure PhD project may be our only defense against deepfakes. The tech’s called steganography, which secretly hides data inside images. This virtual mark is nearly impossible to remove by scammers, allowing an image to be traced. A PhD grad rolled the tech into a small startup called Steg AI, which was mostly ignored until gen AI images exploded. Suddenly, invisible watermarking was critical for national security. Now major AI companies rely on Steg AI to prove their images aren’t fake. Reality is being propped up by something we can’t even see.
fun stats
🍪 $125 million. Funding raised by Mythic to challenge Nvidia. The AI chip startup claims its architecture has a 100x energy efficiency advantage over top GPUs.
🔮 $4 trillion. Global spending on AI infrastructure by 2030, per Nvidia.
🚖 $100 billion. Valuation Waymo is targeting as it raises $15 billion. The Google startup has fully driverless rides in 5 major cities, with plans to be in over 20 by 2026.
🛎️ $1.6 trillion. Tesla valuation as of mid-December. The robots + robotaxi startup is testing fully driverless rides in 1 city.