Starving Genies
Kent Beck · email · 2026-04-03 source ↗ Reader 🖥️
AI Bubble DebateAI Employee vs Tool FramingCompute & Energy InfrastructureExplore / Expand / ExtractThe Genie (Beck)Usage Limits & Rationing
The simultaneous usage throttling by AI model providers is best understood not as an engineering constraint but as a narrative constraint: all three face the same investor-story shelf life, forcing synchronized demand-side rationing while the race to bend the supply curve up determines who wins the next wave.
theoretical · essay / newsletter analysis — Applies Beck's 3X framework (Explore/Expand/Extract) to explain why AI model providers simultaneously throttled usage. Part industry analysis, part framework application.
Structure
| Part | Role |
|---|---|
| 3X Expand phase primer | Establishes the staircase growth model and the concept of rate-limiting resources in Expand |
| Genies section: the synchrony observation | States the empirical puzzle: all major AI providers cut limits at once |
| The Bottleneck? — differential diagnosis | Systematically eliminates chips, compute capacity, and inference economics as the binding constraint |
| It's a Race | Identifies the escape condition: whoever bends supply up first breaks the cartel |
| What's Next? — developer and user impact | Explores how throttling segments users and creates conversion pressure |
Key terms
- 3X (Explore/Expand/Extract) Explore / Expand / Extract
Beck's framework for company lifecycle phases: Explore (find what works), Expand (grow fast past bottlenecks), Extract (optimize margins once growth slows) - rate-limiting resource The Genie (Beck)
The specific constraint that creates the ceiling on growth at any given moment in the Expand phase - the story / narrative AI Bubble Debate
The investor-facing justification for continued funding of unprofitable growth; has a shelf life - genies AI Employee vs Tool Framing
AI coding assistants / model providers (Claude, ChatGPT, Gemini etc.) — entities that grant wishes but are now constrained - bending the supply/demand curve
Actively changing the trajectory of resource availability (supply) or consumption (demand) to overcome a ceiling
Propositions
- explicitcorroboratedGrowth in the Expand phase is a staircase, not a curve: you hit successive rate-limiting resources and must overcome each one.conf: highsupports ←Canon 3X: Explore/Expand/Extract · p7 — Canon's own Expand-phase tactics ('intense focus on the next emerging growth bottleneck', 'throttle growth, discard non-essential features') are the general framework of which starving-genies' claim — that Expand-phase growth is a staircase of successive rate-limiting resources to overcome one at a time — is the specific restatement applied to AI providers.
- explicitAll major AI model providers throttled usage simultaneously, which is a signal, not a coincidence.conf: high
- explicitThe binding constraint on AI model providers is not chips, physical compute capacity, or inference economics — it is the investor narrative about the path to profitability.conf: high
- implicitThe synchrony of the throttling is itself the strongest evidence for narrative as the constraint, because engineering constraints would produce variation across companies.conf: high
- explicitUsage limits are investor-signaling devices: evidence of managing toward profitability, not admissions of resource exhaustion.conf: high
- explicitcorroboratedThe company that bends the supply curve up first (cheaper inference, custom silicon, new capacity) wins the next wave of expansion.conf: highsupports ←How the Escalating AI Wars Benefit You · p5 — Anthropic extending Fable access and raising Claude Code limits 50% in direct response to OpenAI's launch is a concrete instance of 'the company that bends the supply curve up first wins the next wave of expansion.'
- explicitThrottling may not be temporary — it could mark the beginning of Extract phase where heavy AI usage becomes a premium product.conf: medium
- implicitNormally, bending demand down during competitive Expand is suicide, but here competitive dynamics are suspended because all providers face the same narrative constraint simultaneously.conf: high
Arguments
The binding constraint on AI model providers is the investor narrative, not engineering. strength: high
Explicit premises:
- All three major providers cut limits at the same time.
- Chip supply differences (Google and Amazon make their own) would produce variation, not synchrony.
- Physical capacity constraints differ by geography and footprint, so would produce variation.
- These companies have essentially unlimited capital, so cost alone is not the driver.
- The same investor class funds all three, at the same stage, hitting the same narrative shelf-life moment.
