Uploaded December 2025 | Updated September 2026, 1 day ago
Philip Kiely, Lead Developer Advocate at Baseten, explains their role as “the infrastructure behind inference,” powering companies from Cursor to Notion to leading healthcare AI startups. They dive into how BaseTen gets GPUs, loads models, and makes them run fast, and why that deceptively simple promise requires deep engineering.
Topics covered:
• The rise of open-source models and why companies are adopting a blended strategy
• How capability thresholds—not benchmark scores—drive real product innovation
• Why open-source momentum is global, with major breakthroughs coming from China
• How BaseTen treats all cloud GPUs as one unified pool using multi-cluster scheduling
• When startups should shift from pay-per-token to dedicated inference
• The explosion of new modalities: voice, vision, 3D, and beyond
• Surprising use cases and the future of world-model-driven applications
• How enterprise buyers are shifting from fear to optimism about AI adoption
Philip also shares why 2026 will bring massive leaps in model quality, infrastructure, and real-world deployments, and why he’s more excited than ever about where AI is heading.
Philip Kiely, Lead Developer Advocate at Baseten, explains their role as “the infrastructure behind inference,” powering companies from Cursor to Notion to leading healthcare AI startups. They dive into how BaseTen gets GPUs, loads models, and makes them run fast, and why that deceptively simple promise requires deep engineering.
Topics covered:
• The rise of open-source models and why companies are adopting a blended strategy
• How capability thresholds—not benchmark scores—drive real product innovation
• Why open-source momentum is global, with major breakthroughs coming from China
• How BaseTen treats all cloud GPUs as one unified pool using multi-cluster scheduling
• When startups should shift from pay-per-token to dedicated inference
• The explosion of new modalities: voice, vision, 3D, and beyond
• Surprising use cases and the future of world-model-driven applications
• How enterprise buyers are shifting from fear to optimism about AI adoption
Philip also shares why 2026 will bring massive leaps in model quality, infrastructure, and real-world deployments, and why he’s more excited than ever about where AI is heading.
