Economic model — community and SMB customer tiers

@@ -5,58 +5,64 @@ slug: economic-model
category: architecture
type: topic
quality: complete
short_description: "PointSav's two-tier commercial structure: a free Community tier that serves as an adoption funnel, and a paid SMB Customer tier targeting regulated small-to-medium businesses that hyperscalers cannot serve economically."
short_description: "PointSav's two-tier commercial structure: a free Community tier that serves as an adoption funnel, and a paid SMB Customer tier targeting regulated small-to-medium businesses that hyperscale billing models cannot serve economically."
status: active
bcsc_class: public-disclosure-safe
last_edited: 2026-05-01
last_edited: 2026-05-22
editor: pointsav-engineering
cites: []
paired_with: economic-model.es.md
---

PointSav's commercial structure has two tiers. There is no Enterprise tier. The decision is not a positioning choice — it is a structural one. Regulated small-to-medium businesses occupy a market segment hyperscalers cannot reach economically, and the platform is designed entirely around serving that segment.
A regulated small business sits in a gap. It is too small for an enterprise AI contract — those begin in the hundreds of thousands of dollars a year — and too regulated to run its data through a consumer AI product. Neither sales motion reaches it.

PointSav's commercial structure is built entirely around that gap. <!--claim id=two-tiers-no-enterprise confidence=structural cites=[]-->There are two tiers and no Enterprise tier: a free Community tier that is the adoption funnel, and a paid SMB Customer tier for regulated small-to-medium businesses.<!--/claim-->

<!--claim id=structural-not-positional confidence=structural cites=[]-->The decision is structural, not positional. The target segment runs AI-tooling contracts of roughly 5,000 to 50,000 dollars a year — too small for an enterprise sales motion, too regulated for a consumer product — and the platform's cost structure is designed to serve it profitably.<!--/claim-->

For a buyer in that segment, the platform's economics are aligned with theirs. This article covers the two tiers, why there is no Enterprise tier, the addressable segment, federation, and the cost asymmetry that protects the position.

## The two tiers

**Community** is the free tier. It operates under an AGPL-3.0-or-later license. A Community deployment consists of one ToteboxOS archive and one ConsoleOS terminal, with local model inference available as an optional component. Community is the adoption funnel: it generates contributors, surfaces edge cases, and makes the substrate legible to future customers. PointSav earns no revenue from Community deployments.
<!--claim id=community-tier confidence=structural cites=[]-->**Community** is the free tier, under an AGPL-3.0-or-later licence. A Community deployment is one ToteboxOS archive and one ConsoleOS terminal, with local model inference as an optional component. Community is the adoption funnel — it generates contributors and surfaces edge cases — and PointSav earns no revenue from it.<!--/claim-->

**SMB Customer** is the revenue tier. It operates under a Functional Source License with an Apache-2.0 fallback after the Delay Open-Source Publication period, plus a commercial license where required. SMB Customer deployments include multi-archive aggregation via os-orchestration, GPU burst capability, federated LoRA marketplace participation, and priority access to updated base models as they are produced. The commercial relationship is an Order Form per customer.
<!--claim id=smb-tier confidence=structural cites=[]-->**SMB Customer** is the revenue tier, under a Functional Source License with an Apache-2.0 fallback after the delay period, plus a commercial licence where required. An SMB Customer deployment adds multi-archive aggregation, GPU burst capability, federated adapter-marketplace participation, and priority access to updated base models. The relationship is an Order Form per customer.<!--/claim-->

## Why no Enterprise tier

The addressable customer for an Enterprise-tier AI platform typically runs annual contract values above $500,000. Hyperscalers are structured for that segment — their sales organisations, compliance certifications, and infrastructure commitments reflect it. PointSav is not equipped to compete there, and should not try.
The addressable customer for an Enterprise-tier AI platform typically runs annual contract values above 500,000 dollars. Hyperscale providers are structured for that segment — their sales organisations, compliance certifications, and infrastructure commitments reflect it. PointSav is not equipped to compete there and does not try.

The segment PointSav targets runs annual contract values of $5,000 to $50,000 for AI tooling. That gap — too small for an enterprise motion, too regulated for a consumer product — is structurally inaccessible to platforms built around hyperscale billing models.
<!--claim id=the-gap confidence=structural cites=[]-->The segment PointSav targets runs annual AI-tooling contract values of 5,000 to 50,000 dollars. That gap — too small for an enterprise motion, too regulated for a consumer product — is structurally inaccessible to a platform built around a hyperscale billing model.<!--/claim-->

The 2026 evidence is consistent: GPU pricing at neoclouds is structurally lower than at hyperscalers, a substantial fraction of hyperscaler customers report billing unpredictability, and hyperscaler capital expenditure commitments for AI infrastructure continue to rise. These trends do not help SMBs; they increase the pricing floor those customers face.
The 2026 market evidence is consistent: GPU pricing at specialist providers runs structurally below hyperscale pricing, a substantial fraction of hyperscale customers report billing unpredictability, and capital-expenditure commitments for AI infrastructure continue to rise. None of these trends help a small business; each raises the pricing floor it faces.

