For the past two years, markets have treated AI as a semiconductor story. And honestly, that made sense. Chips are tangible, exciting, and sit at the glamorous end of the value chain where margins are fat and every earnings beat feels like confirmation of a new industrial era.
But markets have a habit of crowding around the obvious winners first. The more interesting opportunity is usually one layer below.
AI is not just a compute story anymore. It is a power story. Every new model, every inference workload, every new data centre needs something far less exciting than a next-gen GPU. It needs electricity. Grid access. Cooling. Storage. Transmission. A system that can deliver reliable power at scale. And that is starting to matter just as much as the silicon.
The IEA has been fairly blunt about this. Global electricity demand is growing hard through 2030, and annual grid investment needs to rise by around 50 percent just to keep up. There are already more than 2,500 gigawatts of projects sitting in grid connection queues worldwide, renewables, storage, data centres, all waiting. That is not a minor operational headache. That is a structural bottleneck, and it is sitting right in the middle of the AI buildout.
So the question is no longer whether AI demand exists. It clearly does. The question is whether the physical world can keep pace.
The second-order winners are starting to show up
The first wave of AI beneficiaries was predictable: chipmakers, cloud platforms, software businesses with a convincing productivity story. The next wave looks a lot more industrial. Utilities. Grid equipment manufacturers. Electrical contractors. Transmission specialists. Battery storage operators. Cooling infrastructure providers. Businesses that have never once put the words “artificial intelligence” in a slide deck, but are sitting right in the path of where the capital needs to go.
This is classic second-order investing. The market narrative catches up to the underlying economics slowly, and the gap between those two things is where returns tend to live.
Part of what makes this interesting is the sheer nature of how AI demand lands. It does not arrive gradually. A data centre is not a theoretical increment on the grid. It is an enormous electricity customer that wants power immediately, wants it reliably, and does not care that grid upgrades and transmission projects can take years to approve and build.
Google is already signing agreements with utilities to curtail up to a gigawatt of data centre load during peak demand periods. Read that again. One of the most sophisticated technology companies in the world is now actively managing its energy flexibility as a core part of running its AI infrastructure. That tells you something important about where the real constraint is forming.
When the marginal unit of AI growth depends on access to power, scarcity shifts. Cheap electricity becomes a strategic asset. Grid stability stops being a regulatory footnote and starts driving earnings. The businesses that can connect, firm, manage, or supply power into this ecosystem gain real pricing power, even if they never show up in a fund manager’s tech basket.
The timing mismatch nobody is pricing properly
Here is the tension that the market still seems to be underestimating. AI capacity can scale in months. Physical energy infrastructure cannot. A new software application ships in weeks. A grid upgrade takes years. A transmission line can take longer than that.
Battery storage helps bridge the gap, but duration matters more than people realise. Reuters recently reported surging interest in long-duration energy storage technologies, pointing to a Google project in Minnesota pairing a 300 MW installation with a 30 GWh iron-air battery system capable of 100 hours of storage. That is a meaningful step beyond the short-duration battery model that has dominated investment conversations until now. It is also a sign that the industry is starting to think seriously about system architecture rather than just generation capacity.
Generating more electrons is not the whole answer. You also have to move them, store them, shape them, and deliver them reliably when needed. Transmission, interconnection, storage duration, demand response, behind-the-meter solutions, cooling and water infrastructure. These are not peripheral details. They are the bottleneck.
Why Australia is worth paying attention to
There is a tendency in local markets to assume the real AI opportunities are all offshore, ideally in a large US technology stock. That instinct deserves some pushback.
Australia’s power system is already showing what the next phase looks like. AEMO reported that renewables including storage exceeded 50 percent of the National Electricity Market’s quarterly energy mix for the first time in the December 2025 quarter. Battery discharge nearly tripled year on year. The pipeline of new projects connecting to the NEM hit a record 64 gigawatts. That is not just an environmental story. It is an investable infrastructure story.
A more renewable-heavy grid creates a premium for flexibility, storage, and grid intelligence. Intermittent generation can be enormously valuable, but only when the broader system can handle variability. AI data centres do not want intermittent power. They want continuous, high-quality, reliable supply. That gap between how renewables generate and how digital infrastructure consumes is exactly where value gets created.
You can already see this playing out locally. AirTrunk’s battery investment attached to its western Sydney data centre is a sign that operators know they cannot simply demand power from the grid and assume the system will absorb it. They need to be part of the solution. That shift matters, and it signals that the next phase of AI infrastructure is going to be more capital intensive, more regulated, and more industrial than most investors seem to expect.
Old economy businesses in new economy value chains
When a hot theme runs into physical constraints, markets eventually rediscover industrial businesses. Not because they become fashionable, but because they own the scarce assets everyone suddenly needs. That tends to produce a sharp rerating in companies tied to engineering services, energy networks, electrification, specialist manufacturing, and infrastructure software.
Australia has plenty of listed businesses sitting in the middle of these value chains without being priced as AI beneficiaries. The company supplying electrical balance of plant, designing substations, upgrading transmission systems, or enabling industrial cooling may have just as much earnings leverage to AI infrastructure growth as a more obviously branded technology name, and in some cases better valuation support to go with it.
The risks are real too
This is not a one-way trade. Rapid data centre expansion is already generating community pushback across parts of the United States over electricity bills, water usage, and local impact. Microsoft’s president acknowledged this week that winning community trust is now a genuine operational constraint, with opposition already cancelling projects in some areas.
The path from demand forecast to actual deployment is rarely clean. Planning delays, political resistance, and regulatory shifts can all slow how quickly any of this monetises. Buying anything adjacent to electricity and declaring victory is not a strategy. The better opportunities are in businesses with genuine competitive advantages, disciplined balance sheets, and exposure to durable bottlenecks rather than just the headline theme.
The bigger picture
Every major technology cycle follows roughly the same arc. The first phase rewards invention. The second rewards deployment. The third rewards the infrastructure that makes adoption possible at scale. Railways needed steel. The internet needed fibre. Cloud computing needed hyperscale campuses. AI needs power. Literally. Not as a metaphor.
When an industry starts talking in gigawatts rather than gigabytes, the businesses that can deliver electricity, move it, stabilise it, and store it deserve a lot more attention than they are currently getting.
That is where the bottleneck is forming. And bottlenecks, identified early, are usually where the best returns are found.
The TAMIM Takeaway
The market’s fixation on chips and software may be causing it to miss the next wave of AI beneficiaries. The real constraint is increasingly electrical infrastructure. Grids, storage, transmission, cooling, and flexible power management are becoming strategic assets in the AI buildout. For investors, that opens up a broader opportunity set in industrial and infrastructure-linked businesses that are not yet being priced as AI winners. The next phase of the AI boom may have less to do with glamour and more to do with transformers, substations, batteries, and grid access.