Singularity Business Models

See which business models are winning as AI rewrites the economy. (working line, not approved)

Models48

Rising29

On watch9

Retiring10

Companies121

Evidence144

Grade A / B / C68 / 76 / 0

Heidi Health The ambient scribe B 2026-09-25 Anthropic (Claude Connectors Directory / plugin submission portal) Apps inside AI assistants A 2026-09-25 Alibaba Cloud (Qwen) Tokens as the unit of sale A 2026-09-24 Meta (Ray-Ban Meta / Oakley Meta, with EssilorLuxottica) AI glasses B 2026-09-21 Xiaomi AI glasses B 2026-09-21 Rokid AI glasses B 2026-09-21 Moonshot AI (Kimi) Open weights with paid layers B 2026-09-11 UBTech Robotics Robots as a service B 2026-09-01 Symbotic Robots as a service B 2026-09-01 Fin (Intercom) Pay per outcome A 2026-09 Vercel (AI Gateway) Model routing A 2026-09 OpenAI (ChatGPT) Tiered intelligence subscriptions A 2026-09 Kuaishou (Kling AI) Generative media subscriptions B 2026-08-20 Micro1 Expert data marketplaces B 2026-08-20 Reddit Content licensed to AI labs B 2026-08-17 Coinbase / Cloudflare (x402 protocol and x402 Foundation) Machine-to-machine micropayments B 2026-08-13 Zoox (Amazon) Driverless rides B 2026-08-11 Circle Internet Group Stablecoin float A 2026-08-05

The map: rung by value family Loop in small type · rows ordered by loop gain

Loop gain: each time AI gets markedly better and cheaper, how much stronger does one turn of this model's loop become? Rung 0 models sit in the family their value is moving toward. Leveraged is a flag, off the ladder. Grey: on watch.

Rung · loop gain	Selling abundance	Selling the finished work	Owning the scarce inputs	Owning the scarce human	Owning the choke points
4. Self-improvingSelf-set · 1		Autonomous businesses (watch)self-improving
3. CompoundsRising · 4		Engineers deployed with the softwarelearning The tiny-team companyself-improving Digital workers priced as labor (watch)self-improvingThe one-person company (watch)self-improving
2. RidesPositive · 23	Tokens as the unit of salecost-curve Tiered intelligence subscriptionscost-curve AI companions (watch)learning	Pay per outcomelearning Vertical AI priced by the worklearning Discovery paid by milestonelearning AI roll-ups of services firmscost-curve The ambient scribelearning Driverless rideslearning Robots as a servicelearning AI-tutored schooling (watch)learningAI-run insurance (watch)learning		Expert data marketplaceshuman-signal Paid AI access to contenthuman-signal Likeness and voice licensing (watch)human-signal	Agentic checkoutcontrol-point AI glassescontrol-point Open weights with paid layersnetwork Open protocols as commonsnetwork Stablecoin floatnetwork Prediction marketsnetwork Machine-to-machine micropayments (watch)networkApps inside AI assistants (watch)control-point
LeveragedBorrowed · 4			Neocloudscapital Compute for equitycapital Sovereign AI capacitycapital Energy underwritten by computecapital
1. Flow throughAbout zero · 6	Credits for AI workcost-curve Prompt to appcost-curve Generative media subscriptionscost-curve			Content licensed to AI labshuman-signal	Optimizing for AI answerscontrol-point Model routingcontrol-point
0. FightsNegative · 10	towardStock media licensinghuman-signal	towardPer-seat software pricingcost-curve Offshore contact-center laborcost-curve Per-word human translationcost-curve Commodity freelance gigscost-curve Headcount-billed IT servicescost-curve The billable hour and the pyramidlearning		towardSearch-referral publishingcontrol-point Question-and-answer sitesnetwork	towardTen-blue-links search advertisingcontrol-point

Spine: the alignment ladder How aligned with the curve?

FLAGLeveragedBorrowed4A flag, off the ladder: the gain comes from expectations of the curve and reverses if it slows.R4Self-improvingSelf-set1The business redesigns its own loop, increasingly by agents.R3CompoundsRising4The loop makes its users more able, so the gain itself grows.R2RidesPositive23Each step of the curve makes its own loop stronger.R1Flow throughAbout zero6The curve helps it, but the gain passes straight through to customers and rivals.R0FightsNegative10The curve shrinks what it charges for (seats, hours, heads, words, clicks).

Complement: value families Where does the money land?

Selling abundance6 on · 1 retiring toward

Volume of cheap intelligence. Loops: cost-curve, learning.

Selling the finished work14 on · 6 retiring toward

The outcome, not the tool or the hours. Loops: learning, cost-curve, self-improving.

Owning the scarce inputs4 on

Compute, energy, capital, capacity. Loops: capital.

Owning the scarce human4 on · 2 retiring toward

Original human signal, expertise, likeness. Loops: human-signal.

Owning the choke points10 on · 1 retiring toward

Where machine choices are made and paid. Loops: control-point, network.

The family says where value lands; the loop says why it compounds. One family can hold several loops.

What the placements show The AI's reading of the v5 map

Rung 2 holds about half the map (23 of 48). The families do the finer sorting inside it.
Selling the finished work is the largest family (20 of 48), and it is where every retiring labor model is heading.
Rungs 3 and 4 hold five models, four of them on watch. Nothing evidenced yet compounds its users or redesigns itself.