AI infrastructure is booming, but margins depend on pricing power, electricity costs and tenant concentrationTechnology cycles tend to reward platforms and applications over physical assetsA handful of Australian companies sit in the AI capability layer rather than the real estate one Special Report: The AI boom has sparked a rush into data centres, but the long-term opportunity may lie with the firms developing models and sovereign capability.Australia is pouring billions into data centres. Private equity is backing AI “factories”. Listed landlords are expanding aggressively. Utilities are planning for sharp rises in water and electricity consumption. Data centre operators are moving to renewables, water treatment and more efficient chips.The infrastructure story dominates headlines. But the more strategic investment question may sit one layer above it.If the past 18 months have been about racks, land banks and megawatts, the next phase may be about who owns the intelligence running inside them.The crowded infrastructure tradeThe market has treated data centres as the cleanest exposure to the AI boom.Companies like NextDC (ASX:NXT) , CDC and Goodman Group (ASX:GMG) offer scale, recurring contracts and hyperscaler tenants. Firmus is preparing for an IPO after securing a $14 billion debt package led by Blackstone.The logic is straightforward. AI requires significant compute. Compute requires significant data centre infrastructure.But infrastructure is capital-intensive, exposed to grid approvals and energy costs, and increasingly reliant on a small group of global hyperscalers. The long-term return profile will depend not just on demand growth, but on pricing power, renewal terms and customer concentration.That shifts attention to the usage layer. The AI stack broadly splits into three tiers:Infrastructure Foundation models and orchestration Applications and vertical solutionsAustralia has committed heavily to the first. It has far fewer scaled players in the second and third. In previous technology cycles, the enabling layer mattered. The outsized returns flowed elsewhere.Railways enabled commerce. The merchants built the fortunes.Fibre enabled streaming. The platforms captured the value.AI is unlikely to be different.Training the modelsOne of the few listed Australian companies operating directly in the capability layer is Appen (ASX:APX) .The company provides training data and model evaluation services to major AI developers. It sits inside the model development process rather than on the real estate surrounding it.Appen’s share price volatility reflects the intensity of change in AI markets. But structurally, its exposure is to model refinement, scaling and deployment.As enterprises experiment with AI and refine outputs for commercial use, demand for curated, high-quality training data becomes more important, not less.That demand exists regardless of which data centre houses the GPUs.Intelligence at the edgeBrainChip Holdings (ASX:BRN) offers a different capability angle. Rather than building facilities, BrainChip develops neuromorphic processors designed for edge AI applications, allowing intelligence to run locally in devices rather than centrally in hyperscale facilities.That positioning aligns with a structural shift toward decentralised AI. If inference increasingly occurs in vehicles, industrial systems, defence platforms or connected devices, value migrates from pure centralised compute toward distributed intelligence.Commercial traction remains developing. But the strategic exposure is to capability, not construction.The sovereign opportunityThe most strategic opportunity sits with firms attempting to build domestic AI capability for regulated sectors.Sovereign Australia AI is developing foundational language models and secure deployment platforms tailored for Australian government, defence and critical infrastructure operators.Its thesis is not about leveraging global hyperscalers but about enabling local capability.Federal AI strategies increasingly emphasise sovereign control and regulatory alignment in sensitive industries. For many agencies and operators, offshore-hosted AI models are not suitable for mission-critical use.If government and regulated industries increasingly prioritise Australian-controlled AI systems, domestic providers could become important drivers of local compute demand.Similarly, while regulated industries have solved for sovereign data storage, AI is still often processing data offshore leaving possible exposure of critical data.That broadens the demand base beyond a handful of global tenants.The long gameThe recent volatility in global tech stocks and warnings about potential overbuild in AI infrastructure are reminders that capital cycles rarely move in straight lines.Data centres will remain essential but owning concrete alone does not guarantee exposure to the highest-margin layer of the value chain.The more sophisticated portfolio construction may involve pairing infrastructure exposure with capability exposure.Back the landlords but also back the companies building models, security frameworks and vertical AI tools that will monetise that infrastructure.This article was developed in collaboration with Sovereign Australia AI, a Stockhead advertiser at the time of publishing.This article does not constitute financial product advice. You should consider obtaining independent advice before making any financial decisions.Originally published as Beyond data centres: The real AI trade is capability not cables