Pinned Tweet
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 🧵
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 🧵
13
8
143
2,568,149
The Nordics are becoming a serious hub for advanced computing, and Verda is part of it. Great to see us named among the AI clouds behind a projected 1 GW of capacity across the region by end of 2027
The Nordics are becoming one of Europe's fastest-growing hubs for AI infrastructure.
From launching Sferical One, a shared AI supercomputer running on two NVIDIA DGX SuperPODs, to AI clouds expecting to bring 1 gigawatt of capacity across the region by 2027, Nordic countries are ope's most impactful technology ecosystems.
Across startups and open models to quantum and more, the Nordics are moving from ambition to application. ➡️ bit.ly/3TD3T2X
0
1
8
227
Verda retweeted
Pro tip: Use PyTorch nested graph breaks RIGHT NOW!!! On my NVIDIA B200 (provided by my friends @verdacloud), I ran a microbenchmark with PyTorch 2.14.1 (CUDA 13.0) and nested_graph_breaks=True. This configuration ran my code 4.53× faster at steady state.
3
2
10
943
Talk 2 of 2 at @PyTorch NA: Rain Jiang and Kevin Li, “WideEP NVL72 with Rust frontend.” Expert-parallel MoE inference across a full NVL72 rack.
🗓️ Oct 21, 12:20 PM
📍 Community Expo
Save it to calendar: events.linuxfoundation.org/pytorch-confer…
🗓️ Oct 21, 12:20 PM
📍 Community Expo
Save it to calendar: events.linuxfoundation.org/pytorch-confer…
0
0
1
139
We’re joining @Supermicro for the @NVIDIAGTC Berlin 2026 keynote watch party. First, a fireside chat on scaling AI with CTO Arturs Polis.
🗓️ Oct 21 10-13:30 CEST
📍Hallesches Haus, Kreuzberg
Free of charge, no GTC pass needed. Waitlist only
🗓️ Oct 21 10-13:30 CEST
📍Hallesches Haus, Kreuzberg
Free of charge, no GTC pass needed. Waitlist only
0
0
4
278
1st talk at @PyTorch: Paul Chang and Erik Schultheis on NVFP4 Training Tips and Tricks,
🗓️ Wed Oct 21 4:20 PM
What they learned training dense LLMs on GB300 at 4-bit: hyperparameters, optimisers, and recipes
Save it in the calendar: events.linuxfoundation.org/pytorch-confer…
🗓️ Wed Oct 21 4:20 PM
What they learned training dense LLMs on GB300 at 4-bit: hyperparameters, optimisers, and recipes
Save it in the calendar: events.linuxfoundation.org/pytorch-confer…
0
1
6
2,355
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.
co/IXECCwS03O
@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.
co/IXECCwS03O
0
3
12
882
Confidential computing on NVIDIA HGX B300 and B200 is now available on Verda.
Your data stays protected while it’s being processed, the stage usually left exposed, with proof you can verify yourself. Up to 99% of full speed in our B300 tests.
Benchmarks and setup: verda.com/blog/confident…
Your data stays protected while it’s being processed, the stage usually left exposed, with proof you can verify yourself. Up to 99% of full speed in our B300 tests.
Benchmarks and setup: verda.com/blog/confident…
0
3
9
303
Verda is heading to @NVIDIAGTC Berlin next week. Booth #3059, Hall 3, Oct 21 and 22.
Before the expo opens, our CTO Arturs Polis will join @Supermicro and @nvidia for a keynote watch party.
Details to follow soon.
Before the expo opens, our CTO Arturs Polis will join @Supermicro and @nvidia for a keynote watch party.
Details to follow soon.
0
0
2
113
Congratulations to @IlhanScheer and @Aleph__Alpha on Kolibri.
78B params, 3.46B active, up to 1M tokens of context, open weights.
Trained by Aleph Alpha on Verda in Europe.
We ran the clusters, built the k8s stack, and contributed to their training efficiency work: verda.com/blog/aleph-alp…
78B params, 3.46B active, up to 1M tokens of context, open weights.
Trained by Aleph Alpha on Verda in Europe.
We ran the clusters, built the k8s stack, and contributed to their training efficiency work: verda.com/blog/aleph-alp…
4
8
76
3,619
What does an additional $189M unlock for Verda customers?
More capacity is part of it. But scaling an AI cloud means much more than adding GPUs.
Here's what we're building next 🧵
More capacity is part of it. But scaling an AI cloud means much more than adding GPUs.
Here's what we're building next 🧵
1
0
1
231
More capacity and the power to run it.
More than 250 MW in 2027, with data center capacity in Finland today and more coming online in Europe, the UK, the US and Asia.
More than 250 MW in 2027, with data center capacity in Finland today and more coming online in Europe, the UK, the US and Asia.
1
0
0
38
A platform ready for production AI.
We're broadening the platform with Object Storage and Managed Kubernetes, while continuing to improve enterprise readiness, security, and developer experience.
Full story: verda.com/blog/what-189m…
We're broadening the platform with Object Storage and Managed Kubernetes, while continuing to improve enterprise readiness, security, and developer experience.
Full story: verda.com/blog/what-189m…
0
0
0
27
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: vllm.ai/blog/2026-09-1…
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: vllm.ai/blog/2026-09-1…
Kimi K3 is a 2.8T-parameter model. We trained a DSpark speculator for it, and the speedup holds up.
Single-stream math reasoning goes from ~110 to ~435 tokens/sec per user. Under concurrent load, up to ~3.5x higher output throughput at matched interactivity. The drafter is a 5B model proposing 8 tokens a step, and on math it gets ~6.4 of them accepted per round.
Training a drafter for a model this big meant going multi-node: Speculators plus a new Mooncake connector streaming hidden states between vLLM inference and training over RDMA. Two GB300 nodes to serve the target, one to train.
vllm.ai/blog/2026-09-1…
4
9
55
7,174
Verda retweeted
Cloud startup Verda raises $189 million in fresh financing, underscoring continued investor appetite for the infrastructure powering AI bloomberg.com/news/articles/…
3
8
27
29,815
It’s no longer just about availability and pricing.
Time to first workload and performance under real load depend on how well hardware, networking, storage, and the serving stack work together.
Time to first workload and performance under real load depend on how well hardware, networking, storage, and the serving stack work together.
1
0
0
507
That’s what we’ve been perfecting for thousands of teams running production AI on Verda.
This funding lets us take that much further.
Learn what’s next:
verda.com/blog/what-189m…
This funding lets us take that much further.
Learn what’s next:
verda.com/blog/what-189m…
0
0
0
380