## The addressable SMB segment

Regulated SMBs share three properties that define the market:
Regulated small-to-medium businesses share three properties that define the market.

1. They are too small for an enterprise sales motion at hyperscaler minimums.
1. They are too small for an enterprise sales motion at hyperscale minimums.
2. They are too regulated for consumer or unmanaged AI — HIPAA, PIPEDA, GDPR, FINRA, and equivalent Canadian provincial frameworks apply.
3. They require their data and their AI to remain on infrastructure they control.

Examples include small clinics subject to health privacy legislation, regional law firms with privilege and confidentiality obligations, mid-cap financial advisors with regulatory reporting requirements, and real-estate operators maintaining corporate document archives. These customers are not edge cases. They are the mainstream of the regulated economy below the enterprise threshold.
Examples include small clinics under health-privacy legislation, regional law firms with privilege and confidentiality obligations, mid-cap financial advisors with regulatory reporting requirements, and real-estate operators maintaining corporate document archives. These are not edge cases. They are the mainstream of the regulated economy below the enterprise threshold.

## Federation

Federation capabilities — the federated LoRA marketplace, the Mooncake KV pool, and base model updates as they are produced — are included in the SMB Customer license. There is no separate federation tier. Every paying customer may participate; privacy-preserving federated learning techniques are mature enough in 2026 to make this structurally sound.
<!--claim id=federation-included confidence=structural cites=[]-->Federation capabilities — the federated adapter marketplace, the shared key-value cache pool, and base-model updates — are included in the SMB Customer licence. There is no separate federation tier; every paying customer may participate.<!--/claim-->

## Continued pretraining as a curatorial investment

PointSav's investment in continued pretraining of the base model — the production of PointSav-OLMo-N variants — is funded by SMB Customer license revenue and benefits every customer when an updated base is distributed in the next platform release. The economic structure resembles that of a Linux distribution: curation is funded by the subscription business; customers run their own installations; the improved base returns to the full subscriber base.
<!--claim id=linux-distro-model confidence=structural cites=[]-->PointSav's investment in continued pretraining of the base model is funded by SMB Customer licence revenue, and the improved base benefits every customer when it ships in the next platform release. The economic structure resembles a Linux distribution: curation is funded by the subscription business, customers run their own installations, and the improved base returns to the full subscriber base.<!--/claim-->

## The cost asymmetry

The economics that protect this position are straightforward. LoRA fine-tuning of a seven-billion-parameter model costs in the range of $30 to $100 per adapter. Continued pretraining of that same model costs $30,000 to $100,000. Pretraining from scratch at that scale costs $500,000 to $2,000,000.
The economics that protect this position are straightforward. Adapter fine-tuning of a seven-billion-parameter model costs roughly 30 to 100 dollars per adapter. Continued pretraining of that same model costs 30,000 to 100,000 dollars. Pretraining from scratch at that scale costs 500,000 to 2,000,000 dollars.

SMBs can do LoRA fine-tuning on their own data. They cannot do continued pretraining. PointSav does continued pretraining. Federation pools the learning from per-customer LoRA adapters into a commons that improves the base for everyone. This asymmetry is structural: it does not depend on any particular competitive outcome. It depends on the economics of compute at scale, which are unlikely to invert.
<!--claim id=cost-asymmetry confidence=structural cites=[]-->A small business can fine-tune adapters on its own data; it cannot fund continued pretraining. PointSav funds the continued pretraining, and federation pools the per-customer adapter learning into a commons that improves the base for everyone. The asymmetry does not depend on any particular competitive outcome — it depends on the economics of compute at scale.<!--/claim-->

## Service availability by tier

Ring 1 services (boundary ingest: filesystem, people, email, and input) and Ring 2 services (knowledge and processing: content, extraction, search, and egress) are available in both tiers. Ring 3 services differ: Community includes optional local model inference; SMB Customer adds GPU burst, external API integration, multi-archive aggregation, federation, and access to updated base models. Support and integration services are community-forum only for Community and contracted for SMB Customer.
Ring 1 services (boundary ingest) and Ring 2 services (knowledge and processing) are available in both tiers. Ring 3 differs: Community includes optional local model inference; SMB Customer adds GPU burst, external API integration, multi-archive aggregation, federation, and access to updated base models. Support is community-forum only for Community and contracted for SMB Customer.

## See also