Get started

Train frontier ideas, not infrastructure teams.

Take a novel architecture from a local experiment to a frontier run without inheriting one vendor’s software stack. SF Tensor optimizes, verifies, places, and operates training across the best available compute.

Plan a frontier runSee the training stack

Input

Training repo
Novel architecture
Scaling plan
01

Optimize

Generate fast kernels

02

Scale

1 to 10,000 GPUs

03

Recover

Checkpoint and resume

Outcome

Frontier run
Maximum utilization
Portable training

01. GPUs, one workflow

1to0GPUs

From local research to a production pre-training fleet.

02. Lower training cost

Up to0%

With cross-provider placement and workload-specific optimization.

03. Target fleet utilization

0%

Through compiled kernels, topology-aware execution, and resilient input.

The new-lab advantage

Research velocity should not depend on owning a hyperscaler.

The scarce thing is the architecture and the people inventing it. We make compute supply fungible by retargeting workloads across vendors, compiling kernels for the actual topology, and operating the fleet—so a small team can move like a large lab.

Live training efficiency

Fleet utilization

Optimizing
step 18,42096% target

01

Bring the training repo

Keep the framework and model code your researchers already use.

02

Compile for the fleet

Search, generate, and formally verify optimized kernels for each target.

03

Scale across supply

Place the job on the best available mix of hardware, providers, and regions.

04

Run the experiment

Monitor, checkpoint, recover, and return the model—not an infrastructure backlog.

What you get

The systems team behind every training run.

Use SF Tensor as the compiler, performance, distributed systems, and fleet operations team that turns research code into a reliable frontier run.

System layer 01

Novel architectures welcome

Bring unusual layers, kernels, communication patterns, and research frameworks.

Automatic kernel optimization

Generate workload-specific kernels instead of waiting months for hand-tuning.

System layer 02

Proof before production

Use symbolic execution to prove optimized kernels equivalent or return a counterexample.

Cross-vendor portability

Retarget training as new accelerators and lower-cost supply enter the market.

System layer 03

Training-native storage

Keep data and checkpoints close to workers with shared, cluster-local caching.

Resilient fleet operations

Handle placement, observability, checkpoints, node failures, and run recovery.

Where it starts

Give every research direction a credible path to scale.

01

Foundation models

Pre-train dense, MoE, multimodal, and domain foundation models from scratch.

One path from ablation to full-scale training.
02

New architectures

Explore custom attention, state-space, memory, routing, and communication designs.

Infrastructure adapts to the research—not the reverse.
03

Alternative silicon

Use emerging accelerators when they are the best technical or economic fit.

Hardware choice without a permanent software commitment.
Also building for enterprise model teams

Your model is the advantage. Infrastructure shouldn’t be the constraint.

Bring the model, data, and ambition. We’ll optimize the kernels, orchestrate the fleet, and deliver the training outcome.

Plan a training runTalk to an engineer

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

SolutionsHomeEnterprisesAI LabsTensor CloudEmma LangTalk to an Engineer
Company
Blog
Manifesto
CareersWe Are Hiring!
Engineering
SF Tensor Stamp