More Trillion Dollar IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts

TAM

Total Addressable Market — the total revenue opportunity for a product or service if it achieved 100% market share. Used here to argue intelligence is the largest TAM in history.

Jevons paradox

The economic observation that as the efficiency or affordability of a resource increases, total consumption of that resource rises rather than falls. Applied here to AI token pricing.

ARR

Annual Recurring Revenue — an annualized measure of subscription or recurring revenue, used to compare Anthropic and OpenAI's scale.

Confidential filing (S-1)

A process allowing companies to submit their IPO registration statement to the SEC privately before going public, letting them negotiate terms without full public disclosure.

Index inclusion

The addition of a newly public stock into a major market index (e.g., S&P 500), which forces passive index funds to buy shares and can significantly boost demand and liquidity.

Lockup period

A post-IPO window during which insiders and early investors are contractually prevented from selling shares, typically 90–180 days. SpaceX pioneered a staged lockup with milestone-based releases.

Model fungibility

The ability to swap one AI model for another interchangeably in an application without losing memory, context, or history — the holy grail for enterprises wanting to route to the cheapest model.

Post-training

Fine-tuning a pre-trained AI model on a specific dataset or set of tasks after initial training, used to specialize a general model for a particular enterprise use case.

Harness

In AI deployment, the orchestration layer (prompt engineering, memory, skills, context management) that wraps a model and determines how efficiently tokens are used.

Inference

The process of running a trained AI model to generate outputs (e.g., answers, code) in real-time, as opposed to training. Inference costs are what companies pay per token.

Distillation

A technique where a smaller model is trained to replicate the outputs of a larger, more capable model, effectively transferring knowledge. Alleged to have been used by Chinese labs on American frontier models.

Vibe coding

Colloquial term for using AI tools (like Lovable or Claude Code) to generate software with minimal manual coding, driven by natural language prompts.

Reconciliation bill

A Congressional legislative process that allows budget-related bills to pass the Senate with a simple majority (51 votes) rather than the usual 60 needed to overcome a filibuster.

Superannuation

Australia's mandatory employer-funded retirement savings system requiring employers to contribute a percentage of wages into a privately managed account owned by the employee. Used as a comparator to Trump Accounts.

P-doom

Shorthand for 'probability of doom' — a metric used in AI safety circles to express the estimated likelihood that advanced AI leads to catastrophic outcomes for humanity.

Agentic

Referring to AI systems that can autonomously plan and execute multi-step tasks, make decisions, and take actions over extended periods without human intervention at each step.

Nerf

Slang for deliberately limiting or restricting a model's capabilities, often for safety or compliance reasons. Used here to describe Claude being restricted on health-related queries.

Accredited investor

A legal designation under US securities law for individuals or entities meeting certain income or net worth thresholds, which grants access to investment opportunities restricted from the general public.

GAAP revenue

Revenue recognized under Generally Accepted Accounting Principles — the standard US accounting rules for when and how revenue is recorded, often used to distinguish from ARR or bookings.

Cron job

A scheduled task in computing that runs automatically at specified time intervals (e.g., hourly), used here to describe automated AI agents running recurring trend-spotting tasks.