Oction Labs
Sovereign AI infrastructure for regulated markets.
Brandon Gill, CEO · Bailey Rhodes, Co-founder · 2026
The thesis
We own the hardware, the data, and the knowledge graph that public-sector and regulated clients can't legally hand to OpenAI or Anthropic.
The problem
Regulated buyers want AI but can't legally buy what's available.
- → Municipalities, healthcare, finance, legal, public-safety cannot ship data to general-purpose LLM providers
- → Compliance (NIST 800-53, HIPAA, SOC 2, provincial sovereignty) requires on-prem or sovereign-cloud
- → Hosted models leak. Open-weight models lack support. Custom builds require teams clients don't have.
What we want to do
Three models, one company.
Model 1 · Phase 1
Sovereign AI as a Service
Deployable today. Bespoke builds for Muni → regulated clients. Stealth: clients buy software, we own hardware. We build + manage the knowledge graph + data ingestion.
Revenue: managed-service fee + initial build fee
Model 2 · Phase 3
Proprietary Asset Model
Pretrained, refined data for agentic models. 4 paths: (a) anonymized Model 1 data, (b) purchased datasets, (c) data partnerships, (d) public-data compression.
Revenue: data-layer subscriptions, enrichment, smart licensing
Model 3 · Phase 2
AI Research Lab
Net Zero Compute (solar + hydro + wind). Canadian-funded data centers. Bi-annual cyber-safety tests as gov deliverable.
Revenue: gov grants, optimization contracts
Why us
Four moats that compound.
Compliance fluency
We read the regulations clients can't operationalize.
Hardware control
We buy, deploy, and operate the compute — not a SaaS dependency.
Data ownership across lifecycle
Raw → anonymized → licensed, all on infrastructure we control.
Net Zero Compute as moat
Canadian power mix + grant alignment is a regulatory + cost arbitrage.
Where we land first
Canadian Municipal markets.
Why Muni: predictable procurement, well-defined compliance (provincial + federal), ≥$250k typical spend per client, sales cycle in months not years
Why we win: general-purpose AI vendors can't get permitted; sovereign deployment is a hard requirement, not a preference
Then adjacent regulated: healthcare admin, public-safety analytics, post-secondary, legal aid, infrastructure ops
Brandon to add: 2–3 named Muni prospects in active conversation
Unit economics — Model 1 per engagement
Built to be conservative.
Initial build fee[Brandon: $150–250k]
Managed-service fee / month[Brandon: $10–25k]
Hardware capex per client (Mac M3 Studio cluster)~$60k
Monthly ops cost per client~$1k–1.6k
Gross margin target (managed service)[Brandon target]
CAC payback[Brandon target]
LTV[Brandon target]
Numbers in brackets are placeholders for Brandon's review. No fabricated precision.
The flywheel
Phase 1 funds the data asset that Phase 3 monetizes.
Every Model 1 deployment generates: (1) anonymized operational data, (2) patterns we encode into the knowledge graph, (3) tested infrastructure templates.
Phase 3 monetizes that asset: subscription data layers, smart licensing, enrichment products.
Phase 1 looks like a service business with a margin ceiling.
Phase 3 is venture-scale lock-in: proprietary data, subscription revenue, smart licensing.
Model 3 · The moat builder
Net Zero Compute. Canadian sovereign.
● Solar + hydro + wind powered data center pilot
● Canadian government grants in pipeline (Innovation Canada, provincial energy programs)
● Bi-annual cyber safety tests as deliverable to gov stakeholders
● Not a profit center — a positioning + cost-arbitrage moat that makes Model 1 + Model 2 defensible
Brandon to add: Active grant applications — names, amounts, decision dates
Hardware strategy
Compliance-friendly. Supply-stable. Operator-friendly.
Compute
Apple Mac M3 Studio with 512GB unified memory. Quiet, mature, energy-efficient, well-supported software stack. Clusterable for client deployments.
Storage
Raw — in-house storage, client-isolated. Anonymized delivery — S3, encrypted, region-locked.
Network
Tailscale-based mesh for ops. Client deployments isolated per-VPC. Wireguard fallback.
Future scale
In-house data center under Model 3 — timing tied to grant decisions. Solar + hydro + wind primary power.
Team
Lean and composite.
Core team — 4–5 co-founders @ ~$120k each
→ Brandon Gill — CEO / Sales
→ Bailey Rhodes — Co-founder, Owner-level
→ [Brandon: founder 3 + role]
→ [Brandon: founder 4 + role]
→ [Brandon: founder 5 + role — if applicable]
Future hires — scaled per Model 1 ramp
Data engineers · ML engineers · Compliance · Security
AI staff (built in-house, not headcount)
Julian Pierce (Systems & Infrastructure) · Lucius Fox (Operations Intelligence) · Atlas (Security & Forensics)
Use of funds
Up to $20M for the data side.
[Brandon: total raise amount + stage + instrument]
Hardware (Mac M3 Studio + storage)~$5M
Engineering hires (post-founders, 18-mo runway)~$6M
Canadian data center pilot (Model 3 seed)~$3M
Founder comp (5 × $120k × 18 mo)~$900k
Working capital + reserves + contingency~$5M
Detailed line-item breakdown in
investment-breakdown-2026-05-29.md.Roadmap · Next 12 months
Stage, don't sprint.
Q3 2026
First paid Muni client closed. First M3 Studio cluster deployed.
Q4 2026
2–3 paid clients. Anonymized data lake online. First Canadian grant submitted (Phase 2).
Q1 2027
5 clients. Data partnership pilot (Model 2 path 3). First ML engineer hired.
Q2 2027
Model 2 subscription alpha. Pilot DC site selected.
Brandon to calibrate: exact pipeline timing per current conversations
Risks · We name them
Top 5.
Regulatory shift — federal/provincial AI policy could lock out on-prem play, or mandate it. Mitigation: track both, build Innovation Canada relationships early.
Hardware supply — Apple Silicon supply tight. Mitigation: multi-vendor procurement, 6-month forward inventory.
Talent — 5-person team is lean. Mitigation: hiring plan gated on revenue, not raise.
Grant timing — Canadian grants slow. Mitigation: don't gate Model 3 build on grants; treat grants as runway extension.
Customer concentration — first 2–3 clients dominate revenue. Mitigation: parallel sales motion, content-driven inbound.
The ask
[Brandon: $X.XM · Stage · Instrument]
To fund the data side: $20M across hardware, hires, pilot DC, founder comp, reserves.
Brandon Gill — CEO
[Brandon: contact channel — email / calendar link / LinkedIn]