What the top podcasts are saying about DeepSeek.
DeepSeek came up in 0 of 106 episodes (0%) in the four weeks to Oct 11, against 6% in the earlier weeks. A change this size could be chance, so we do not call it a trend.
Share of episodes, four weeks at a time
What the shows said about DeepSeek
Key takeaways from our write-ups that name it, newest first, no more than two from one show. Each links to the full write-up.
Regulated enterprises refuse to deploy Chinese open-weight models like Qwen and DeepSeek due to IP, data protection, and governance barriers, turning to them only under intense cost duress.
Why CFOs Will Force Token Tiering and US Open AI · 20VC with Harry Stebbings · week ending Oct 11DeepSeek founder Liang Wenfeng uses quant fund profits to fund AI research with zero commercial milestones, prioritizing small-team science over quarterly revenue.
Jordan Schneider: China AI Strategy and the 15:1 Compute Gap · TBPN · week ending Sep 6Cost variations between models are extreme: Fable completed the full task for $550, GPT Sol and Grok cost roughly one-tenth of that amount, and DeepSeek delivered at one-twentieth of the price.
DHH on Benchmarking Frontier LLMs and Cross-Model Code Reviews · Lex Fridman Podcast · week ending Aug 30Deepseek AI, a private entity, recently revealed astonishing financial health, including an 85% inference margin and $1 billion in API revenue, making the company cash flow positive.
Deepseek CEO: $1B API, 85% Margins, AGI-First Focus Pays Off · TBPN · week ending Jul 26Large language model (LLM) providers like Anthropic and DeepSeek are actively moving to design or build their own custom AI chips.
Anthropic, DeepSeek Build Chips: Control or Margin Grab? · 20VC with Harry Stebbings · week ending Jul 12Axiom Math's AI achieved a perfect 120 score on the challenging Putnam exam in December 2025, outperforming the best human (110 points) and leading LLM Deepseek (103 points).
Axiom Math's Formal AI Crushes Putnam Exam with Perfect Score · Latent Space · week ending Jun 7Open-source models like DeepSeek V4 Pro, despite their quirks, can be tuned to surpass proprietary giants like Claude Opus 4.7 for demanding coding tasks.
DeepSeek Outperforms Claude Opus: Command Code's AI Repair Logic · Latent Space · week ending Jun 7
Earliest mentions in our archive
The first episode from each show that said it, oldest first. The matched words are marked.
“If you talk to any AI lab in America, they say the thing that's bottlenecking them is compute. There are quotes from the DeepSeek founder, or Qwen leadership or whatever. They say the thing they’re bottlenecked on is compute.”
“…approximately 6 months. Oh, they have them now in Deep Seek 4 for sure. They've got some level. Well,…”
OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide…
“…like you name it. Everything from GPTOSS all the way to um Moonshot model to Deep Seeks um…”
Baseten CEO Tuhin Srivastava on Custom Models, and Building the Inference Cloud
“…easy. Deep Seek is raising a monster 7 billion round at 50 billion valuation making…”
Apple-Intel Chip Deal, U.S Adds 115K Jobs, DeepSeek Eyes $50B Valuation | Diet TBPN
“…literally 25 days ago everybody was talking about how DeepSeek is so good. And there were a lot of people talking…”
⚡️Making DeepSeek v4 outperform Opus 4.7 with Taste — @AhmadAwais , CommandCode.ai
“…uh to be solved And every time we see and we saw it like with deepseek one year ago and like we continuous see…”
Nebius Co-Founder on AI Infrastructure Bubbles | How Price Elastic is Demand for Compute
Most recent mentions
The newest episodes that count toward the chart, no more than two from one show.
“It’s hard as well now because so many of the architecture changes — you look at Kimi, for instance, or DeepSeek — they’re doing these architectural modifications not just with dropping the pre-training loss in mind, but with how the models are going to be used in the real world.”
AI researchers debate how close we are to recursive self-improvement
“…has been quiet. Um this is the CEO of Deep Deep Seek. Um, and then there was this…”
WE'RE BACK: Meta Addiction, Jalapeño, Tim Cook's Last Day, Flocked
“…frontier-level models like Kimmy or DeepSeek and GPT-5 6 all up to 5,000 TPS.”
The Inference Frontier: from 100 to 10,000 tokens per second — Sean Lie, Cerebras CTO
“... which took forever. I forget how long Kimi actually spent on it. And then I also did it with DeepSeek V4 Flash first. And Flash failed the same way that Luna did. It couldn't do it. And then I did it with Pro,…”
DHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux | Lex Fridman Po…
“…of which is for 18 months since the deepseek moment right when the deepseek moment happened the markets fell 40%.”
OpenAI vs Anthropic IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump…
Which shows say it
| Show | Last four weeks | All weeks counted |
|---|---|---|
| Latent Space | 0 | 6 |
| TBPN | 0 | 5 |
| 20VC with Harry Stebbings | 0 | 3 |
| Dwarkesh Podcast | 0 | 3 |
| All-In Podcast | 0 | 1 |
| Lex Fridman Podcast | 0 | 1 |
| No Priors | 0 | 1 |
More trends
Episodes in the newest four weeks:
How we count. An episode counts when its transcript says “DeepSeek” or “Deep Seek” at least twice; plurals count too. We count a shortened re-upload of an episode we already have only once, and we skip the sponsor reads and show promos we can spot. A model marks them in every transcript and our code checks each mark against the words; we also skip the lines around phrases like “brought to you by” and any passage a show reads word for word in three or more episodes. The share is the counted episodes divided by all episodes from the tech, business and science shows we cover in the same four weeks, so a busy month of releases does not raise it by itself. The mix of shows does change from month to month, and one daily show can move a number. Weeks run Monday to Sunday, and the week in progress is left out until it ends. The chart starts with the week of March 30, 2026, when our archive became steady. A few older episodes can still appear under the earliest mentions, and some weeks have gaps. Transcripts are automatic captions, so we also match the spellings they use. The counts cover only the shows we summarize. Download the counts (JSON) or CSV.