Anjney Midha (AMP) revealed xAI runs at sub-10% MFU while best-in-class labs hit 60–70%, proving that scaling AI is a systems problem, not just a GPU count problem.

AMP is building a compute grid that pools FLOPs like megawatts, aiming to make GPU utilization as efficient and tradable as electrical power.

Stephen Balaban (Lambda) says AI compute is massively underbuilt; a 2023 GPU now leases for more than its original ship price, proving GPU compute never became a commodity.

Lambda's financing stack uses off-take agreements and SPVs to fund gigawatt-scale factories, turning compute into an investible asset class with institutional logic.

Lip Bu Tan (Intel) outlined his turnaround plan: a collaboration with Elon Musk on Terafab, a roadmap centered on CPUs for agentic AI inference, and fresh investments from Nvidia and Softbank.

He described Intel's culture shift from a 'legacy spreadsheet' company to one that makes fast decisions and prioritizes customer satisfaction.

Kyle Daigle (GitHub) says commits will jump from 1 billion to 14 billion in 2026; agents already created 17 million pull requests in March alone.

GitHub is adding agentic code review, model routers that auto-select the optimal model per task, and shifting to usage-based pricing to support agent-native workflows.

Claude Code can now generate live, shareable artifacts—PR walkthroughs, dashboards, incident pages—that auto-update from session context, turning agent work into collaborative web views.

Artifacts are private to the organization, include version history, and require no extra infrastructure; a single link lets teams follow an investigation as it unfolds.