HostedAI Inc., a software platform and marketplace that increases the efficiency of graphics processing unit infrastructure, today announced it has raised $19 million in early-stage funding to run multiple workloads on the same GPUs simultaneously.

By pooling artificial intelligence workloads together on GPU resources, HostedAI takes underused compute and allows cloud operators known as “neoclouds” to achieve up to five times the output on existing infrastructure.

With the AI buildout frenzy becoming the largest infrastructure project in world history, GPU frameworks are still rented at the card, server and rack level. However, many customers only need a small fraction of the available capacity.

“The GPU infrastructure powering AI today is incredibly wasteful,” said Chief Executive Ditlev Bredahl. “Customers are forced to rent GPU as a static resource, because that is the only way operators can sell it — and AI workloads leave about 60% of that GPU idle, on average.”

At the infrastructure operator level, hardware can support much higher levels of use, which means they can make more money on the capital they poured into building data centers and buying servers. This lack of use can be extremely punishing to neocloud providers, which are losing significant amounts of money on idle compute.

HostedAI said it also aims to allow customers to diversify which providers they can reach throughout the market, which is dominated by only a handful of hyperscaler cloud providers in the United States. European companies and public sector organizations seeking to maintain data sovereignty or regional efficiency would have a better alternative at cheaper cost.

The company added it’s federated GPU market would deliver a platform to essentially publish spare capacity and subscribe to capacity from other providers to extend geographic reach.

AI developers access the company’s system with consumption-based pricing, not renting servers. This leads to significantly lower costs for developers and businesses.

HostedAI is entering a maturing landscape of on-demand GPU marketplaces, including Together Computer Inc., Runpod Inc. and Vast.ai Inc. While these incumbents focus on either massive serverless scale or unmanaged peer-to-peer hardware, HostedAI carves out its position by blending commodity pricing with regional availability. This targets the middle layer of developers and businesses who need low-latency and production-grade reliability without paying too much.

The end result essentially provides neoclouds the flexibility to provision idle GPU power that that would otherwise go to waste. HostedAI believes this is a win-win that will optimize that infrastructure and bring in more customers to use that spare capacity. “The GPU market has a waste problem, not a scarcity problem,” added Bredahl.

Creandum led the seed round, with additional investment from People Ventures and Repeat VC.

Image: SiliconANGLE/Microsoft Designer