🕋 | 🇪🇬 | 👨💻 | ✨️ | Like/RTs ≠ Endorsement | Views my own
Alexandria | Kraków
Joined January 2012
- Tweets568
- Following286
- Followers1.4K
- Likes209K
Abdelrahman retweeted
شيخ الأزهر أحمد الطيب: نشدد على أهمية التقارب بين المسلمين ولمّ شمل الأمة ونتمنى أن نعيش لنرى اليوم الذي يجتمع فيه المسلمون على عدوهم لا على أنفسهم، خلال استقباله وزير خارجية كازاخستان
#مزيد
Abdelrahman retweeted
اكتشفوا إن في ثغرتين في PHP نفسها CVE-2026-109618و CVE-2026-109619.
دول Remote Code Execution وتقييمهم الرسمي 9.2.
والباتش هيبقى لـ PHP 8 بس، أما PHP 5 و 7 خلاص مفيش دعم ليهم.
لسه مفيش تفاصيل تقنية كاملة ولا استغلال معروف، ومفيش حل مؤقت غير التحديث.
Abdelrahman retweeted
DeepSeek just published the price of a token, and it should make you angry.
From their own DeepInfer docs: 1 cluster serves 155 trillion tokens a day, at $38 of revenue per GPU-hour, and thats at off-peak prices. Their API is $0.30 per million input tokens and $1.20 per million output at peak.
Yuchen Jin did the math: even if they rented B300s at $7 an hour, thats about 82% gross margin. Other estimates go higher.
So cheap AI is not charity and it is not "dumping". At these prices serving a frontier-class model is a very profitable business, if your inference stack is good enough.
Now remember every time a US lab told you that intelligence is expensive. That $200 a month is a fair price. That limits exist because compute is scarce.
Part of that is true. Part of it is a business model built on you not knowing what a token actually costs.
DeepSeek just showed everyone the numbers and gave away the engine. Every inference company can now compete on price with the same software. Thats how AI gets cheap for everyone.
“A single cluster currently serves 155 trillion tokens/day, generating $38 in revenue per GPU-hour (at off-peak pricing).”
Huge flex from DeepSeek. Not only did they open-source their inference engine, but they also open-sourced their revenue numbers.
They didn't disclose which GPUs they're using here, but DeepInfer supports both NVIDIA Blackwell and Huawei Ascend.
Even assuming they're renting B300s at $7 per GPU-hour, that's ~82% gross margin.
Abdelrahman retweeted
a few more things i found interesting about DeepInfer:
1. DeepSeek is targeting two very different accelerator stacks from day one: NVIDIA SM100 GPUs + Huawei Ascend 950 NPUs.
2. they didn’t just open-source the inference engine. they also shipped deepseek-recipe, written in Rust with Python bindings, to translate between backends and API formats like Chat Completions, Messages and Responses.
3. one of the numbers floating around from the DeepSeek infra write-up is around $38 per GPU-hour in revenue. i wouldn’t use that to estimate margins, but it shows how much value they’re squeezing out of each deployed accelerator.
4. the interesting comparison for me now is DeepInfer vs vLLM vs SGLang. DeepInfer is purpose-built around DeepSeek’s own model architecture, so even a 10–20% serving efficiency gain would be meaningful at scale.
5. the bigger unlock is for third-party providers. if they can copy even part of DeepSeek’s serving stack, we should see more providers competing on DeepSeek inference instead of all paying the same infra tax.
the first DeepInfer post was about open-sourcing the stack.
this one is why that stack could matter economically.
DeepInfer is probably the most interesting DeepSeek release this week for me:
1. it’s not another model drop. DeepSeek open-sourced the actual inference layer they use to run their models at scale.
2. it already supports both NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs, which makes the Ascend side especially interesting.
3. DeepSeek also released deepseek-recipe, with Rust + Python bindings to connect inference backends to API formats like Chat Completions, Messages and Responses.
4. i’m especially interested how it compares with vLLM and SGLang. DeepInfer is built specifically for DeepSeek models, so there’s a real chance the efficiency gap is meaningful.
5. the bigger thing for me is what this unlocks for the community. custom engines like Strata, gufo, halogen and others can now study DeepSeek’s own optimization choices instead of guessing from the outside.
DeepSeek is starting to open-source more than weights.
they’re opening up the infra stack too.
Abdelrahman retweeted
IMO, today was another DeepSeek moment
Every 2nd/3rd tier inference provider now has a model and a recipe that can generate 90%+ margins on their compute
DeepSeek is also promising future model support and optimizations
This creates a huge incentive for inference providers to serve, as well as push their customers to use more DeepSeek models and utilize the DeepInfer inference framework
DeepSeek gets proliferation and entrenchment, while the inference providers get to make 90% margins
Abdelrahman retweeted
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Abdelrahman retweeted
Today we're launching @VeezaAI (YC F26) - an AI tourist visa agent for people with weak passports.
Yes, you read that right. Weak passports.
If you have a US or EU passport, you book a flight and go. For the rest of the world, every trip starts with forms, bank statements, and fighting for an appointment.
We've suffered through this for 30 years. Those days are coming to an end.
Pick your destination. Veeza AI does the paperwork and books the appointment.
start at: veeza.ai
