Conference Presentation, Keynote, Product Demonstration
Jack Collier, Chief Growth & Marketing Manager, IO.Net: Why Open Models Need Open Infrastructure
- Global compute supply is projected to require doubling every six months to match skyrocketing demand, with 65% of worldwide compute expected to be controlled by just three major providers.
- The market is forecast to remain 70% reliant on closed-source APIs for current usage, though 83% of developers express a preference for open-source models, which are anticipated to gain share in specific verticals and offer costs near $3.50 per million output tokens compared to up to $15 for closed alternatives.
- IO.net aims to expand its active GPU network from 10,000 to over 1,000 GPUs sourced from 138 countries, supporting on-demand serving in Ray, Mega Ray, bare metal, and container environments while maintaining intelligence infrastructure usage exceeding one billion tokens weekly.
- The platform plans to introduce training-as-a-service and fine-tuning capabilities later this week, enabling connections to any Hugging Face open-source model with datasets, and will offer free training for the first five sessions under an alpha program.
- Future ecosystem growth includes the emergence of permissionless services for fine-tuning and inference, alongside startup incentives such as cloud credits and hackathon access via QR code to support on-demand model building without hyperscaler gating.