Parasail raises $32 million to run AI model requests more cheaply

In short: Parasail raised $32 million to grow a cloud service that helps companies run AI model requests, aiming to make those requests faster and cheaper.

What happened

Parasail, a startup led by CEO Mike Henry, announced a $32 million Series A funding round. The company sells computing power for “inference,” which means running an AI model to get an answer, not training it from scratch.

Henry told TechCrunch that Parasail processes about 500 billion “tokens” a day. Tokens are small chunks of text that AI models read and write (like counting words in bite sized pieces). Many developers want lots of tokens quickly, because their apps may send huge numbers of requests to AI models.

Parasail says it lowers costs by pulling together computing power from many places. Some of its graphics chips, called GPUs (specialized chips often used for AI), are its own. But it mainly rents capacity from about 40 data centers across 15 countries, then shifts work around to avoid busy periods, like taking side streets when a highway is jammed.

The company is betting that more businesses will mix and match different AI models. For example, Andreas Stuhlmüller, CEO of research assistant startup Elicit, told TechCrunch his customers use cheaper, openly available models for early screening, then use a more capable model for the final answer.

Why it matters

As more products rely on AI, the cost of running all those requests can add up quickly, especially when software uses AI “agents” (tools that break a job into steps and make many calls). Companies like Parasail are trying to become the utility provider that helps keep those costs down.

Source: TechCrunch AI