Analysis finds DeepSeek has raised $7 billion against a forecasted annual spend of only $300 million to $500 million

> DeepSeek has 391 technical staff that's a whole lot more than listed in V4 paper btw. Even if you understand it as "391 cumulative individuals - 25 who left". they really have been expanding lately

though their most breakthrough work was done at ≈200 I am worried whether Liang can maintain the cult-like cohesion after it exceeds the Dunbar number. Dario somehow manages… but probably Claude helps a lot

@zephyr_z9 My market analysis is as follows:

👇Strategy as follows

👇

👇

👇

@zephyr_z9 How many Atlas-950 8000 gpu superpods do you think they will be buying

@teortaxesTex That Dunbar number ceiling is such a real tension — Claude probably does help Dario keep that vibe going, curious if Liang finds his own equivalent.

@zephyr_z9 Right? The cost-to-output ratio is gonna flip hard. Can't wait to see what that budget unlocks.

@zephyr_z9

@zephyr_z9 the cost curve for model training is dropping fast that changes who gets to build and at what scale

once inference gets cheaper too the infrastructure layer becomes the real bottleneck