Advania · Yggdrasil

Full AI capability. Zero sovereignty compromise.

Advania's trusted AI platform. Built for the Nordic regulated enterprise.
Discovery deck · 2026 Advania Sverige AB
Yggdrasil
02 · The thesis
§02 · The thesis

The most capable AI models live in US cloud.
The most sensitive Nordic data must stay on Swedish soil.

Every regulated organisation in the Nordics faces this contradiction every day. Yggdrasil exists to resolve it architecturally, not by policy.

Yggdrasil
03 · Problem
§03 · The impossible choice

Four problems. One root cause.

01 · Paralysis

Compliance blocks adoption

Legal and security teams block AI outright. No approved solution meets sovereignty requirements.

02 · Shadow AI

Employees use it anyway

ChatGPT, Claude, Copilot, all unapproved. Sensitive data leaks to US providers with no audit trail.

03 · Tradeoff

Quality vs. sovereignty

On-prem-only means lower-quality models. Frontier models need cloud APIs. The choice is lose or cheat.

04 · Invisible spend

Finance can't see it

GPU spend is opaque. Per-team, per-project, per-use-case cost attribution does not exist.

Yggdrasil
04 · Shadow AI
§04 · Shadow AI is already happening
33%
of EU workers use generative AI at work
15%
do so through official, sanctioned channels

The gap is where confidential data goes. No audit, no governance, no control. Just a plausibly deniable browser tab.

Yggdrasil
05 · Tradeoff
§05 · Quality–sovereignty

The false binary.

Until now you could have full capability or full sovereignty. Never both. Yggdrasil collapses the tradeoff.

Path A · stay compliant

Self-host a 13B open-source model. Accept the quality gap.

Path B · keep up

Route everything to the frontier. Accept the compliance gap.

Path C · Yggdrasil

Every request routed by sensitivity. Full frontier for public. Swedish infra for sensitive. One platform.

Yggdrasil
06 · Yggdrasil
§06 · At a glance

One platform. Every model.
Governed by architecture, not policy.

A unified, production-grade foundation for building, operating, and governing AI systems on Advania infrastructure.

Yggdrasil
07 · Three pillars
§07 · Three pillars

Three pillars. One trust layer.

PILLAR 01

Unified access

One API for on-prem open-source models on Advania infrastructure, and frontier models when your policy mode permits. Stream, batch, cache, version, track cost.

PILLAR 02

Architectural sovereignty

T1 and T2 data never leaves Swedish infrastructure. Enforcement at the Secure Router, not in the policy document. Misconfiguration cannot breach sovereignty.

PILLAR 03

Built-in governance

Immutable audit trail, per-tenant cost attribution, guardrails for PII and prompt injection, compliance tooling for EU AI Act, GDPR, NIS2. Audit-ready by default.

Yggdrasil
08 · Who it's for
§08 · Who it's for

Five segments. One platform.

Seg 01

Public sector

Cannot use US cloud AI by policy. GDPR and Schrems II non-negotiable.

Seg 02

Healthcare

Article 9 patient data. Clinical AI cannot route outside Sweden.

Seg 03

Financial services

Hybrid classification. Trading data on-prem. Research can use frontier.

Seg 04

Legal & professional

Client confidentiality. Matter data on-prem. Case law can use frontier.

Seg 05

Startups & scale-ups

Enterprise governance from day one. No re-architecting.

Yggdrasil
09 · Architecture
§09 · Platform architecture

Four layers. Two spines.

L1 · APPLICATIONS & INTEGRATIONS
MaaS · SecureIQ · RFP Assistant · Customer apps · SDKs · REST
L2 · PLATFORM SERVICES · NINE CAPABILITIES
C-03 Model Access C-04 Knowledge C-05 Prompts C-06 Agents C-07 Workflows C-08 Evaluation C-09 Governance C-10 Data Integration C-11 Connectors
⊥ CROSS-CUTTING · C-01 SECURE ROUTER · C-02 OBSERVABILITY
On every inbound request. Nothing bypasses. Every event audited.
L0 · FOUNDATION
OpenShift · NVIDIA GPUs (Swedish DCs) · vLLM · Qdrant · OpenTelemetry · IAM
Yggdrasil
10 · Secure Router
§10 · C-01 Secure Router

The compliant path is the only available path.

