TLDR: GB300 NVL72 systems are now setting the new standard for AI infrastructure, representing a significant leap beyond even the impressive GB200 deployments. From hyperscale cloud providers to specialized AI infrastructure companies, the scale and sophistication required for commercial competitiveness is becoming clear. Australia must ensure it can supercharge these systems without breaking the bank: competitive energy prices, sensible policy, and interstate cooperation (or competition) might determine if Australia can power the next generation of AI right here on home soil, or if these rack-scale supercomputers end up humming somewhere else in the world instead. The window for establishing Australia as a viable AI infrastructure hub is narrowing as companies worldwide rapidly scale their deployments in more cost-effective regions. Once these investments are made offshore it is unlikely it will make economic sense to move them back in the next decade.

NVIDIA’s next tranche of AI chips, the GB300, has arrived and it is both more efficient and more powerful than the GB200. In a groundbreaking announcement this month, Dell and CoreWeave have delivered the first NVIDIA GB300 NVL72 rack. An iteration upon the GB200 chip, the NVL72 configuration packs 36 Grace CPUs and 72 Blackwell GPUs into a single rack. The switch to GB300 series chips pushes the performance density of these rack-scale compute units even further. Hyperscalers and researchers at the major AI research labs are clamboring to place orders as these systems become the gold standard for AI research and inference. Access to this kind of compute efficiency will power new kinds of agentic and real-time services that will make the capabilities supported by tools like ChatGPT and Copilot currently look modest by comparison.

This latest iteration redefines the key metrics related to AI compute—it is no longer about how fast the compute density is or how much memory per chip NVIDIA has managed to pack in: It’s all about energy.

image

These next-generation systems act like one giant GPU for industrial-scale AI workloads. But bleeding-edge performance doesn’t come cheaply: the NVL72 configurations will reportedly set you back about USD $3 million per rack, according to HSBC. However, the initial capital investment isn’t the only challenge with these new systems: they devour electricity.

Both GB200 and GB300 NVL72 racks consume around 120kW of power for the compute hardware alone—what’s remarkable about the GB300 is that it delivers 50% more performance than GB200 within essentially the same power envelope (the GB300 uses only marginally more power). Dell’s deployment with CoreWeave features liquid-cooled Dell IR7000 racks specifically designed to handle these enormous power demands. Powering and cooling these ultra-high-density systems means energy costs become a make-or-break factor. This brings us to a pressing question: Is Australia, with its current energy prices and policies, a competitive place to run these next-generation AI “factories”?

Energy-Hungry AI: Why Power Costs Matter

According to the International Energy Agency, global data centre electricity consumption is set to more than double to around 945 TWh by 2030. Despite the focus on efficiency, the absolute power needed is still enormous. Consuming 120kW continuously, a single NVL72 rack could translate to about 1 GWh per year. Cheap power isn’t just nice to have—it’s a strategic advantage when running large-scale AI infrastructure. Power efficiency means more compute capacity for the same running cost—that means cheaper and more ubiquitous AI services.

While the GB200/GB300 racks themselves might consume ~120kW, the complete system including cooling infrastructure could easily push total power consumption to 180-250kW per rack in hot climates or facilities without optimal cooling tower access.

According to Supermicro’s specifications for GB200 NVL72 systems, the liquid cooling infrastructure requires up to 250kW capacity CDU with redundant PSU, plus 1.3MW capacity in-row CDU, and additional power for a possible 180kW/240kW capacity liquid-to-air solutions to chill the coolant before reuse. This means that beyond the 120-132kW consumed by the compute hardware itself, operators need substantial additional power for the cooling infrastructure—potentially adding another 50-100% to the total power consumption.

The Commercial Reality: Can Australia Compete in the Global AI Infrastructure Race?

The AI infrastructure boom is accelerating globally, with major deployments happening across multiple vendors and cloud providers. The Dell-CoreWeave partnership is just one example—companies like Microsoft, Google, Amazon, and Meta are all rapidly scaling their AI infrastructure, while specialized providers like CoreWeave, Lambda Labs, and others compete for enterprise customers. These deployments consistently favor regions with favorable energy costs and policies, sending a clear signal about what matters for commercial viability.

For Australia to attract commercially-viable AI investment, it needs to demonstrate not just that it can power these systems, but that it can do so cost-effectively at enterprise scale.

