We build data centers 10x faster

We deploy 1-10MW distributed data centers in existing industrial buildings with surplus power, which is 10+ times faster than standard data center deployment. We call our network of data centers the Rewbi Edge Inference Grid, which is designed for secure, reliable, and fast access to compute. Using our vertically integrated approach, our pilot B300 site was deployed in 39 days and is now live. 

I’ve spent the past decade in energy infrastructure investment, including working with Stem’s ~30GW portfolio of powerplants. Based on my experience, I have realized that the fastest way to get power is by leveraging sites that already have a surplus.

Our Pilot B300s went live in 39 days

Building a new data center the traditional way takes 2-10 years to start from scratch. Instead, our pilot:

• Had power online in 29 days, compared to a 700+ day standard (24x faster)

• Set up the site in 34 days

• Installed the hardware on site in 37 days (10x faster)

• Produced our first tokens in 39 days

Our projects are easier to finance and carry significantly less political risk because of their size and geographic diversification. Each data center has nearly uncorrelated power risk because outages almost always impact a 20 mile diameter or less, increasing minimum compute availability for inference.

Our solution to the power problem

The wait time for new power is 2-10 years. The US grid only adds about 55GW of new capacity each year, and Texas alone already has a 474GW large load queue. For inference (as much as 90% of AI compute demand), there is no reason to wait in line for permission to build a large site.

Instead, we’re focused on developing the 26,000 existing sites with on average 2.7MW of surplus power remaining (70GW). We’re actively developing 3 sites (~20MW total) to serve approximately unbounded customer demand. Are you going to be a part of our next build? You can start your reservation at https://www.rewbi.com/compute