jordi hays

Frontier labs are locking in multi-year, multi-hundred-billion-dollar cloud capacity commitments, driving hyperscaler capex to unprecedented levels and making cloud infrastructure the primary bottleneck for AI progress.

AMD's Helios rack-scale system and Anthropic's 2 GW commitment signal a shift toward integrated, rack-level AI infrastructure deployments where silicon, networking, and software are co-optimized at massive scale.

Nvidia-led Open Secure AI Alliance rallies $18T market cap to frame open-weight models as defensive tools, creating economic pressure that makes a US ban on open-source AI unlikely.

OpenAI's Project Camila (20B, Georgia) and Anthropic's 2 GW AMD deployment reflect a shift to gigawatt-scale, purpose-built AI campuses with dedicated power and networking, redefining data center economics and geography.

DeepSeek's disclosed unit economics — 85% inference margins, $1B API revenue, 10-month hardware payback on 3-5 year depreciation — demonstrate that inference can be highly profitable at scale, validating the economic model for AI-native companies.

Security vendors can use restricted frontier models to patch vulnerabilities before similar capabilities diffuse to attackers through open source, but the stable and narrowing capability gap makes the defensive lead temporary and raises the value of rapid deployment.

State-level suits against Meta (NM, KY, LA) are following the tobacco playbook: states sue for Medicaid/mental-health costs, settle for perpetual revenue-linked payments that get securitized. With Meta's $200B revenue, a tobacco-scale settlement would be existential; financialization of state payment streams already underway.

Large US banks are exploring a consortium acquisition of Fiserve's debit networks (Star, Excel) to own end-to-end payment rails, capturing more economics and building toward AI-agent checkout surfaces — a structural rebundling that threatens card network leverage and mirrors Capital One/Discover.