AI Investments Create Intangible Assets
AI Value Assessment — Lesson 2 of 10
When a company spends $1 million on a new manufacturing machine, the asset appears on the balance sheet the next day. When the same company spends $1 million building an AI-powered demand forecasting system, the balance sheet often shows nothing but an expense. The machine depreciates visibly over its useful life. The AI investment creates assets that compound in value — but remain invisible to conventional accounting.
This invisibility is the core distortion in how organisations and markets value AI investment. Every AI programme, whether it succeeds or fails at its primary objective, produces a portfolio of intangible assets that have standalone economic value. Understanding what those assets are, how they accumulate, and how to measure them transforms the way boards and investors evaluate AI spending.
AI investment is not simply an operating expense. It is capital formation — creating data assets, trained models, algorithmic IP, and organisational capability that have measurable economic value. Organisations that recognise and track these assets make fundamentally better investment decisions than those that treat AI as a cost.
The Four Asset Classes of AI Investment
Every AI programme, regardless of its specific application, produces assets that fall into four distinct categories. Each has different characteristics, different valuation approaches, and different strategic implications.
Asset Class 1: Curated Data Assets
Raw data is not an asset. Curated, structured, labelled, and validated data is. The process of building an AI system requires transforming messy, incomplete, siloed data into clean, connected datasets that can train models. This transformation — which often represents 60-80% of an AI project's cost — creates a data asset with properties that persist long after the original project concludes.
A curated dataset can train multiple models. It can be combined with other datasets to create new analytical capabilities. It can be licensed to partners or sold outright. In M&A transactions, acquirers increasingly assign specific valuations to proprietary data assets, particularly when the data cannot be readily replicated.
A logistics company spent $3.2 million over two years building a route optimisation AI. The model itself delivered $1.8 million per year in fuel and time savings. But the curated dataset — 18 months of GPS traces, delivery timestamps, traffic patterns, and weather correlations across 12,000 routes — was independently valued at $7.5 million during a strategic review. Three separate AI initiatives subsequently used the same data asset, reducing their development costs by an estimated 40% each.
Asset Class 2: Trained Models and Algorithms
A trained machine learning model is the product of three expensive inputs: data, compute, and expertise. Once trained, the model can generate predictions, classifications, or recommendations at near-zero marginal cost. The model itself — its architecture, its weights, its performance characteristics — is an intangible asset with identifiable economic value.
Trained models meet the identifiability criteria under IAS 38: they can be separated from the entity (licensed, sold, or transferred), and they generate measurable future economic benefits. In practice, however, most organisations do not recognise models as assets because they are developed internally and the accounting standards impose strict recognition thresholds for internally generated intangible assets.
Asset Class 3: Algorithmic IP and Technical Architecture
Beyond the specific models, AI programmes create broader technical assets: feature engineering pipelines, model serving infrastructure, monitoring frameworks, and the architectural patterns that enable AI deployment at scale. These assets reduce the cost and risk of future AI initiatives, creating a compounding advantage.
Organisations that have built robust MLOps (machine learning operations) infrastructure can deploy new models in days rather than months. This speed advantage translates directly into competitive positioning and revenue — a new product recommendation model, for instance, generates value from the moment it goes live.
Asset Class 4: Organisational AI Capability
The least visible but often most valuable asset class is organisational capability: the skills, processes, cultural norms, and institutional knowledge that enable an organisation to conceive, build, deploy, and govern AI effectively. This includes the tacit knowledge held by data scientists, the cross-functional collaboration patterns between engineering and business teams, and the governance frameworks that ensure responsible AI deployment.
