Together AI has partnered with Pearl Research Labs to tackle the rising cost of AI inference by tying GPU workloads to crypto rewards. The two firms plan to use a system called Proof of Useful Work — a variation on traditional mining that pays out tokens for computations that actually train or run models. The idea is to let GPU providers earn crypto while doing inference, offsetting what's normally a pure expense.
How Proof of Useful Work Works
Proof of Useful Work is exactly what it sounds like: instead of hashing random numbers for a block reward, miners submit useful computations — in this case, AI inference jobs — that get verified and rewarded. Pearl Research Labs brings the underlying protocol, while Together AI contributes its inference infrastructure. The partnership's stated goal is to reduce AI inference costs by letting providers recoup some of their hardware overhead through token rewards.
Why Inference Costs Are the Target
Running large language models and generative AI is expensive, especially at scale. Most GPU time is paid for up front, with no way to earn back the compute cost. By layering a crypto reward on top, Together AI and Pearl Research Labs are betting that the token incentive can make inference cheaper — or at least more sustainable — for the people running the machines. The model effectively turns a cost center into a potential revenue stream.
What Comes Next
Neither company has released a timeline for when the Proof of Useful Work integration will go live. The partnership was announced this week, and both sides are in the early stages of technical collaboration. How the reward rate will be set — and how it balances against the value of the underlying token — isn't public yet. That math will determine whether the system actually cuts costs or just adds complexity.
