Comment by 332451b
11 hours ago
DeepSeek claims $38 of revenue per GPU, per hour for Flash V4. That is $332,880 per GPU, per year. This matches their DeepInfer numbers: 25,600 output tokens/s per GPU sold at $0.60 per million -> $55 minus some other overhead to arrive at $38.
For the bigger Anthropic and OpenAI models (and associated higher prices) there are many unknown factors. But presumably they also work hard on optimizations, and it's not so crazy to think they had or still have a lot of wiggle room to adjust prices.
How can their revenue numbers match benchmarks from their inferencing code? Are they really claiming that they sell 100% of their capacity at all times, 24/7? Wouldn't they require a lot of queued batch traffic to achieve that? Or are they defining revenue in some weird way where what they mean is potential revenue?
It's sad how far down you have to scroll to get the real answer. It's because Anthropic margins on inference are absolutely bonkers and it's easy for them to drop prices by 80% and STILL have positive free cash flow
Missed the memo. Whats the proof of positive free cash flow?
Yes, but DeepSeek went absolute goblin-mode on optimization and I would not be shocked to discover that OpenAI and Anthropic were behind on that front.
OpenAI and Anthropic have more compute and the best frontier models to throw at the optimization problems. They have published articles on these efforts. For example, Dots, and Anthropic's upcoming "persistent agent" is because of these optimizations. Both companies often publish articles detailing a lot of their technical efforts. They just get buried among the flood of other news.
Yes, I would not be shocked either.
Sonnet 5.5 is 16x more expensive at $10 per million output tokens. Kimi K3.1 is also $10 per million output tokens, and its number of active parameters is only 6.5x larger than V4.1 Flash (104B compared to 16B). These things are not directly comparable of course, but just to give an idea.
Even if inference of those models needs 16x, 32x, or 48x the number of GPUs compared to V4.1 Flash, still seems like potentially great revenue per GPU.
My friend you are looking at the wrong number.
Its cash flows to the GPU that matter. Not revenue.
They published all/most of their work and it would be unbelievable if Anthropic had not already used that research in their own architecture.
What is the TPS of DeepSeek? All of the LLMs I have tried are slow as molasses. \
I use TensorX (EU hosted) and get 330TPS
I pretty reliably get 200 tokens per second on deep infra.
> $38 of revenue per GPU
its per GPU-cluster, not a single gpu.
> its per GPU-cluster, not a single gpu.
No, it isn't. It's revenue per hour per GPU.
Based on translating their Chinese text on Firefox, it says "per card":
> The current service throughput of a single cluster is 155 trillion tokens per day, and the overall revenue is $38 per hour (at the lowest price) per card.
With Kagi translate (best), it says GPU
> A single cluster currently serves 155 trillion tokens per day, generating overall revenue of $38 per GPU per hour (at off-peak electricity rates).
https://zhuanlan.zhihu.com/p/2095560295930508953
Does Deepseek say how much those Clusters cost?
Don't forget that Deepseek pays salary in RMB which is 1/7 of the value of USD.
Given that they open sourced their inference framework, how realistic would be to buy hardware and become a provider on OpenRouter? These numbers are incredible.