Intrica — Quantum-inspired AI for the optimal decision under uncertainty

01The problem

The hardest problems in finance and energy are decisions under uncertainty.

Optimal control over many interacting variables — what to hold, when to dispatch, how to hedge — explodes combinatorially. Classical methods like Monte Carlo and dynamic programming slow to a crawl or fall back on approximations, exactly where the value is.

Latency

Too slow.

Solutions arrive after the decision window has already closed.

Fidelity

Too coarse.

Approximations smooth over the tail risk and the upside that matter most.

Scale

Doesn’t scale.

Every added asset, constraint, or scenario multiplies the cost.

The space of possible futures explodes exponentially. Predicting it is not the same as knowing what to do — that is the gap.

02The breakthrough

A new primitive for AI.

Computing the optimal decision under uncertainty is an open problem in AI — the possibilities explode faster than any model can handle. We’re building a new kind of model that fuses modern machine learning with a breakthrough from quantum physics, where that same explosion was tamed decades ago: compressing it into something computable, without losing what matters.

Modern AI

Neural nets · Transformers · LLMs

Compute probabilistic tokens. Trained on data.

Quantum algorithms

Optimization · Circuits

Compute exponentially large states. Derived from equations.

Tensor network models

Agile as neural networks. Rigorous and powerful as quantum physics.

→Learns the structure of a problem.A trainable model class with the expressive reach of neural networks, plus an explicit dial — the bond dimension — that trades accuracy against cost under provable error bounds. The learning runs deeper than data: active-learning algorithms probe a problem’s equations directly, detecting the hidden, quantum-like structure that tensor networks exploit to capture an astronomically large space with only a few parameters.

→Solves otherwise-intractable equations.Classical solvers discretise space into a grid, and the number of points blows up exponentially with dimension. Tensor networks store the solution in compressed, low-rank form — making high-dimensional partial differential equations (PDEs) solvable where finite-difference and finite-element methods break down.

→Reasons under uncertainty.Probabilistic inference, done exactly: the full joint distribution is held compactly as a network; contracting it returns marginals, conditionals, normalising constants, and expectations directly. No sampling and no stochastic error — the same query returns the same answer every time.

→Generates the full range of outcomes.As generative models — Born machines — tensor networks learn a distribution, expose an exact and tractable likelihood for any configuration, and draw independent samples directly, without the mixing and convergence problems of Markov chain Monte Carlo.

Every result is deterministic, auditable, and orders of magnitude faster than simulation — on classical hardware.Quantum physicists spent thirty years uncovering a secret: behind impossibly complex systems lies a simple, repeating structure you can compress onto an ordinary computer. Intrica’s founding insight is that markets, power grids, and supply chains are built the same way. Modern AI lets the engine learn that hidden structure and exploit it — computing the best move outright instead of guessing at it, far faster, with no quantum computer needed.

03How it works

From a problem to the optimal call, in three steps.

One engine, one workflow — the approach we’re building, designed to fit a problem and run on the hardware you already have.

01 · Represent

Represent

We cast a control problem as a tensor network — capturing the full high-dimensional structure, no shortcuts.

02 · Compress & solve

Compress & solve

Quantum-inspired algorithms shrink the state space and solve it — answers in a usable timeframe, at a fidelity classical methods can’t match.

03 · Calibrate

Calibrate

Machine learning tunes the model to live market and operational data — decisions grounded in reality, not assumptions.

On classical hardwareNo quantum computer required. Architected to ride quantum hardware as it matures.

04Use cases

The first markets.

One engine for wherever high-dimensional uncertainty meets real money — where the gap between a defensible answer and the optimal one is measured in capital, risk, and P&L. Benchmarked in finance.

Finance

Option pricing and risk under correlated uncertainty, and portfolio optimization under real-world constraints — fast enough to run intraday, so a desk can act while the market is still open. Under the Fundamental Review of the Trading Book (FRTB), market-risk capital at top US banks rises $100B → $138B — making faster, auditable pricing a capital question.

See our benchmarks

Energy & commodities

Stacking revenue across several markets at once, coordinating entire battery fleets, and weighing today’s profit against long-term battery wear — the hard version of the problem, where every decision constrains the next and simple scheduling leaves money on the table.

Pricing derivatives under correlated uncertainty is one of the hardest computational problems in finance — the foundation every trading and risk decision is built on, where Monte Carlo has been the standard for twenty years. We benchmarked European option pricing under Black–Scholes — five assets with all-to-all correlations — sweeping the required accuracy.

Monte Carlo · incumbent

Sampling

Draws millions of random scenarios and averages them into an estimate — accuracy improving only as 1/√N, every run seed-dependent.

Intrica

Compression

Holds the full range of outcomes in one structured object — solved once, deterministically: the same inputs always return the same answer.

Our results

~410×

Faster than Monte Carlo
at 10⁻⁶ relative accuracy

27×

At 10⁻⁵ — the advantage
compounds as precision tightens

100%

Deterministic
& auditable

fig. 01 — speedup over Monte Carlo vs required relative accuracy, log–log · European options, Black–Scholes · 5 assets, all-to-all correlations · crosses parity near 3×10⁻⁴ and grows to ~410× at 10⁻⁶

Finance and energy are the first markets, not the last — the same engine computes the optimal decision wherever high-dimensional uncertainty meets high stakes.

05Proof

Already real, with paying partners.

Paid pilots underway in finance and energy, plus direct lines into the institutions at the frontier of European quantum technology.

Commercial

Finance · Energy

Paid proofs-of-concept underway with Tier-1 customers in both sectors.

Network

University of Geneva
Innosuisse
Open Quantum InstituteCERN
CEA GrenobleHub Quantique

06Why us

A rare combination, built for exactly this.

Deep expertise in tensor-network algorithms and machine learning, with direct lines into Europe’s quantum institutions. Based in Geneva, Switzerland.

Julian Thoenniss, PhD

Project Lead

Tensor-network modelling; built the prototype behind the speedup benchmark for option pricing. Innosuisse grant awardee.

Prof. Xavier Waintal

Scientific Advisor

Director of Research at CEA Grenoble and of the Maison du Quantique, Alpes. Senior author of Kwant, a foundational quantum-transport library.

Samuel Nyckees, PhD

Quantum algorithm engineer

Tensor networks for deep quantum simulation and neural networks.

Thomas Kloss, PhD

Quantum-algorithm researcher

Frontier tensor-network computation; GPU computing; code library developer.

Quantum-inspired AI for the decisions classical models can’t compute.