From rented infrastructure to sovereign AI. The next phase of Sentigen.
80% of CIOs plan to repatriate cloud workloads. 93% of executives say AI sovereignty is critical. 55% of enterprise AI now runs outside public cloud, up from 12% in 2023, a 4.6× increase. $936/month handles 200-500 users. Scale by adding servers, not per-user fees.
SOVEREIGN INFRASTRUCTURE
Most SaaS companies rent everything: servers, databases, AI APIs. They pay more as they grow. We're building the opposite: a sovereign stack where infrastructure cost stays flat while revenue scales. One server at $936/month handles 200-500 users. Need more? Add another server, not another per-user fee.
The $936/mo covers a Hetzner GEX130 (RTX 6000 Ada GPU, 48GB VRAM) plus backup and DNS. Full production stack including app server and self-hosted database: ~$1,034/mo. At ~500 users, add a second inference node ($1,200/mo total). At 1,000: $2,500/mo. At 5,000: $8,000/mo. Breakeven vs cloud at just ~8 users.
Seven layers that replace cloud vendors. Click any layer to see what it does and why it matters.
SaaS cost assumes $45/user/month in cloud infrastructure (API calls, database, hosting). Sovereign cost shows actual stepped hardware costs. Margin is gross margin on $100/user ARPU.
| Users | SaaS Cost | Sovereign Cost | Gross Margin |
|---|---|---|---|
| 50 | $2,250/mo | $400/mo | 82% |
| 100 | $4,500/mo | $500/mo | 89% |
| 500 | $22,500/mo | $1,200/mo | 95% |
| 1,000 | $45,000/mo | $2,500/mo | 94% |
| 5,000 | $225,000/mo | $8,000/mo | 96% |
ON-CHAIN TRUST LAYER
Today, enterprise identity runs through vendors like Okta at $12-18/user/month, with 5 major breaches in 18 months (including one exposing all 18,400 customers). We replace that with three blockchain-based primitives that cost pennies and have no central point of failure. Users don't need to know anything about crypto. They just use Face ID.
UDA System
Digital keys that can’t be faked, copied, or revoked by a vendor.
Instead of usernames and passwords managed by a company like Okta, each user gets a cryptographic token (like a digital passport) stored on the blockchain. The token proves who you are and what you can access. No central server to hack. No password database to leak.
Technical Detail
ERC-721 tokens on Base L2 (Coinbase’s blockchain). Four access tiers. Wallet login via Coinbase Smart Wallet — users just use Face ID, no crypto knowledge needed. One revoke cascades through the entire delegation tree.
Cost Comparison
Okta: $12-18/user/month. UDA: $0.01-$0.05 one-time mint per credential.
$3/mo Audit
A tamper-proof receipt that proves nothing was changed after the fact.
Every day, the platform creates a cryptographic fingerprint (hash) of all activity logs and writes it to the blockchain. If anyone — including us — tries to alter the logs, the fingerprint won’t match. Auditors can verify compliance independently, without trusting the company.
Technical Detail
Daily Merkle root of all platform logs anchored on-chain. Timestamped and verifiable by anyone with a blockchain explorer. No enterprise audit infrastructure needed.
Cost Comparison
Traditional compliance infrastructure: $50K-500K/year. On-chain anchoring: $3/month.
Agent Identity
Every AI agent has a verifiable identity and signs its work.
When an AI agent drafts an email, generates a report, or makes a recommendation, it cryptographically signs the output. This creates an unforgeable record of which agent produced what, when, and under whose authority. Critical as AI agents start acting autonomously in business.
Technical Detail
Every AI agent gets a Base wallet. Signed outputs create cryptographic proof of provenance. On-chain registry enables cross-platform verification — an agent’s reputation follows it everywhere.
Cost Comparison
AI agent economy: $7.3B in 2025, projected $47-52B by 2030. The platform that controls agent identity controls the market.
Four levels of access, each represented by a blockchain token. Think of it like keycards at a building. Different cards open different doors. The blockchain makes the cards unforgeable and instantly revocable.
Full platform control, token minting, delegation
Can do everything: create new access tokens, grant permissions to others, and revoke access across the entire tree.
API access, agent deployment, tool creation
Can build on the platform: connect integrations, deploy AI agents, create custom tools. Cannot grant access to others.
Workspace management, user invites, reporting
Can manage day-to-day operations: invite team members, run reports, configure workspaces. Cannot change the platform itself.
Scoped tool execution, signed outputs, rate limits
AI agents get their own identity with strictly limited permissions. Every action is signed and auditable. Rate-limited to prevent runaway costs.
