Nvidia licenses Poolside’s “model factory” and staff for $6B: Nvidia agreed to license Poolside’s AI model-building infrastructure (“model factory”) and 109 staff in a $6B deal, while Poolside raises $1B for the remaining company at a $12B valuation. The “license, don’t acquire” structure is drawing wide attention. Newcomer · Bloomberg
Salesforce partners report little real revenue from Agentforce: A partner survey finds partners aren’t generating meaningful revenue from the agent platform Agentforce. Agent commercialization demand is loud, but ecosystem monetization is lagging. The Register
AI data startup Micro1 hits a $500M gross run rate: Amid the AI training boom, data-labeling startup Micro1 reached a $500M annualized gross run rate — evidence that training-data pipeline demand remains strong. TechCrunch
Orbital data-center startup Starcloud raises $250M: Even as launch options dry up, Starcloud secured $250M to expand space-based data centers. TechCrunch
The harness, not the model, is becoming the moat: Nvidia’s demo underscored that the orchestration/inference stack around an AI model — not the model itself — is where durable value sits. As models commoditize, entry barriers are shifting to surrounding infrastructure. TechCrunch
Moving frontier MoE models to local hardware: Post-training GLM-5.2 with NVFP4 quantization and an open-source project running 290B+ Mixture-of-Experts on a gaming PC both surfaced this week. Real demand is building for local inference, driven by cost and data-sensitivity. Patronus AI · GitHub
The agent wave is turning CI into a bottleneck: As agents generate a flood of changes that break CI, test-selection is cutting queueing from hours to minutes. The infrastructure bottleneck of the agent boom is moving to code review and CI. human systems
Model-licensing and “model factory” infrastructure stacks: The Poolside deal shows model-building infrastructure can be an independent asset. There’s experimental value in turning training/fine-tuning pipelines into reusable products. Newcomer
Local inference and quantization tooling: With a growing push to run frontier MoE models locally/at the edge, NVFP4-style quantization pipelines, local MoE runtimes, and the deployment/monitoring tooling around them are practical niches. Patronus AI · GitHub
Test-selection and CI observability for the agent era: Intelligent test selection and CI queue management built for the large change-sets agents produce is an obvious demand. human systems
Salesforce Agentforce partner-monetization lag: A gap between agent-platform hype and actual partner revenue is now visible. Watch how fast the agent market moves from “demand signal” to real monetization. The Register
Frontier-model regulation and safety governance: With the Nvidia/SpaceX/MS AI-safety initiative and Brundage’s guardrails argument stacking up, how frontier-model vetting and safety ops get institutionalized remains open. The Guardian