Sent to 16 readers
Race to AGI Daily Digest - Thursday, January 22, 2026
Share:
TLDR
South Korea passed the first full AI safety law focused on frontier models, pushing AI security from guidance into statute.
Meta’s Superintelligence Lab delivered its first internal frontier models, signaling a serious bid to compete with top AGI labs.
Anthropic is pursuing a $10B+ mega‑round on a reported $9B revenue run rate, deepening the capital concentration around a few AGI players.
Nvidia’s $300m bet on Baseten highlights how inference infrastructure is becoming as strategic as model training.
The Full Story
Following Monday’s Anthropic coworker launch, Tuesday’s AI stock slide, and Wednesday’s huge bet on Humans&, today the frontier race gets a new player: the lawmaker.
On screens, the AI chip complex is suddenly smiling again. Intel, AMD, Arm, Baidu, and Alibaba all ripped higher, helped by the backdrop of Nvidia’s coming H200 exports to China and a market that still believes in the GPU story. You can track those chip‑security tensions in our export narrative here ->.
Now layer on what South Korea just did. The country passed the first comprehensive AI safety law that explicitly targets frontier‑scale models, with OpenAI and Google squarely in mind law recap ->. That plugs straight into our "AI Security Stack Races to Catch Up" storyline: it’s not just tools like Dam Secure and HackerOne anymore, it’s national rules about who can train what, and how.
Meanwhile, the model war at the top is getting sharper. Meta’s new Superintelligence Lab shipped its first internal frontier models, a clear signal that Meta doesn’t plan to sit out the AGI club model details ->. And Anthropic is now chasing a $10B‑plus mega‑round on a reported $9B revenue run rate, with backers like Microsoft, Nvidia, Sequoia, and Lightspeed lining up again funding story ->. If you want to see who’s at the center of this capital vortex, start with Anthropic’s profile here ->.
Building on yesterday’s focus on agentic tools, Nvidia is now backing Baseten with a $300m round to scale AI inference infra deal breakdown ->. That’s the plumbing that keeps enterprise agents like Claude Cowork and Ringg AI’s multilingual voice agents running in production; Baseten itself is worth a look company profile ->.
In the background, Hacker News is debating "How AI destroys institutions" thread ->. So by Thursday, the picture is pretty stark: agents are moving into real workflows, money and GPUs are piling into a few labs and infra players, and now policymakers are racing to put guardrails around whatever those labs build next.
Get This Delivered Daily
Join thousands of AI professionals who start their day with Race to AGI.