From a healthcare race that’s partly defensive and partly IPO positioning to a departure that exposes a fundamental disagreement. Here’s what news mattered this week.
Jan 17, 2026
∙ Paid
Every week I spend hours tracking what’s actually happening in AI—not the headlines, but the strategic shifts underneath them. The past two weeks had over 200 announcements. Five of them matter: a healthcare race that’s partly defensive and partly about IPO positioning, a departure that exposes a fundamental disagreement about where AI is headed, a physical AI convergence that’s forming the first real flywheel, a training data scramble that reveals what’s actually scarce, and a capability tipping point that just produced a browser in a week. This is what’s worth your attention and what it tells us …
From a healthcare race that’s partly defensive and partly IPO positioning to a departure that exposes a fundamental disagreement. Here’s what news mattered this week.
Jan 17, 2026
∙ Paid
Every week I spend hours tracking what’s actually happening in AI—not the headlines, but the strategic shifts underneath them. The past two weeks had over 200 announcements. Five of them matter: a healthcare race that’s partly defensive and partly about IPO positioning, a departure that exposes a fundamental disagreement about where AI is headed, a physical AI convergence that’s forming the first real flywheel, a training data scramble that reveals what’s actually scarce, and a capability tipping point that just produced a browser in a week. This is what’s worth your attention and what it tells us about where this is all going.
Two weeks ago, OpenAI announced a healthcare product. Five days later, Anthropic announced one too. The same week, Nvidia unveiled its next-generation chip architecture at CES. Google DeepMind announced a partnership with Boston Dynamics. Yann LeCun departed Meta with a parting shot that LLMs are “a dead end for superintelligence.” And buried in the noise, OpenAI started asking contractors to upload real work from their previous jobs.
Read superficially, these are separate stories. Read carefully, they’re all the same story: the AI industry is discovering what actually compounds, and it’s not what most people think.
Top Stories this week:
OpenAI and Anthropic both launched healthcare products within five days—partly defensive, partly IPO positioning
Yann LeCun departed Meta, called LLMs “a dead end for superintelligence,” and revealed Meta “fudged” their Llama 4 benchmarks
Nvidia launched the Rubin platform at CES; Google DeepMind partnered with Boston Dynamics to put Gemini in Atlas robots
OpenAI and Handshake AI are asking contractors to upload real work from previous employers—the easy training data is exhausted
Cursor built a browser from scratch using GPT-5.2 in one week: 3 million lines of code, and it kind of works
Lots to cover so let’s get into it!
Listen to this episode with a 7-day free trial
Subscribe to Nate’s Substack to listen to this post and get 7 days of free access to the full post archives.