rory o'driscoll

SpaceX buying Cursor exemplifies a new dynamic: entities with massive compute (Colossus) acquire AI applications to turn their capacity into revenue. The application's gross margin problem becomes the compute owner's revenue opportunity. This could drive more compute-rich players (neoclouds, sovereigns) to acquire application-layer companies.

Anthropic advocates a government model-approval process that would structurally disadvantage open-source and Chinese models while entrenching closed-source incumbents; the proposal's requirement for Chinese participation makes it practically a stall tactic to freeze competition.

Nvidia's 70% revenue guidance (vs 44% consensus) signals hyperscalers (Google, Oracle, CoreWeave, OpenAI, Anthropic) are accelerating capex, not tapering; this pushes out the 'normalization' date where free cash flow turns positive, keeping compute demand intense for at least 12 more months.

Extrapolating 80% growth and 70% share gets Nvidia to $1T revenue by 2030; the $64T question is whether economic ROI sustains the next $2T capex increment. Market currently prices mid-20s PE assuming growth deceleration, not continuation.

Nvidia is recycling entire free cash flow into neoclouds, model companies, and data providers — each investment expands GPU TAM and locks in demand, creating a self-reinforcing cycle where capex becomes revenue.

CIOs moving from cost-per-token to cost-per-completed-task; every company will need tiered model strategy (frontier for complex, cheap for volume); developer token consumption exploding as agents run 24/7, pressuring budgets.

AI infrastructure companies sell picks and shovels — observability, networking, communications APIs — to hyperscalers and model builders who need 100x more capacity, requiring no product reinvention.

Palantir proves enterprise AI requires (1) true outcome-based contracts ($2B deal tied to $8B savings) and (2) forward-deployed engineers who implement solutions — generic SaaS sales motions fail for AI transformation.