Just launched DeepSeek, 2 more coming and deploying thousands of Blackwell GPUs to increase capacity!
What model should we launch next?
DM if you want free credits (no promises) - state your usecase!
May I interrupt your endless stream of posts about Jensen to notify you about us revenue growth surpassing 200M?
Thank you for your attention to this matter.
Downscale medical imaging to fit in GPU memory and the model loses the detail that catches faint findings.
@epsilonhealth built its next-gen radiology AI models on a dedicated NVIDIA Blackwell Ultra cluster on Verda, now running 5x more experiments/week and training 3x faster.
Full story: https://t.co/IXECCwS03O
Google just ruined OpenAI's dev day lmao
Gemini 4 Pro leaks absolutely dominates both Anthropic and OpenAI
This means Fable 5.5 and GPT 6.1 has to price match Gemini 4 Pro now IPO DELAYED 🤣
We’ve been working with the @vllm_project team to accelerate open-source AI.
Thanks to the new DSpark trained on Verda’s GB300 NVL72, Kimi K3 just got ~4x faster in vLLM: ~110 to ~435 tok/s per user on math reasoning.
Full technical report: https://t.co/jIHJbOisN2
Today we are announcing $189M in new funding and becoming Europe’s latest unicorn.
This funding accelerates what we’ve been doing from day 0:
Building the most reliable and performant AI infrastructure for our customers.
Read more in the thread 🧵
EU will not be developing Frontier AI - this is how I read the statement from the president @vonderleyen "But the next leg of this race is not only about frontier models". The next leg is exactly about the frontier! Giving it up will not only hurt EU economy but push us in the dependence of Chinese AI.
I have been talking to many leading experts on AI in EU in recent months, and I do not believe that any of them was consulted in the preparation of SOTEU - we simply disagree with the view presented in this speech. Unfortunately, the EU strategy on AI does not seem to be prepared based on any technological foresight. Let me briefly give 3 reasons. I will be hated by disagreeing but AI is really the key aspect of our future and the start of the "II Cold War" between USA and China is visibly showing this.
1) The HuggingFace incident shows that EU is defenseless in the case of such attacks, i.e., HuggingFace had to retool to Chinese models. The reasonable reaction to these incidents shall be speed-up the development of AI in EU. There will be more such incidents and wee need our models to be able to defend ourselves. Winning this cyber-war will not be possible without frontier AI.
2) Industrial AI is the way to go, and tailor-made model trained on our data are already happening. The problem is that our industry is rolling out in large scale internal models based on Chinese open source solutions. I have been talking to many companies right not and this is what they are doing. Industrial AI needs frontier - running the a company is a hard task, so we need the smartest AI to do it. Not having our own frontier models already makes us dependent on Chinese technology, and will make our companies less competitive.
3) Navier-Stokes problem has been solved using AI. It is clear now that scientific progress and innovation in the next "leg" will driven by AI. This progress is, however, driven by access to internal models. None of us in EU could have prompted these models to solve these problems open for tens of years that the best mathematicians tried to solve. Not having our own frontier will hurt innovations in EU... which anyway are lagging.
In my opinion the reasonable reaction to the world state of AI shall be "EU shall speed up development of AI". In particular this is what I was telling today at TVP World: https://t.co/CbCeOlXPpe.
EU will not be developing Frontier AI - this is how I read the statement from the president @vonderleyen "But the next leg of this race is not only about frontier models". The next leg is exactly about the frontier! Giving it up will not only hurt EU economy but push us in the dependence of Chinese AI.
I have been talking to many leading experts on AI in EU in recent months, and I do not believe that any of them was consulted in the preparation of SOTEU - we simply disagree with the view presented in this speech. Unfortunately, the EU strategy on AI does not seem to be prepared based on any technological foresight. Let me briefly give 3 reasons. I will be hated by disagreeing but AI is really the key aspect of our future and the start of the "II Cold War" between USA and China is visibly showing this.
1) The HuggingFace incident shows that EU is defenseless in the case of such attacks, i.e., HuggingFace had to retool to Chinese models. The reasonable reaction to these incidents shall be speed-up the development of AI in EU. There will be more such incidents and wee need our models to be able to defend ourselves. Winning this cyber-war will not be possible without frontier AI.
2) Industrial AI is the way to go, and tailor-made model trained on our data are already happening. The problem is that our industry is rolling out in large scale internal models based on Chinese open source solutions. I have been talking to many companies right not and this is what they are doing. Industrial AI needs frontier - running the a company is a hard task, so we need the smartest AI to do it. Not having our own frontier models already makes us dependent on Chinese technology, and will make our companies less competitive.
3) Navier-Stokes problem has been solved using AI. It is clear now that scientific progress and innovation in the next "leg" will driven by AI. This progress is, however, driven by access to internal models. None of us in EU could have prompted these models to solve these problems open for tens of years that the best mathematicians tried to solve. Not having our own frontier will hurt innovations in EU... which anyway are lagging.
In my opinion the reasonable reaction to the world state of AI shall be "EU shall speed up development of AI". In particular this is what I was telling today at TVP World: https://t.co/CbCeOlXPpe.
Agentic AI is putting the CPU back at the center of infrastructure.
It's why our Founder & CEO @Ruben_Bryon and COO Jorge Santos visited @Arm to meet CEO @renehaas237 and EVP of Cloud AI Mohamed Awad.
The Arm AGI CPU brings the efficient orchestration, data movement, and sustained performance AI-era infrastructure requires, and it's what we're building on, aiming for a fully Arm-native stack from orchestration to inference.
Thanks to Rene, Mohamed, and the Arm team for having us.
Ever wonder something stupid like whether Kimi K3 does better if you give it more active experts? How does work distribute across different experts - ie. are few doing most of the work?
https://t.co/bihZAfRSZo
TLDR;
- No, more experts is not automatically better, assuming this is tied to how the model is trained. Maybe there is some easy gains if compute is cheap enough but at least not by doing it the most naive way.
- Some workloads have higher affinity to certain experts, but really depends on what benchmark you're checking.
- You can't run a benchmark once to make conclusions, which can make testing on large models quite compute intense.
DSpark update: Turns out with a little Speculators+Mooncake, I'm able to scale training on GB300 NVL72!
9 vLLM nodes serve the full GLM 5.2 FP8 verifier -> Mooncake RDMA store -> 6 nodes train the DSpark with FSDP (DP=24). 125k prefill tok/s, 1.5 steps/s, full online training :)