Our stack at a glance
SuperPOD-Class Compute
To power the next phase of Causal AI research, Alembic is deploying a NVIDIA SuperPOD cluster with Equinix in San Jose, marking our second privately owned supercomputer. Built on the NVIDIA AI Enterprise software suite, this DGX system supports Alembic’s continuous-learning spiking neural networks and spatio-temporal graph construction, enabling real-time causal inference at scale.
With multi-exaflop compute, dual-site redundancy, and a fully optimized CUDA-X and cuGraph stack, Alembic’s researchers can push the boundaries of dynamic causal graph analysis without the latency or resource constraints of shared infrastructure.
The result: near real-time insights for partners like Accenture, delivered by a private AI fleet purpose-built for scientific discovery.
8× GB200 NVL72 racks: Each NVL72 forms a 72-GPU NVLink® domain (36 Grace CPUs + 72 Blackwell GPUs) acting like one massive GPU for breakthrough performance.
Equinix Fabric connectivity: Software-defined, low-latency interconnection links our AI core with customers, partners, and clouds, globally.
WWT collaboration: World Wide Technology assisted design, procurement, and deployment. Their Equinix-hosted AI Proving Ground validated our solutions.
Privacy, residency, and governance by design
No PII collected or stored.
Alembic focuses on anonymous, aggregate signals and rigorous time-series reconstruction to align with enterprise privacy standards.
Data stays in a secure Equinix environment.
We maintain strict data residency and governance controls; connectivity is via private, software-defined links (Equinix Fabric).
FAQ
Is this an internal cloud or an AI “factory”?
Both. It’s a private, DGX SuperPOD with GB200 systems designed to queue, schedule, and isolate workloads while scaling to massive jobs. NVIDIA Docs
Why not just rent GPUs?
Predictable cost, better data control, lower latency to customers, and the ability to tune for our causal workloads — all reasons we built private. Our messaging emphasizes speed, causation (not correlation), and enterprise ready outputs. We use highly optimized custom kernels to squeeze the highest performance possible for the NVIDIA GPUs. We require the flexibility that owning our own hardware gives us.
How much data can Alembic handle?
We regularly process billions of rows and have run implementations exceeding 100B rows annually, driven by our GPU cluster.
Build the private AI backbone that proves what drives revenue for the world’s biggest brands.
Go to Careers >Learn about the history of Alembic, Created in California