Sacra estimates that Multiverse Computing generated €11.1M in recognized revenue in FY2025 (roughly $12.5M), up approximately 297% from €2.8M in FY2024. The company reached approximately €100M in annualized recurring revenue in January 2026, up from €11.1M at the end of 2025.

FY2025 revenue came entirely from services based on quantum and AI solutions, with the Singularity optimization platform accounting for most recognized income. CompactifAI model compression and deployment began contributing in the second half of 2025, but most large contracts signed late in the year had not yet flowed fully into statutory revenue at year-end.

The gap between €11.1M in recognized FY2025 revenue and ~€100M in January 2026 ARR stems from the timing of large enterprise and OEM contracts with high annualized values that were signed or began delivering near year-end. Customer count remained around 100 across 10 industries, indicating that growth came from larger contract sizes and expansion within existing accounts rather than logo acquisition. Named customers and partners include Allianz, BBVA, Bosch, Iberdrola, Telefónica, and PwC.

The company reported that Q1 2026 sales were 96 times Q1 2025 levels and that annualized revenue had grown more than 10x since the June 2025 Series B. FY2025 net loss was approximately €24M, amid R&D and international expansion spending, while the company carried €43.3M in positive working capital at the end of 2025.

In July 2026, Multiverse Computing announced a Series C targeting up to €500M ($570M) at a €1.5B ($1.7B) pre-money valuation; the round was reported at a $2.3B post-money valuation. Participants include Forgepoint Capital International, BNP Paribas Solar Impulse Venture Fund, Bullhound Capital, Santander Alternative Investments, Tikehau Capital, HP Inc., Orange Ventures, Scania Invest, NAventures (National Bank of Canada), Qatar Development Bank, Zouk Capital, SETT, the EIC Fund, Hazten Scale-Up Fund, and Kutxa Fundazioa.

The company previously closed a €189M ($215M) Series B in June 2025, led by Bullhound Capital, with participation from HP Tech Ventures, SETT, Forgepoint, CDP Venture Capital, Santander Climate VC, Quantonation, and Toshiba.

Founded in 2019 in San Sebastián, Spain, Multiverse Computing raised earlier seed and Series A rounds, including an oversubscribed €25M Series A. Total lifetime funding is reported at $897.4M.

Multiverse Computing develops software that reduces the size and operating cost of AI models, as well as quantum-inspired optimization tools for financial and industrial problems. Both product families use tensor networks, a technique from quantum physics that represents complex data structures in compact form.

Multiverse's model compressor, CompactifAI, restructures a large language model's internal weight matrices into smaller, interconnected mathematical objects called tensor networks. Unlike conventional pruning or quantization tools, which delete parameters or lower numerical precision, CompactifAI rewrites parts of the model's architecture to store a compact recipe for reconstructing important relationships rather than a full table of numbers. Multiverse reports that supported configurations can require 50–95% less memory and run on fewer or cheaper processors, typically with accuracy losses in the low single digits.

Developers and enterprise teams can access CompactifAI through a hosted API, select a model from its catalog, replace the endpoint, and pay per token. The catalog includes Multiverse's Quasar 438B and HyperNova 60B, alongside compressed versions of Llama, Mistral, Qwen, and other models. For organizations that keep data on their own infrastructure, Multiverse produces compressed model variants for deployment on-premises, in a private cloud, or on edge devices. A mobile app launched in March 2026 runs small models offline on phones and tablets, while a routing layer can send more complex queries to the cloud when connectivity is available.

Singularity, the company's original optimization platform, allows portfolio managers and operations analysts to load constraints into Excel or Python, select a solver, and receive optimized asset weights, schedules, or pricing outputs without writing quantum code. It supports classical, quantum-inspired, and hybrid quantum-classical backends. Applications include portfolio optimization, derivatives pricing, FX trading, and risk modeling, with early deployments at BBVA, Crédit Agricole, and the Bank of Canada.

Luminary, launched in September 2026, helps customers select models for deployment. It generates synthetic customer journeys, runs candidate models through multi-turn conversations with tool calls and policy constraints, and compares success rate, compliance, latency, and cost per completed task. The product can feed customers into CompactifAI by testing whether a smaller, lower-cost compressed model performs as well as a frontier model on a specific workflow.

Multiverse Computing sells to enterprises, financial institutions, telecom operators, device manufacturers, defense organizations, and governments through direct sales, AWS cloud marketplace distribution, and channel partnerships with systems integrators such as EY and Plain Concepts.

The company uses several monetization models. The CompactifAI API charges per million input and output tokens, like a commercial inference endpoint, with lower cost of revenue because compressed models require fewer GPUs per request. Private and on-premises deployments are individually quoted and likely bundle a model license, compression fees, integration work, and annual support. Singularity is licensed SaaS delivered through SDKs, APIs, and Excel plugins, with paid implementation projects. Luminary could add subscription revenue from repeatable model evaluation and compliance testing.

The tensor-network methods that reduce customer infrastructure costs also lower the GPU spend required to host the CompactifAI API. Better compression can improve model performance while reducing cost of revenue. Multiverse's cost structure consists primarily of specialized research talent in tensor networks and quantum algorithms, GPU compute for model healing and inference hosting, and enterprise solution engineering across multiple geographies.

Expansion is consumption-led for the API and account-led for enterprise contracts. As a customer's application gains traffic, token consumption grows. Enterprise and OEM compression projects can expand into fleet-wide device licensing, sovereign infrastructure deployments, or cross-sales from CompactifAI to Singularity optimization. The company grew from four founders to over 160 staff by mid-2025, with the Series C intended to fund further expansion across the US, Canada, Europe, Asia, and the Middle East.