Implicit premises:
- Synchronized behavior among competitors sharing the same constraint type (but not the same physical resources) implies a shared non-physical constraint.
- Investor patience is finite and operates on roughly shared timelines across a funding cohort.
Inferences:
- Since engineering constraints would produce variation and we observe synchrony, the constraint must be non-engineering.
- The shared investor class and stage explains the synchrony; the constraint is narrative credibility.
Whoever achieves better inference unit economics first will break the competitive equilibrium and win the next expansion wave. strength: medium
Explicit premises:
- The current throttling is a cartel-like equilibrium sustained by shared narrative pressure.
- Distillation, caching, smarter routing, and custom silicon can bend the supply curve up.
- The company that can open the throttle while competitors cannot gains decisive advantage in Expand.
Implicit premises:
- The Expand phase rewards growth above all; rationing is inherently unstable if one player can escape it.
- The technology improvements listed are feasible in the near-to-medium term.
Inferences:
- A unilateral supply-curve improvement breaks the equilibrium because Expand dynamics punish rationing.
Narratives
- The genie metaphor — AI assistants as wish-granting genies that have 'all gone to rehab at the same time,' starving their users of capability.
→ Frames throttling as withdrawal of a quasi-magical resource, making the user impact visceral and setting an irreverent, insider tone. - The retcon narrative — successful companies retroactively rewrite near-death Expand crises as inevitable success.
→ Warns against survivorship bias in interpreting company histories; implies current throttling may look different in retrospect. - The developer mid-flow cliff — a power user hitting a daily cap and being forced to write a blog post instead.
→ Autobiographical humor that concretizes the impact of throttling on the most valuable user segment.
Questions addressed
- ✅ Why did all major AI model providers throttle usage at the same time?
The shared binding constraint is investor narrative shelf-life, not engineering limits, which explains the synchrony. - ✅ What is the actual rate-limiting resource for AI model providers right now?
Investor confidence in the path to profitability — 'the story.' - ⭕ Are the usage limits temporary or the beginning of a new equilibrium?
Explicitly left open. Beck frames it as the key diagnostic: are we in a brief Expand pause or entering Extract? - ⭕ Which company will break the equilibrium by bending supply up first?
Not answered, but identified as the signal to watch.
Gaps
- unstated assumptionBeck assumes the three providers are essentially in a shared Expand phase, but they may be at different lifecycle stages (e.g., Google's AI is embedded in a profitable search business, while Anthropic is pre-revenue at scale). The 3X phase may not be uniform across players.
Read next: Compare with analysis of each provider's actual business model maturity and investor structure.closed byCanon 3X: Explore/Expand/Extract — Beck's own canonical statement of 3X (p9 here) explicitly assumes a portfolio runs mixed phases simultaneously by design ('some Extract products that pay the bills & pay for a portfolio of Explore projects') — the framework does not assume phase uniformity across projects, it is built to manage non-uniformity. This answers the gap's question about whether the model requires a shared phase: it doesn't, even though starving-genies' application of it to three AI providers reads as if it does. - alternative explanation not consideredRegulatory signaling is not discussed. Providers may be synchronizing on usage limits partly to demonstrate responsible scaling to regulators, not only to investors.
Read next: Look at EU AI Act timeline and responsible scaling commitments from Anthropic, Google, OpenAI. - evidence gapThe 'narrative shelf-life' explanation is offered through elimination rather than direct evidence. No investor statements, board dynamics, or financial data are cited.
Read next: Cross-reference with investor communications, earnings calls, or insider reporting on AI funding dynamics. - scope limitationThe analysis treats 'AI model providers' as a unified category but doesn't address open-source models (Llama, Mistral) which face different economic dynamics entirely and could disrupt the equilibrium from outside the cartel.
Read next: Consider how open-weight models change the competitive dynamics Beck describes.closed byHoward Yu's Post — American AI models went from 70% of usage to 30% in 12 months — That gap noted the analysis treated model providers as one category and ignored open-source models disrupting the equilibrium from outside the cartel. This is that disruption made concrete: Chinese open-weight models at ~1/9 the price now take the majority of routed volume, breaking the pricing equilibrium from outside.