0
Applied AI·August 11, 2026·1 min read

IBM bets $240m on cheap, open-source inference to take on the hyperscalers

Share

Inference cost is becoming the competitive front door for enterprise AI, not model bragging rights. If IBM and Together AI can make Blackwell-class inference cheap and open on IBM Cloud, expect procurement teams to start benchmarking vendors on $/token and portability rather than logo alone.

Applied AI

Sources: Nvidia is developing a Nemotron 4 model with 1T+ parameters, up from Nemotron 3 Ultra's 550B parameters but smaller than leading Chinese open models

Nvidia pushing Nemotron 4 past 1T parameters as an open model is about selling more of the stack—GPUs, software, and reference workloads—rather than chasing benchmark glory. If you’re building on open weights, expect a more opinionated Nvidia ecosystem where the model, tooling, and hardware are tightly coupled.