It is a basic economic calculus that tells the story: A modest 10-rack deployment of these advanced systems would consume roughly 1.2 MW continuously for the compute hardware alone. Including cooling infrastructure, the complete system could reach a total power consumption of 1.8-2.5 MW.

At Australia’s average electricity price of $0.27/kWh, this translates to $4.3-5.9 million per year in electricity costs. The same deployment in Texas or other low-cost US regions at $0.08/kWh would cost just $1.3-1.8 million annually—a difference of $3-4 million per year. Over a typical 5-year infrastructure lifecycle, this represents $15-20 million in additional operating costs that Australian operators must somehow absorb or pass on to customers. Put simply, every cent per kWh difference adds roughly $175,000-220,000 per year to operating costs for a 10-rack deployment.

Its not all bad news: Homegrown cooling innovators

Interestingly, Australia does have some advantages in advanced cooling technologies. DUG Technology, based in Western Australia, has developed sophisticated immersion cooling solutions that can dramatically reduce cooling power overhead compared to traditional liquid cooling systems. DUG’s immersion cooling technology, originally developed for their own high-performance computing operations in Perth, could potentially offer a more energy-efficient path for AI infrastructure deployment in Australia’s challenging climate conditions.

Australia is also pioneering innovative natural cooling solutions. The Pawsey Supercomputing Centre in Perth utilizes a groundbreaking geothermal cooling system that pumps 21°C water from the Mullaloo aquifer through heat exchangers before reinjecting it back into the aquifer. This CSIRO-developed system is estimated to save 14.5 million litres of water compared to conventional cooling towers in its first two years alone, and is designed to scale with future hardware additions. The system carefully manages thermal energy while avoiding negative impacts on the aquifer through strategic cold-water injection bores.

However, even with these innovative cooling solutions, the core economic challenge persists: while better cooling technology can reduce total power consumption, the electricity used still costs three times more in Australia than in competing regions like the US or China. In other words, saving 30% on cooling power through advanced technology still leaves Australian operators paying substantially more per kilowatt than their international competitors.

Global Power Price Check: US, China, Europe… and Australia

To gauge Australia’s competitiveness, let’s compare electricity costs globally. Broadly, the U.S. and China enjoy relatively cheap industrial power. Industry electricity prices in both countries hover around $0.08 USD per kWh (8 cents) on average. China’s rates have inched up only slightly in recent years—about $0.088 in 2024, from $0.084 in 2019—thanks to deliberate policies to keep energy costs low for manufacturers. The United States likewise benefits from abundant domestic energy; American industry power prices increased only ~21% from 2019–2023, and remain in the single-digit cents per kWh in many regions.

Now contrast that with Europe. In the wake of the 2022 energy crisis, European electricity costs spiked dramatically, underscoring a serious competitive gap. By 2023, industrial power prices in the EU were about 158% higher than in the U.S. on average. Major EU economies that rely on expensive imported gas saw huge jumps—e.g. Poland’s industrial rates up 137%, the UK’s up 124% since 2019. Even traditionally lower-cost France saw a ~93% hike (though France’s prices stayed below the European average, buoyed in part by its nuclear fleet). Germany, after years of Energiewende policies and heavy taxes, has some of the priciest electricity globally (~~$0.36/kWh for households), though large industrial users often receive partial relief. The net effect is that Europe’s high energy costs are holding back its tech and manufacturing growth, and EU leaders worry about businesses fleeing to regions with cheaper power.

However, Europe does have a significant R&D advantage that Australia lacks: IMEC (Interuniversity Microelectronics Centre), the world’s leading semiconductor research institute. Recognizing the importance of energy costs for advanced manufacturing, IMEC is establishing its first facility outside Belgium in Málaga, Spain, with the Spanish government committing €100 million for the initial phase. Spain’s selection isn’t coincidental—it has some of the lowest industrial electricity costs in Europe, making it an attractive location for energy-intensive semiconductor research and development. This strategic investment demonstrates how even within high-cost Europe, countries with relatively cheaper energy can attract world-class technology infrastructure.

So where does Australia fit in? Unfortunately, Australian electricity isn’t cheap by global standards, especially not since recent price rises. The average electricity price in Australia (all sectors) was roughly $0.27 USD per kWh as of early 2024—more than three times U.S. or China levels. We pay well above what Americans or Chinese do for power. A few years ago, Australia’s prices had actually dipped to an 8-year low (around 19 US cents/kWh in 2021) thanks to increased renewables and lower demand. But the global gas turmoil in 2022 hit Australia’s east coast hard as well: wholesale electricity rates surged, and retail tariffs followed. So today, running an AI data center in Australia means paying a premium for energy—an obvious disadvantage when energy is a primary operating cost.

