I had grand plans this week to do some deep writing – compose my 2025 annual letter, reflect on one year since the “DeepSeek Moment”, and riff on other topical issues in the fast changing AI world. From Claude Code and Clawbot, to the back to back releases of Qwen 3 and Kimi 2.5, there is no shortage of topics. 

But mother nature has a way of its own. Over the weekend, Nashville got pummeled by one of the worst snow plus freezing rain storms in its history, where most of the city lost power.  We lost power on Sunday morning around 5am, so we scrambled to get our little family to a hotel, while entire neighborhoods got paralyzed by falling trees and broken power poles. Below are some pictures of the scenes of destruction around the city. Our neighborhood looks as destroyed as these pictures; I did not take any as I concentrated on driving on the icy road to get our family out of the snowpocalypse.

We are all doing fine now and consider ourselves the lucky ones. There is no way to tell when power will be restored where we live, so we are getting comfortable here in the hotel, along with other parents around us, all juggling life to do what they can for the people they are about most.

Instead of my normal essay-like post, I’ll share a compilation of the previous posts I wrote about DeepSeek over the last year. I admit that it is a bit of a lazy way to “celebrate” the one year anniversary of the “DeepSeek moment”. Though as you go through these old posts, you will see three big themes emerge that illustrate the impact of DeepSeek that, I believe, will last for many years and cycles to come:

  • Pushing open source by default into frontier AI development
  • Using software specs to guide hardware roadmap (especially in China)
  • Pursuing AGI with no business model in its more enduring advantage
No Business Model: DeepSeek’s Enduring Advantage
No other frontier AGI-pilled lab anywhere has this edge
InterconnectedKevin Xu
The Real DeepSeek Moment Just Arrived
Last week, DeepSeek released a new version of its model to much fanfare. Except the fanfare did not come from the model’s performance improvement, hybrid nature (combining both reasoning and non-reasoning into a single model, like GPT-5), or strong agentic capabilities. It came from an unprompted comment from the
InterconnectedKevin Xu
From DeepSeek-R1 to gpt-oss
Finally, some positive-sum competition between the US and China
InterconnectedKevin Xu
China’s Structural Advantage in Open Source AI
Not just my words, Ion Stoica’s too.
InterconnectedKevin Xu
($) DeepSeek Diffusion
It’s about open vs closed, not (just) about the US vs China
InterconnectedKevin Xu
($) DeepSeek’s Three Idiosyncratic Advantages
Internet’s only bilingual newsletter on the intersections of tech, business, investing, geopolitics, and US-Asia relations. New post every week by Kevin Xu (@kevinsxu)
InterconnectedKevin Xu

When mother nature speaks, all you can do is adapt and forge ahead, much like Mr. Market in many ways. And Mr. Market sure had quite the freakout last year, today.

I hope this compilation of my DeepSeek posts is still a useful read, as we wait for power and normalcy to return.