AI Intangible Asset Classes
| Asset Class | Examples | Persistence | Transferability |
|---|---|---|---|
| Curated Data | Training datasets, labelled corpora, feature stores | Long-lived; value increases with volume | High — can be licensed or sold |
| Trained Models | Prediction models, classifiers, recommendation engines | Medium — requires retraining as data shifts | Medium — transferable but context-dependent |
| Algorithmic IP | MLOps pipelines, feature engineering, model architectures | Long-lived; reusable across projects | High — can be productised |
| Organisational Capability | AI literacy, governance processes, cross-functional workflows | Long-lived but fragile (key person risk) | Low — embedded in organisational culture |
How Assets Accumulate Over Time
AI assets follow a compounding pattern that is fundamentally different from tangible asset depreciation. A machine loses value from the day it is installed. A curated dataset gains value as more data is added. A trained model improves as it processes more real-world examples. An organisation's AI capability deepens with every project completed.
This compounding characteristic has two important implications.
First, early AI investments create disproportionate long-term value because they establish the foundation — the data, the infrastructure, the skills — upon which all subsequent AI initiatives build. Cancelling an AI programme after a disappointing first year may destroy assets that would have generated substantial returns over three to five years.
Second, the gap between AI-mature and AI-immature organisations widens over time. An organisation with five years of curated data, battle-tested models, and experienced AI teams can deploy new AI initiatives faster, cheaper, and more reliably than a competitor starting from scratch. This is the AI equivalent of the competitive moat — and it is an intangible asset.
The compounding nature of AI assets creates a strategic paradox for CFOs. The projects with the lowest first-year ROI often create the highest long-term asset value, because they are the foundational investments in data infrastructure and organisational capability. Evaluating AI solely on short-term returns systematically biases against the most valuable investments.
Mapping AI Assets to Accounting Standards
The question of whether and how AI assets appear on the balance sheet depends on whether they were acquired or internally generated.
Acquired AI Assets
- Recognised at fair value under IFRS 3
- Separately identifiable in purchase price allocation
- Amortised over estimated useful life
- Subject to annual impairment testing
Internally Generated AI Assets
- Mostly expensed as incurred under IAS 38
- Development costs may be capitalised if strict criteria are met
- Data assets rarely recognised on balance sheet
- Organisational capability never recognised
Under IAS 38, internally generated intangible assets can only be recognised on the balance sheet during the development phase (not the research phase), and only when the entity can demonstrate: technical feasibility, intention to complete, ability to use or sell, probable future economic benefits, availability of resources, and reliable measurement of expenditure.
In practice, many AI projects blur the line between research and development, making it difficult to identify the precise point at which capitalisation should begin. The result is that most internally generated AI assets never appear on the balance sheet — creating the measurement gap discussed in Lesson 1.
This accounting treatment does not mean the assets have no value. It means that management, investors, and boards must develop supplementary frameworks for tracking and valuing AI assets outside the formal financial statements. This is where the Opagio Valuator and the AI ROI dashboard (covered in Lesson 8) provide structured approaches.
Practical Asset Identification
For any AI initiative, the asset identification exercise is straightforward. At each stage of the AI lifecycle, ask: what assets have been created that persist beyond this specific project?
Audit your data assets
Catalogue every curated dataset, labelled corpus, and feature store created by AI projects. Note volume, freshness, uniqueness, and potential for reuse.
Inventory your trained models
Document every production model: its purpose, accuracy, training data, compute cost, and estimated replacement value.
Map your technical infrastructure
Identify reusable pipelines, feature engineering code, model serving platforms, and monitoring tools that accelerate future AI work.
Assess organisational capability
Evaluate your team's AI skills, governance maturity, and cross-functional collaboration patterns. Identify key person dependencies.
What Comes Next
This lesson has established that AI investment creates four classes of intangible assets, each with distinct characteristics and valuation implications. In Lesson 3: The 4-Layer AI ROI Framework, we introduce the structured framework for measuring AI value across cost reduction, revenue growth, competitive advantage, and strategic optionality — integrating the asset creation perspective with operational financial metrics.
Ivan Gowan is CEO of Opagio, the growth platform that helps businesses and investors measure, manage, and grow intangible assets. Before founding Opagio, Ivan held senior technology and leadership roles across financial services and digital platforms for 25 years. Meet the team.