Corsair · Computex 2026 · Taipei

Four AI300 workstations at Computex

For Computex 2026, we put together several demos at Corsair. One used a cluster of four AI300 workstations to explore local models as an alternative to paid frontier-model APIs.

Four Corsair AI300 workstations at Computex 2026, with a display comparing cloud AI spend to local inference. — Four Corsair AI300 workstations and the live comparison display at Computex 2026.

I set up the infrastructure and built the workflow, using kyuz0’s cluster setup guide as a reference.

The demo compared costs and outputs on basic tasks. We saw comparable results from some local models on those tasks. The question was where local models could do the job well enough to make them a practical option.

Running locally avoids per-token API charges, though hardware and electricity still have a cost.