Alembic cuts AI runtime from 5 days to 20 minutes | Equinix customer story
~360× faster
Algorithm runtime, down from 5 days per customer1 trillion
edge calculations computed per customer, per runMore data than answers, until causal AI connects it
Fortune 500 companies generate more data than most teams know what to do with. Alembic's causal AI does not just help customers find a single answer in that data, it surfaces the chain reactions behind it, focusing on why something happened rather than only what happened. Delivering that depth of analysis at enterprise scale demands more than raw compute power, it demands infrastructure built specifically for the problem.
We don't just partner with Equinix to do things that are hard. We literally partner with Equinix to do things that have never been done before, ever.
Tomás Puig Founder and Chief Executive Officer, Alembic Technologies
Custom hardware, co-designed around the math
Alembic's customers do not just need help finding a needle in a haystack, they need to understand every lever and chain reaction driving their business, and they need to move at the speed of the market. Running that kind of causal AI at enterprise scale eventually outgrew Alembic's existing H100 cluster, with the most demanding customer runs taking days rather than minutes.
Alembic worked directly with Equinix to co-design and co-develop its embedded machines and solvers, marrying custom hardware to its mathematics rather than treating infrastructure as an afterthought. That partnership led to the deployment of a Blackwell SuperPOD, replacing the aging H100 cluster and unlocking a step change in what Alembic's models can process.
Secure GPU capacity from anywhere in the world
Operating at this scale means Alembic sometimes needs to requisition GPU capacity from across the globe in the middle of a training job, over a secure encrypted line. For customers such as Fortune 500 banks, that also means working within strict secure interconnect requirements that Alembic says are not even available through other providers.
An algorithm took five days to complete for a single customer. Five weeks ago, we took delivery of our brand new Blackwell SuperPOD and that customer ran in twenty minutes.
Tomás Puig Founder and Chief Executive Officer, Alembic Technologies
From a five-day wait to a 20-minute answer
The Blackwell SuperPOD deployment turned Alembic's most demanding workload from a five-day wait into a 20-minute answer, and the secure interconnection underneath it lets Alembic pull GPU capacity from anywhere in the world mid-job. That combination of custom hardware, global reach and Fortune 500-grade security is what lets Alembic keep pushing causal AI past what its customers thought was possible. Talk to an Equinix expert about building private AI infrastructure around your own workloads.
