The vision
Sovereign AI safety compute
AI safety compute should be owned by, and answerable to, the safety organizations using it.
Why compute
Compute is the indispensable resource of AI safety work.
Powerful AI models are released weekly, and AI safety has to keep pace. We believe that securing sovereign compute controlled by AI safety organizations is one of the most effective ways to safeguard the frontier of AI.
Today, most AI safety organizations don’t control the compute they need to do their work. Pacific Compute was founded to put these safety organizations in control of their own compute resources.
Who owns it
The safety organizations doing the work.
Pacific Compute is a not-for-profit organization whose members are organizations working to safeguard our collective AI future. Pacific supports AI safety work with safety-only compute for research, model evaluations and interpretability. We also provide support for labs, evaluators and academic partners through a small, dedicated engineering team.
We negotiate with neoclouds, labs and other compute providers to ensure AI safety has access to affordable, safe and accessible compute resources on demand. By working to secure compute access, we help maintain researcher independence, autonomy and security.
What it should cost
Toward the price of electricity.
We believe that the price of compute for AI safety work should converge on the price of electricity. By reducing the prices AI safety orgs pay for compute, we multiply the effective budgets of these organizations, helping them achieve their missions. By providing dedicated AI safety compute to lab partners, we help ensure that GPU access for safety is never competing with capability work.
Sovereign safety clusters can save up to 40% compared with today’s on-demand prices.
How to pay for it
Organize the market
We rent compute at a discount today, and plan to move to long-term contracts and finally full ownership. By aggregating demand for safety compute, we are able to offer AI safety organizations better prices for compute.
Our path to fully owned, sovereign compute runs through projects where financial or political factors have prevented a data center from becoming operational. We’re currently exploring how these sites can be transformed into safety-only clusters alongside community, lab and finance partners.
Where
We have started with a cluster in Singapore for open models; others in the U.S. and the EU will follow.
The Singapore clusterEnsuring safety-only work
We use privacy-respecting monitoring systems to ensure our clusters are always working towards our goals in AI safety. The world of compute verification is currently being rapidly developed, and we’ve partnered with academics, labs and evaluation organizations to make our cluster a useful testing environment for new ideas such as telemetry-based workload classification, kernel metadata, and other compute verification research directions.
Put compute to work.
Bring a model, an evaluation, or the funding to make it happen.