Datalane .se
SE · STOCKHOLM 59.33°N · 18.07°E
Pilot customers · Q3 2026 Sweden / EU

Sovereign AI
inference, made in
Sweden.

Datalane runs any open-weight model on hardware we own and operate inside Swedish borders. An OpenAI-compatible API for European teams that need their AI to live somewhere defensible.

Request early access → Why sovereign infrastructure?
Jurisdiction
Sweden & EU only
API
OpenAI-compatible
Models
Any open weight
Billed in
SEK · EUR
Data residency
100% SE
Never leaves the EU
Sites
3 · active
Stockholm · Luleå · Malmö
Energy
100% renew.
Hydro + wind PPAs
Latency to Nordics
< 20ms
CPH · OSL · HEL
§ 01 — Sovereignty

Where your inference runs decides who controls your data.

Most AI inference today happens on US-headquartered hyperscalers, subject to US extraterritorial law. For European public sector and regulated industries, that's no longer an acceptable trade-off. Datalane closes the gap — same developer experience, different jurisdiction.

01 / 03
Jurisdiction

Data stays in Sweden.

All compute, storage and routing inside Swedish borders, operated by a Swedish AB with no foreign parent. No CLOUD Act exposure, no Schrems II hand-waving — your DPA names a Swedish court, full stop.

02 / 03
Independence

Owned hardware, owned stack.

We operate the GPUs, the network and the control plane. No upstream hyperscaler in the path between your application and a token. Per-tenant key custody — we cannot read your traffic.

03 / 03
Compliance

GDPR & EU AI Act, day one.

Built so your data-protection, ISO 27001 and EU AI Act provider documentation lines up out of the box. Audit log access included; sub-processor list short and stable.

§ 02 — Platform

One endpoint. Any open model.

Datalane is a serverless inference layer. You bring an open-weight model; we expose it behind an OpenAI-compatible API. Switching models is one line of code, not a procurement project.

  • Bring any open model. Llama, Mistral, Qwen, DeepSeek, Gemma — we host what you pick. Custom fine-tunes welcome (safetensors / GGUF).
  • OpenAI-compatible API. Point your existing SDK at api.datalane.se/v1 and ship — chat, completions, embeddings, tool use, streaming.
  • Serverless inference. Pay for tokens, not idle GPUs. Cold starts amortized across tenants; warm pools for steady traffic.
  • Per-tenant isolation. Dedicated namespaces, network and KMS keys. No shared cache, no prompt leakage between customers.
llama-3.3-70b mistral-large-2 qwen-2.5-72b deepseek-r1 gemma-3-27b + any open weight 
python · drop-in# OpenAI SDK · zero rewrite
from openai import OpenAI

client = OpenAI(
    base_url="https://api.datalane.se/v1",
    api_key=os.environ["DATALANE_KEY"],
)

response = client.chat.completions.create(
    model="llama-3.3-70b",
    messages=[
        {"role": "system", "content": "Du är hjälpsam."},
        {"role": "user",   "content": prompt},
    ],
    stream=True,
)
# → tokens streamed from Stockholm, never via US-DC
§ 03 — Infrastructure

Built in Sweden. Run in Sweden.

Stockholm is our primary GPU campus and control plane. Luleå handles bulk and long-context workloads on hydroelectric power. Malmö gives us an edge into continental Europe. Every byte stays inside Swedish jurisdiction.

69°N — Arctic Circle
65°N
62°N
59°N
55°N — Baltic
Luleå secondary SE-LLE-01 · hydro
Stockholm primary SE-STH-01 · control plane
Malmö peering SE-MMA-01 · EU edge
Fig. 02 · Three sites, one jurisdiction

Stockholm — SE-STH-01

Primary
59.33°N · 18.07°E · Stockholm metro
  • Main GPU campus — NVIDIA H200 & B200 clusters on InfiniBand fabric.
  • Control plane, customer onboarding, API gateway, KMS.
  • NetNod IX peering. <30ms RTT to Copenhagen, Oslo, Helsinki, Berlin.
  • Tier III equivalent; 2N power, N+1 cooling, on-site Swedish staff.

Luleå — SE-LLE-01

Secondary
65.58°N · 22.15°E · Norrbotten
  • Hydro-powered GPU pods for bulk batch & long-context workloads.
  • Free-air cooled; PUE under 1.15 year-round.
  • Fixed-price hydroelectric PPA — predictable energy cost.
  • Hot standby for control plane state; cross-site replication every 60s.

Malmö — SE-MMA-01

Peering
55.60°N · 13.00°E · Öresund region
  • Edge peering point for traffic routed to continental EU.
  • Latency shaving for German, Benelux and Polish customers.
  • No customer data at rest — pure transit and cache.
Silicon
H200 · B200
InfiniBand fabric · isolated tenancy
PUE
< 1.15
Free-air, hydro-powered north
Peering
NetNod
Stockholm IX + Malmö edge
Personnel
SE-cleared
No foreign access path
§ 04 — Pricing

Two ways to pay. Both in your currency.

Pricing in Swedish kronor (SEK) by default. Euro (EUR) on request — no dollar conversion, no FX surprises. Quoted on signup, contracted in your jurisdiction.

Mode · Flexible

Pay-as-you-go

Metered per million tokens, billed monthly. No commitment, no idle cost. Built for prototypes, internal tools and bursty production traffic.

  • Per-token metering, sub-second granularity
  • No minimum spend, scale to zero
  • Shared warm-pool capacity
  • Standard SLA, support during EU business hours
Quoted in SEK / Mtok · EUR on request
Mode · Fixed

Reserved capacity

Reserve dedicated GPU capacity at a fixed monthly rate. Priority scheduling, predictable latency, predictable invoicing. Built for production at scale.

  • Dedicated GPU pool, your model loaded warm
  • Priority scheduling, guaranteed throughput
  • 12-month term, predictable monthly invoice
  • 24/7 on-call, named technical account manager
Quoted in SEK / month · EUR on request

Indicative pricing shared under NDA during pilot onboarding. Public price list published when general availability opens, in Q4 2026.

§ 05 — Early access

Onboarding pilot
customers, now.

Tell us about your workload. We're prioritising Swedish public sector, finance, healthcare and EU-regulated industries for the first cohort.

  • One-week response. A founder replies, not a sales bot.
  • Pilot DPA included. Swedish law, data residency in writing.
  • Switch costs nothing. If you're on OpenAI today, you're an env-var away.
Replies within one Swedish working week.