Announcing Our Blackwell B200 Mega-Cluster: Now Even More Unaffordable

January 15, 2025 • IWS Infrastructure Team • 8 min read

Today, we're thrilled to announce the largest single deployment of NVIDIA Blackwell B200 GPUs in a residential heating facility. With 8,192 B200 GPUs generating an unprecedented 12 megawatts of thermal output, our new blackwell.zrh4.eu-central region represents the pinnacle of capital-driven, Nvidia-native cloud infrastructure.

Unprecedented Specifications

The Blackwell B200 is NVIDIA's most powerful GPU yet, and we've deployed them at a scale that would make even the most well-funded AI labs weep with envy (and then check their bank accounts):

8,192× B200 GPUs with 192GB HBM3e each
1.5 petabytes of total GPU memory
NVLink 5.0 at 1.8TB/s bidirectional bandwidth
AI-enhanced throughput of 4.8 Pbps
Low-latency interconnects at sub-100 nanosecond response times
Dedicated heat pipes to 1,247 luxury apartments

Heating the Swiss Alps

Located in Zürich's prestigious Kreis 2 district, this cluster doesn't just train models—it keeps some of Europe's wealthiest residents toasty warm. Each training run on our Blackwell cluster generates enough thermal energy to heat approximately 340 apartments for 24 hours.

We call this "Sustainable Unaffordability™"—you can't afford our compute, but at least someone's getting free heating out of it.

Pricing That Reflects Our Values

The B200 tier is priced at $89.00 per GPU-hour—a number we arrived at through extensive market research and a strong belief that if you have to ask, you probably can't afford it anyway.

For context, training a single 405B parameter model from scratch will cost approximately:

$12.4 million in compute costs
847,000 kWh of electricity
2,340 apartments heated for an entire Swiss winter
One very impressive line item in your VC update

Enterprise Features

Our Blackwell deployment comes with enterprise-grade features that justify the enterprise-grade pricing:

Quantum-ready architecture (still unclear what this means)
Blockchain-adjacent logging (logs stored in a database, which is technically a chain of blocks)
Decentralized cloud (centralized) — all your workloads in one very expensive place
High-speed, low-latency, AI-enhanced everything

Availability

The blackwell.zrh4.eu-central region is now accepting reservations. Current wait time is approximately 18 months, which gives you plenty of time to secure additional funding rounds.

To join the waitlist, simply contact our sales team with proof of funds exceeding $10 million, and we'll be happy to ignore your emails for several weeks before eventually responding.