Classify → Policy → Route. Every request. Deterministic first, LLM-assisted second. Fail-closed default. When classification is uncertain, the request stays on-prem.

Misclassification cannot cause a sovereignty breach, because the architecture does not permit the wrong path.

Yggdrasil
11 · Sensitivity model
§11 · T1–T4 sensitivity model

Four tiers. Deterministic routing.

TierClassificationExamplesRoute
T1 Restricted Patient records, trading data, personnummer, matter files On-prem · always
T2 Confidential Internal memos, draft contracts, non-public financials On-prem · always
T3 Internal General business content, internal documentation Policy-dependent
T4 Public Published content, marketing, public data Frontier-capable
Yggdrasil
12 · Audit
§12 · Every decision audited

Every routing decision.
Audit-logged. Immutable.

Customer-accessible audit API. OpenTelemetry export to your SIEM. Article 30 records generated automatically. SIEM integration on day one.

// audit.event.routing
request_id: "req_3f7a...b12"
tenant: "tenant_nordic_health"
tier_classified: T1
classifier: "deterministic"
policy_mode: "strict"
route: "on_prem_vllm"
model: "gpt-sw3-30b"
signed: "sha256:8d2f..."
Yggdrasil
13 · Capabilities
§13 · Capabilities

Eleven capabilities. One governed platform.

C-01
★
Secure Router
C-02
★
Observability
C-03
Model Access
C-04
Knowledge
C-05
Prompts
C-06
Agents
C-07
Workflows
C-08
Evaluation
C-09
Governance
C-10
Data Integration
C-11
Connectors

★ Cross-cutting · Passes through every request · Nothing bypasses.

Yggdrasil
14 · Governance
§14 · Governance is the platform

Compliance is not a feature.

GDPR

Data stays in EU jurisdiction for T1/T2. Article 30 records automatic. Right-to-erasure API.

Schrems II

Addressed by architecture. No US-controlled infrastructure in the T1/T2 data path.

EU AI Act

Human-in-the-loop. Transparency declarations. Conformity documentation.

NIS2

Security controls, incident detection, reporting workflows built in.

ISO/IEC 27001

Operates within Advania's certified management system.

ISO 9001

Operates within Advania's ISO 9001:2015 quality management system.

Yggdrasil
15 · Deployment
§15 · Deployment tiers

Two ways to deploy. Swedish soil either way.

Shared tier

Advania's Nordic GPU pool

  • ◆ Consume from Advania's Nordic GPU pool
  • ◆ Curated model catalog per policy mode
  • ◆ Multi-tenant with namespace isolation
Best for: Teams scaling AI adoption, variable workloads.
Dedicated tier

Your own GPU cluster

  • ◆ Your own GPU cluster on Advania infrastructure
  • ◆ Full model catalog, tenant-level customisation
  • ◆ Physical tenant isolation
Best for: Regulated sectors, Strict mode, high-volume workloads.
Yggdrasil
16 · Pricing
§16 · Pricing structure

Pricing reflects what you're buying: a trust layer.

01
Platform subscription
Annual · per tenant
Access to governance, observability, admin console, connectors, SLA-backed support. The trust layer.
02
Inference consumption
Shared tier
On-demand per token. Provisioned throughput on roadmap.
03
Dedicated capacity
Dedicated tier
Committed GPU cluster, all-in pricing aligned to allocated capacity.

Indicative pricing under NDA. Tailored per workload and support tier.

Yggdrasil
17 · Steps
§17 · Three steps to deployment

Discovery → Pilot → Design partner.

01
60 minutes

Discovery session

Structured conversation. Map AI usage, data classification, compliance obligations. Output: one-page sovereignty assessment.

02
Two weeks

Pilot scope

One high-value use case. Two-week pilot. Success criteria up front. Sandbox runtime.

03
Thirty days

Design partner

Formal agreement. Preferred commercial terms. Roadmap influence. Privileged GA access.

Yggdrasil
18 · Book a session
§18 · Next step

Book a discovery session.

60 minutes. No prep required. You'll walk away with a one-page sovereignty assessment specific to your organisation, whether or not you engage further.

Book now: ai@advania.com Read the platform overview →
MR
Matías Rodríguez Paz
Head of AI Product · Advania
EN
Errol Norlum
Chief Data & AI Officer · Advania
Slide overview · Press Esc to close · Click a slide to jump
←/→ · Space · Esc for overview · F for fullscreen
01 / 18