The problem

The cloud bill you can't predict.

European companies run their stack on US hyperscalers — where the meter, not the bandwidth, sets the bill:

• Everything is metered — every request, operation, and gigabyte, ingress and egress. The bill tracks activity, not bandwidth, and it moves every month.
• You can't forecast it. Usage spikes, the invoice spikes — after the fact.
• It gets worse at scale. At mid-scale the metered bill runs several times owned-hardware cost — same workload.

The Mittelstand wants European infrastructure. What it gets is a US invoice with a German postcode.

Why now

Three trade-offs just became one product.

• Sovereignty is policy, not sentiment — EU digital-sovereignty and GDPR are hardening from preference into procurement requirement.
• AI compute demand is exploding — the workload most exposed to residency and cost pressure.
• Bill-shock fatigue is real — buyers now put "what will this actually cost?" on the whiteboard.
• And now it's buildable, too — an AI agent fleet does the operating labour that used to need a datacentre team, so small teams can run big infrastructure.

Sovereign + affordable + reproducible was a pick-two. It is now one product.

The wedge

Pay for capacity, not activity.

• Flat, predictable, and yours to forecast — owned European capacity and bandwidth, priced like a colo. No per-operation meter, ingress is free, and the number is one you sized — not last month's traffic.
• Same APIs — migration is a config change, not a rewrite.
• EU residency you control by architecture — GDPR-native, on-device where the data allows.
• Scale-to-zero underneath — idle costs us almost nothing, so flat pricing is the architecture. (newer platform IP · [BUILDING])

Storage is the near-term wedge. Instrumented compute is the compounding arc.

The moat

The operating system, not the GPUs.

Anyone can rent GPUs. The defensibility is in how the platform is run:

• An AI agent fleet operates the infrastructure — durable execution, a persistent knowledge graph, full IaC reproducibility, and energy-aware scheduling, all without a human in the loop.
• Event-driven, scale-to-zero underneath — wakes on real work, idles at near-zero cost. (capability · [BUILDING])
• A small team runs infrastructure that normally needs many — that cost structure is the margin.

The control plane is the product. The racks are where it earns.

Team · founder fit

Built by an operator who ships.

James Westover — founder / CTO, building real products with an AI fleet.

• Sonobi — led an AWS→bare-metal migration — ~10× lower cost per impression at 250M+ daily impressions. (This business, already done once.)
• Uber, Staff Engineer — ran Location Platform at 10M+ requests/second; founded Uber's company-wide Python Platform — built 200+ shared libraries that carried ~90% of its Python-2 services (≈400 of ~4-5k microservices) to Python 3; GPS anti-fraud — $30M+ in prevented losses.
• HERE — autonomous-driving data pipelines at 250M+ events/day; grew the capture team from 2 to 14 engineers.
• PhD physicist (ABD) + MBA — systems thinking with a P&L vocabulary.

Berlin-based · Silicon Valley pedigree · GDPR by design. Building with an AI fleet.

The ask

€2M seed · ~24-month runway

An accelerant, not burn — the flywheel already spins on consulting revenue and grants; this buys speed.

Range: floor €1.2M · target €2M · every rack is a measurement, not a bet.
James Westover · [email protected] · westoverlabs.eu

Business model

Sell width, never a meter.

• Free — a narrow pipe. Throttled pressure, not a metered volume.
• Base — a wide pipe. Generous flat capacity; widen for pressure, never a GB cap.
• Higher tiers — wider pipes. Width is the only dial — ingress free, never per-op.

Where we price: €10/TB-month, flat, ingress-free, no egress meter, EU-sovereign, owned-hardware — below the hyperscaler, sovereign where the discount clouds are not.

One pricing dial — width. The meter is the thing we replace.

Unit economics · MODELED, pre-validation

Attractive — and honest about what's still a model.

Input prices (HDD, power, colo, transit, competitor storage) are real, current market quotes; the product economics are pre-validation — no external €10/TB customer yet, and we say so. The demand-gated rack ladder de-risks the two binding uncertainties (price realization, fill) one measured rung at a time.

Capital ladder

Gated capital, not a moonshot.

Capital is deployed on a ladder, each rung unlocked by evidence from the last:

• Customer demand → validates the first node
• Node telemetry → validates the workload economics
• Measured utilization → validates more compute
• Facility load + tariffs → validate energy and storage

An instrumented sovereign-compute flywheel — prove the economics on iron before scaling them. Every rack is a measurement, not a bet.

Go-to-market

A funnel that already earns.

• Free — a bot delivers an honest first read on AI-readiness, a repo review, and a security check. Indexable per specialist and built to be found.
• €49 Repo Health Check (paid hero) — the 3–5 things worth fixing first, walked through with you.
• €350 repo-wide review — adversarial bot review; grant write access and we open the fix PRs, characterization tests first.
• €500 "with the expert" — the review plus James's hour and a board-ready action plan. A few slots a week, by application.

The AI engineering team is itself a live, bookable revenue lane — consulting funds the runway today.

The team is the thesis

A founder + an AI fleet running a real company.

• The agent fleet plans, codes, reviews, ships, and operates the infrastructure — around the clock. The operating model on the "working system" and "moat" slides is the team.
• the platform monorepo — ~3,400 pull requests (2,800+ merged) in ~12 months, agent-driven: an AI reviewer gates and a fleet of named agents ships the work.
• Ship velocity, proven: Beatrice — an on-device, fully-offline voice-to-voice AI reading companion (iOS), taken from an empty repo to a working voice demo and a CI-built app in a single day, agent-driven, now on TestFlight.

We don't demo the company with the fleet. The fleet runs the company you're being pitched by.