ThursdAI > Guest Directory > Chris Alexiuk > September 10, 2026
← All guests  —  NVIDIA / DeepSeek V4.1 Flash

Chris Alexiuk

NVIDIA · DeepSeek V4.1 Flash

Chris Alexiuk joined Alex Volkov on ThursdAI to break down DeepSeek V4.1 Flash. The NVIDIA product research engineer walked through what makes the release interesting under the hood: the disaggregated serving design DeepSeek baked into the model rather than bolting on after the fact, and why the KV cache is the quiet secret behind what inference actually costs.

Full segment

Chris on ThursdAI

Unlisted · 28:52 · download from DescriptWatch on Descript / download →
Bangers

They baked disaggregation in

DeepSeek didn't bolt disaggregated serving on afterwards — V4.1 Flash was designed with it baked in.

Unlisted · download from DescriptWatch on Descript / download →

KV cache is the cost secret

The KV cache is the quiet lever behind what serving a model like V4.1 Flash actually costs.

Unlisted · download from DescriptWatch on Descript / download →
Share it

Suggested posts

𝕏 Post
DeepSeek V4.1 Flash is a really interesting release once you look under the hood. I joined @altryne on ThursdAI to break it down — the disaggregated serving design they baked in, and why the KV cache is the real story on inference cost. Full segment: https://thursdai.news/guests/llm_wizard/sep-10-2026
LinkedIn
I joined Alex Volkov on ThursdAI this week to break down DeepSeek V4.1 Flash. We went under the hood of the release: the disaggregated serving design DeepSeek baked into the model rather than bolting on afterwards, and why the KV cache is the quiet lever behind what inference actually costs. Full segment and clips: https://thursdai.news/guests/llm_wizard/sep-10-2026 ThursdAI goes live every Thursday and is one of the highest signal AI shows around. Worth a follow.

ThursdAI — The weekly AI podcast, hosted by Alex Volkov. Every Thursday, live.

Subscribe Free →