The AI landscape is split between two paradigms. On one side, proprietary models from OpenAI, Anthropic, and Google operate behind API walls with closely guarded model weights. On the other, Meta's Llama, Mistral, and a growing ecosystem of open-source models offer comparable capability with open weights, permissive licences, and community development.
For investors and technology leaders, this is not an abstract debate. The choice between open-source and proprietary AI has direct implications for the intangible assets a company builds, the competitive moats it can defend, and the valuation it can command.
Proprietary AI development — building custom models on proprietary data with internal engineering teams — creates several categories of intangible assets:
Proprietary models and algorithms. Custom-developed AI models are technology intangible assets that can be protected through trade secrets (and in some jurisdictions, patents). They have measurable replacement cost and can generate attributable revenue.
Training data assets. Proprietary datasets used to train and fine-tune models are data assets with independent value. Unlike the model itself (which depreciates rapidly), high-quality proprietary data retains value across model generations.
Organisational knowledge. The team's expertise in building, training, deploying, and maintaining AI systems constitutes human capital — an intangible asset that is difficult to replicate and essential for continued AI capability.
Open-source AI creates a different intangible asset profile:
Integration and customisation expertise. The value lies not in the model (which anyone can access) but in the team's ability to adapt, fine-tune, and deploy open-source models effectively. This is human capital rather than technology capital.
Fine-tuning data and processes. While the base model is open, the company's fine-tuning data and training recipes create proprietary capability layered on top of a shared foundation.
Vendor independence. Open-source AI avoids API lock-in, meaning the company controls its AI infrastructure and can switch or modify models without vendor permission. This reduces risk but does not directly create a competitive moat.
Proprietary AI creates technology assets that appear (or should appear) on the balance sheet. Open-source AI creates human capital and process assets that do not. Both approaches can generate competitive advantage, but the nature and durability of that advantage differs fundamentally — and so does the valuation impact.
The critical question for investors is which approach creates more durable competitive advantage.
| Factor | Proprietary AI | Open-source AI |
|---|---|---|
| Model defensibility | High (trade secret protection) | Low (model weights are public) |
| Data defensibility | High (same for both, depends on data) | High (same for both) |
| Replication time for competitor | 1-3 years | Days to weeks (for base model) |
| Vendor lock-in risk | Low (you own everything) | Low (you control everything) |
| Cost to maintain | High (full stack responsibility) | Medium (community maintains base) |
| Talent requirements | High (need full ML team) | Medium (need deployment/fine-tuning skills) |
| Innovation speed | Self-dependent | Benefits from community innovation |
| Regulatory compliance | Easier (full control, full documentation) | Harder (provenance questions on base model) |
Proprietary AI is more defensible when:
Open-source AI creates better outcomes when:
Two healthcare AI companies provide clinical decision support. Company A built a proprietary model on licensed medical literature and 5 years of clinical outcome data — a genuine data moat. Company B fine-tuned Llama 3 on the same medical literature (publicly available) with 18 months of clinical data. Company A's intangible asset position is significantly stronger: proprietary model, deeper data moat, and longer replication time. Company B has lower costs but weaker defensibility.
Proprietary AI assets can be valued using:
Open-source AI deployments are harder to value as discrete assets because the base model has no scarcity value. The value resides in:
The total value of an open-source AI deployment is typically lower than a comparable proprietary system — but so is the investment. Return on investment may favour open source even when absolute asset value is lower.
Companies claiming proprietary AI capability that is actually a thin layer on top of open-source models represent AI washing risk. If the "proprietary AI" can be replicated by anyone with access to the same open-source model and a modest fine-tuning budget, the valuation premium is unjustified. Due diligence must distinguish genuine proprietary AI from open-source wrappers.
In practice, most sophisticated AI organisations use a hybrid approach: open-source foundation models fine-tuned on proprietary data and deployed through proprietary infrastructure. This captures the cost efficiency and innovation speed of open source while building defensible intangible assets through proprietary data and deep integration.
The Opagio Growth Platform helps organisations assess their AI intangible asset portfolio, including the balance between proprietary and open-source components and the defensibility of each layer.
The choice between open-source and proprietary AI is ultimately an intangible asset strategy decision. Proprietary AI creates larger, more defensible technology assets but at higher cost and risk. Open-source AI creates smaller, less defensible assets but with better cost efficiency and faster innovation. For most organisations, the optimal strategy is hybrid: open-source foundations with proprietary data and integration layers. For investors, the key is assessing which layers of the AI stack are genuinely proprietary and defensible — because that is where the durable value resides.
Ivan Gowan is Founder and CEO of Opagio. His technology leadership at IG Group (LSE: IGG) included evaluating build-vs-buy decisions across the full technology stack. Learn more about the Opagio team.
Not all AI creates competitive advantage. This article examines the four types of AI moats — data, model, integration, and network — and provides a valuation framework for assessing the defensibility and durability of proprietary AI systems.
Read more →Large language models have become core business assets, but valuing them is uniquely challenging. This article examines the economic characteristics of LLMs that affect valuation, the distinction between foundation and fine-tuned models, and practical approaches to LLM asset valuation.
Read more →AI-enabled companies trade at 25-40% higher multiples than peers, but not all AI capability is created equal. This article examines what drives the AI valuation premium, how to assess whether it is justified, and what happens when the premium unravels.
Read more →Get the latest insights on intangible asset growth and productivity delivered to your inbox.
Book a free consultation to see how the Opagio Growth Platform can help your business.