When an Admin grants access to a Partner, that Partner can delegate to their team. The key insight: if you revoke the Partner's access, everyone they granted access to loses access instantly. One blockchain transaction. No chasing down individual accounts. No waiting for IT to process tickets. Instant, complete, verifiable.
Scenario
Revoking Partner A's token (dashed border) instantly kills access for Team Lead and their AI Agent too. Everyone in the dashed branch. Partner B and their agents are unaffected. One transaction, complete and irreversible.
Compare to Today
Traditional offboarding: 3-5 IT tickets,
2-4 weeks, often incomplete.
On-chain: 1 transaction, instant.
No seed phrases. No MetaMask. Users sign in with Face ID through Coinbase Smart Wallet. Behind the scenes, the system verifies their blockchain token to determine access level.
What the user sees
What happens behind the scenes
SCALING ECONOMICS
Not every customer needs their own server. Some want the cheapest option. Others need complete isolation for regulatory reasons. The sovereignty ladder matches infrastructure to what each customer actually needs, and is willing to pay for.
L1: Shared + RLS
Apartment Building
Everyone shares the same database, but each customer can only see their own data, enforced at the database level, not the app level. Like apartments with soundproof walls and separate locks.
Best for: Free tier, small teams, startups
L2: Separate Schema
Townhouses
Each customer gets their own set of database tables on shared hardware. Like townhouses: shared foundation, but your own walls, your own plumbing. A breach in one schema cannot reach another.
Best for: Mid-market, data-sensitive orgs
L3: Isolated Container
Detached House
Each customer runs in their own isolated container: their own database, their own AI models, their own everything. Like a detached house on a shared street. Completely independent but sharing the underlying hardware.
Best for: Enterprise, regulated industries
L4: Dedicated Stack
Private Estate
The customer gets their own physical server in a data center of their choosing. Nothing is shared. They can white-label the entire platform as their own product. Like a private estate with its own utility connections.
Best for: Government, defense, sovereign AI mandates
Most customers use L1 (free/cheap). A few pay premium for isolation. The blend creates strong margins.
500 L1 users + 20 L2 tenants + 5 L3 tenants + 1 L4 client = 526 customers
Total Customers
Blended Revenue
Total Infra Cost
Blended Gross Margin
AI Costs Collapsing 10× Annually
GPU costs halve every 2.5 years. Cloud H100 rentals fell 64-75% from 2023 to 2025. Today's $936/month stack will deliver 2-3× performance within 2-3 years at the same price — or handle 2-3× more users.
Enterprise Sovereignty Demand
Sovereign cloud market: $97B in 2024, growing 23-25% CAGR to $649B+ by 2033. 67% of Fortune 500 companies now run on-premise AI infrastructure.
On-Chain Infrastructure Matured
Minting an ERC-721 credential: $0.01-$0.05 on Base L2. Compare to Okta: $12-18/user/month — with 5 major security breaches in 18 months. The UX barrier dissolved with passkey wallets.
ROADMAP
Each phase generates revenue independently. The on-chain layer is additive. The platform works without it. This is a migration, not a rewrite.
V1.1
Self-Host Stack
Supabase, Inngest, Langfuse on bare metal. Caddy + WireGuard for networking.
V1.2
GPU Inference
vLLM + Qwen3-8B for NANO/MICRO tiers. Portkey routing between local and cloud models.
V1.3
UDA Access Control
ERC-721 tokens on Base L2. Wallet login via SIWE + Coinbase Smart Wallet.
V1.4
Audit Anchoring
Merkle root of daily logs anchored on-chain. Tamper-proof compliance for $3/month.
V2
Agent Identity + Delegation
Every AI agent gets a Base wallet. Signed outputs. Cascading delegation tree with single-revoke kill switch.
V3
Composable Access + Agent Economy
Agents as economic actors. On-chain reputation. Cross-platform tool marketplace.
Future
Autonomous Agent Marketplace
Agents hire agents. Reputation-gated access. Revenue-sharing smart contracts.
Base L2
Coinbase L2 for access tokens and audit anchoring
vLLM
High-throughput local inference engine
Qwen3-8B
Open-weight model for NANO/MICRO tiers
Portkey
AI gateway with fallback routing
Caddy
Automatic HTTPS reverse proxy
WireGuard
Encrypted VPN mesh between nodes
The AI agent economy reached $7.3B in 2025, projected to hit $47-52B by 2030. Gartner predicts 40% of enterprise apps will embed AI agents by end of 2026. The platform that controls identity, access, and provenance for these agents becomes the operating system for the AI economy. That's what we're building.