In The Press: Ausbiz - Does Google’s nuclear deal reveal the next AI opportunity?

Daniel Reaper, Portfolio Manager, joined AusbizTV to discuss why power and not demand is the binding constraint on the AI data centre build-out, and how nuclear baseload (via Constellation Energy) and gas bridging solutions are emerging as key ways to play it. Daniel notes there is what he describes as localised pushback starting to happen on the build out of new data centres, however, this doesn’t mean they’re not being build, it means they are simply pivoting to new locations, though this does impact speed of buildout and costs as sites become more competitive.

Watch Here: Does Google’s nuclear deal reveal the next AI opportunity?

AI is not running out of demand. It is running out of power.

The narrative around AI data centres has turned noisy. Water use, local opposition, higher power bills, and talk of a bubble have all featured heavily. Much of the water debate is overstated. Older facilities relied on evaporative cooling towers that consumed large volumes of water. Newer AI data centres largely use closed-loop systems, recycling roughly 90–95% of the coolant. The political friction is real, but the physical constraint that actually matters is electricity.

Data centre electricity consumption was about 415 TWh in 2024—roughly 1.5% of global supply. It is projected to roughly double to around 950 TWh by 2030, lifting its share to near 3%. That remains a modest slice of the overall grid, but it represents a large share of incremental demand. Local pushback is already slowing projects. In the US, roughly 120 projects worth nearly $200 billion were blocked or delayed in the first half of 2026, and about 70% of Americans say they would oppose a data centre in their own town. Europe restricted or rejected more than 70 projects in a four-month window, worth around $42 billion. Australia is seeing the same pattern. Capacity is still being built, Goldman Sachs expects US data centre capacity to reach 64 GW by the end of 2026 and 90 GW by the end of 2027—but timelines are stretching and sites are being rerouted.

The core problem is simple: without firm power, the data centres do not get built at the pace the AI build-out requires.

Nuclear as the long-term baseload solution

Consistent baseload power is the missing piece, and nuclear is the technology best suited to supply it at scale. We are less interested in pure uranium mining exposure further up the supply chain—those equities tend to be volatile and speculative. The more attractive positioning sits with companies that own and sell the power itself.

Constellation Energy is the clearest example. It has secured multi-decade offtake agreements with hyperscalers. Microsoft has contracted the entire output of the restarted Three Mile Island Unit 1 (roughly 835 MW) for 20 years. Unit 2, which suffered the 1979 partial meltdown, is not returning. Unit 1 was shut in 2019 on economics; the Microsoft contract has made the restart viable.

Google’s arrangement is larger still—3.6 GW across two contracts. The first covers 890 MW of uprates on 11 existing Constellation reactors over 20 years, with initial megawatts expected from 2028 and full delivery by the end of 2032. This is effectively the equivalent of a large new reactor without constructing one. The second is 2,700 MW of existing generation for 15 years. Constellation continues to sell into the wholesale market, but Google has locked in a long-term slice of that output. These deals illustrate the pricing power that accrues to owners of reliable generation when the buyers are the hyperscalers themselves. Constellation is a multi-year holding in this theme.

Broader nuclear exposure can also be accessed via ETFs such as NUKZ (Range Nuclear Renaissance) and NLR (VanEck Uranium and Nuclear). Small modular reactors remain an important longer-term growth vector—Google, Amazon and Meta have signed agreements with Kairos, X-energy, Oklo and TerraPower—but none of these projects has yet delivered power. First electrons are unlikely before 2030 at the earliest. Constellation’s current value is in the existing fleet: restarts and uprates.