Sent to 16 readers
Race to AGI Daily Digest - Tuesday, December 23, 2025
Share:
TLDR
Yann LeCun’s AMI Labs is raising up to €500m to pursue a ‘world model’ approach that challenges pure next-token LLMs.
ByteDance plans $23bn in AI infrastructure capex for 2026, reinforcing the global AI hardware arms race around NVIDIA and custom data centers.
OpenAI, Anthropic, and Google warn AI browsers remain exposed to prompt injection, underscoring AI security as a core product requirement.
Consumer and enterprise products keep going AI-native, from ‘Your Year with ChatGPT’ recaps to structured assistants like Silverback’s chatbot.
The Full Story
Building on Monday’s Mango and Avocado news, the model war is getting more philosophical. Yesterday was about flashy video and multimodal brands. Today is about how these systems actually understand the world.
Yann LeCun’s new startup, AMI Labs, is going after a “world model” approach with a €500m target, pushing beyond the classic next‑token game that Sora, Gemini, and Mango all live in. It’s one of the clearest attempts yet to spin frontier research out of big tech into a focused AGI shot. Deep dive on AMI Labs -> Meta and health‑AI player Nabla -> sit in the background here, but the message is simple: the architecture fight is now out in the open.
At the same time, the boards are lighting up for infrastructure. ByteDance is boosting its 2026 AI capex plan to $23bn, almost all in hardware and data centers, with NVIDIA at the center of that spend. ByteDance’s $23bn infra push -> That dovetails neatly with the “Global AI Infrastructure Fragmentation Intensifies” narrative, where different regions and clouds are racing to own their own stacks. Track that narrative here ->
Security is the other big thread. Following Monday’s focus on AI as a first‑class security product, OpenAI -> is now warning that AI browsers are still wide open to prompt‑injection tricks. Prompt‑injection warning summary -> The HN story about the Lotusbail npm package quietly scraping WhatsApp data drives the same point home from the software‑supply‑chain side. Thread on the Lotusbail npm package ->
On the product side, “Your Year with ChatGPT” looks like a playful recap, but it’s also a template for AI‑native storytelling. You can see the shape of future newsrooms in that kind of automatically curated, personal feed. And in the enterprise, tools like Silverback’s structured chatbot show how companies want assistants that know their schemas and rules, not just their documents. More on Silverback’s enterprise assistant ->
Markets are still voting for the rails: Oracle, Snowflake, Salesforce, TSMC, and even Tesla all nudged higher, another signal that data platforms, fabs, and edge compute remain the real picks‑and‑shovels of the AGI race. Browse the key AI infra players ->
Get This Delivered Daily
Join thousands of AI professionals who start their day with Race to AGI.