Articles
warpsim: a microscope for PTX kernels
We run PTX kernels on the CPU instead of the GPU: a deterministic interpreter turns each kernel launch into a replayable trace of every thread, register write and memory access.
Proving Kernels Correct Instead of Testing Them
We formally prove optimized ML kernels correct instead of testing them: symbolic execution shows a FlashAttention tile is equivalent to naive attention at d=64
STFS: Stop Making Every Worker Hammer S3
STFS turns repeated ML training reads into cluster-local cache hits instead of repeated S3 traffic: 97% less origin load, sub-200ms p95 reads after the first epoch.
Introducing SF Tensor
If we succeed, the world we enable will allow growth across the entire economy to accelerate, by letting enterprises grow AI on their own terms.
Tensor Cloud Launch
We're launching the private beta of Tensor Cloud — a cloud-agnostic interface that lets you run jobs across different providers without dealing with the messy details of each one.
Introducing The San Francisco Tensor Company
At the San Francisco Tensor Company, we believe that the next great wave of computing won't be defined by a single chip, a single cloud, or a single framework.
Train the models only you can build. One stack for enterprise post-training and frontier pre-training.
All System Operational© 2026 San Francisco Tensor Company