Multiverse Computing competes across two markets, efficient AI infrastructure and quantum-inspired optimization. Competitors include vertically integrated platform owners, independent compression specialists, and classical optimization incumbents.

NVIDIA is Multiverse's most strategically important competitor in AI compression. Its Model Optimizer and TensorRT-LLM bundle quantization, pruning, distillation, sparsity, and optimized kernels at zero incremental cost for customers already running NVIDIA GPUs. The tools are coupled to NVIDIA hardware and distributed through its existing cloud, OEM, and enterprise channels.

Multiverse competes through hardware independence: CompactifAI models can run on Intel Xeon CPUs, edge devices, and non-NVIDIA accelerators, as well as GPUs. A July 2026 demonstration showed compressed Llama 3.3 70B running on Intel Xeon 6 with roughly doubled throughput. NVIDIA could incorporate comparable factorization techniques or make conventional compression sufficient for most GPU-centric customers.

Pruna AI offers a unified open-source framework that combines quantization, pruning, distillation, and compilation in a single developer workflow. Its open-source distribution lowers adoption friction, while support extends beyond LLMs to image, video, and speech models. Ora Computing is an emerging tensor-network specialist whose technical approach overlaps directly with CompactifAI.

Multiverse has deeper enterprise relationships, more financing, and a larger proprietary research base. However, Pruna can incorporate new open-source tensor-factorization methods without replicating Multiverse's research organization. Ora presents a separate risk: tensor-network compression could become a recognized category rather than a Multiverse-specific capability.

Singularity competes with mature classical solvers from Gurobi, IBM CPLEX, and FICO Xpress, which benefit from decades of validation, large operations-research talent pools, and existing enterprise licenses. These tools are easier to audit, explain to regulators, and integrate into legacy risk and portfolio systems.

Among quantum-inspired platforms, Fujitsu's Digital Annealer and Toshiba's SQBM+ target the same combinatorial optimization problems, backed by established enterprise procurement relationships and global systems-integration capacity. D-Wave is both a competitor and an infrastructure supplier because Singularity has historically included D-Wave's Leap Hybrid solver. Quantagonia's HybridSolver competes with Singularity's hardware-agnostic approach through familiar MILP and QUBO interfaces. Financial-services firms spend more than $4B annually on high-performance computing potentially addressable by quantum solutions, but classical solvers retain advantages in reliability, explainability, and procurement credibility.

Multiverse Computing's expansion strategy runs along three axes: moving from compression into a full AI deployment stack, pushing models from data centers onto edge devices and sovereign infrastructure, and entering new geographies and industry verticals.

CompactifAI started as a model compression engine and has expanded into an efficient-AI platform spanning hosted inference, private deployment, model evaluation, and model development. The company now distributes its own models, including Quasar 438B, alongside compressed variants of third-party models. Luminary adds an evaluation layer that can direct customers toward CompactifAI deployments.

The forthcoming Foundry product would add GPU cluster management, workload monitoring, and serverless AI, creating a control plane that captures recurring infrastructure fees. Partnerships with Cohere for enterprise AI and Cerebrium for serverless inference integrate Multiverse technology into enterprise procurement rather than limiting it to standalone compression projects.

Compression expands Multiverse's addressable market from data-center inference into smartphones, PCs, vehicles, drones, telecom equipment, and industrial systems. The CompactifAI mobile app runs offline AI on phones, while OEM relationships with HP and Scania indicate a potential route to per-device licensing across hardware product lines.

Sovereign AI represents an expansion vector in Europe, Canada, and the Middle East, where regulated institutions seek to deploy models inside controlled infrastructure without relying on US hyperscalers. The Arsys collaboration on the European 8ra sovereign-cloud initiative and the PINQ² partnership in Canada provide institutional routes into government and defense procurement. Quasar's status as a European-built model could differentiate it commercially as the EU AI Act increases demand for auditable, locally deployed models.

The July 2026 agreement with EY targets specialized models for financial services, government, telecom, and energy, combining Multiverse's compression technology with EY's distribution and sector relationships. The Mitiga partnership translates climate-risk data into asset damage and underwriting metrics, extending Multiverse from infrastructure software into AI-native decision products.

Geographically, the company is expanding from its European base into the US (San Francisco office), Canada (PINQ² and NAventures), the Middle East (Qatar Development Bank investment), and Asia (Marubeni partnership for Japan rollout). The US provides access to the largest enterprise software ecosystem, the Middle East offers sovereign fund capital and national AI programs, and Asia's electronics and automotive manufacturing base creates potential OEM channels for compressed on-device models.

Compression commoditization: NVIDIA's Model Optimizer already bundles quantization, pruning, and distillation at zero incremental cost to GPU customers, open-source frameworks such as llama.cpp and vLLM provide free alternatives, and foundation-model providers increasingly release optimized small variants directly, which could reduce CompactifAI's tensor-network approach from a differentiated capability to an unnecessary layer if simpler methods prove sufficient for most production workloads.

Revenue concentration and recognition risk: The jump from €11.1M of FY2025 recognized revenue to ~€100M of January 2026 ARR, with a roughly stable customer count of ~100, implies dependence on a small number of large contracts, while the company's prior FY2025 revenue forecast of €30M versus the €11.1M statutory outcome indicates that annualized metrics and management projections can diverge materially from audited results.

Product sprawl: Singularity, CompactifAI, proprietary models, the hosted API, the mobile app, Luminary, and the forthcoming Foundry distribute R&D, sales, and management attention across seven partially overlapping products while the company remains unprofitable, with FY2025 losses of approximately €24M.