This stark gap in energy costs is causing prospective investors to ask: what is Australia doing to rein in energy prices, and is relief in sight?

The State-by-State Power Price Reality in Australia

One complicating factor is that Australia’s energy costs aren’t uniform nationwide. State policies, resources, and grid setups vary, leading to notable differences in electricity prices across the country. For example, Victoria and South Australia—states that have rapidly expanded wind and solar generation—have recently seen significantly lower wholesale electricity prices than the coal-dependent states. In Q4 2024, Victoria’s average wholesale price plunged to just A$45/MWh, about half the national average, thanks to lots of renewable energy pushing prices down. South Australia (SA), with its high renewables mix, similarly enjoyed low prices, even frequently dropping to negative prices during windy, sunny periods. Tasmania, powered largely by hydro, also tends to have moderate wholesale prices. Meanwhile, the northern NEM states (New South Wales and Queensland) were hit with prices of A$127–143/MWh in the same quarter—roughly triple Victoria’s rate—due to heavy reliance on aging coal plants and expensive gas. Those coal stations not only face higher fuel costs but also more frequent outages, driving up “cap payments” (price spikes) for reliability.

Average wholesale electricity spot prices by region in Australia (Q4 2023 vs Q3 2024 vs Q4 2024). States with higher renewable generation (VIC, SA, TAS) saw substantially lower prices by late 2024, whereas NSW and QLD experienced price spikes due to reliance on high-cost coal power, which averaged over $120/MWh on the National Energy Market in 2024 according to the Australian Energy Regulator report.

Even Western Australia (WA), which isn’t connected to the eastern National Electricity Market, merits a look—and in fact, it might be Australia’s dark horse for cheap energy. WA runs its own grid (the SWIS) and has a unique policy: a domestic gas reservation. Gas producers in WA must reserve 15% of output for local use, ensuring local gas prices stay relatively low (often half the price east coast users pay). This paid off big during the 2022 crisis. While Eastern states saw wholesale electricity costs skyrocket to record levels (the June quarter 2022 averaged an “unprecedented” A $284/MWh on the NEM), Western Australia’s average electricity price was around A $64/MWh—literally about one-quarter the cost—because its gas-fired generators had access to cheap fuel and the state had less exposure to global LNG prices. WA’s grid is also less coal-intensive (around 38% coal vs ~60% in the east, circa 2022), and is rapidly adding renewables.

Not all Aussie states are equal for powering an AI data center. If you’re purely shopping for the lowest cents per kWh, recent data suggests Victoria (in periods of high renewables output) and Western Australia (with its domestic gas reservation policy) could be front-runners. Queensland and New South Wales, on the other hand, have had the highest wholesale prices of late (thanks to the high price of coal), which would directly hurt an energy-intensive AI operation’s bottom line. Tasmania offers clean hydro power and has attracted some industrial users with special energy deals in the past, though its total grid capacity is smaller. South Australia has abundant renewables, but one must ensure backup supply for when the wind isn’t blowing (SA often leans on battery storage and interstate interconnects).

For an investor eyeing Australia, these differences mean it’s worth comparing state policies and power contracts carefully. The “best” state could be one that offers a mix of low cost and reliable supply. Right now, Western Australia is making an aggressive pitch that it’s that place—touting the lowest delivered electricity pricing in the nation and huge investments in new solar/wind projects to stay that way. Meanwhile, Victoria is proudly advertising its falling prices as renewable projects come online. The competitive dynamic even within Australia might benefit energy buyers, as states vie to lure data center projects.

Policy Pain Points: What’s Driving Australia’s High Energy Costs?

To understand how Australia can improve, we have to examine which policy settings might be hurting its energy competitiveness. A few key factors stand out:

  • Gas and Fuel Costs: Until recently, eastern Australia lacked a domestic gas reservation policy, meaning local gas generators had to pay export-level prices for fuel. When global gas spiked in 2022, electricity prices on the east coast blew out accordingly. This was a policy choice—unlike WA, the east didn’t shield itself from international price swings. The federal government scrambled to implement temporary caps on gas (and coal) prices to tame power bills. While those interventions helped somewhat, they were a reactive band-aid. In the long run, either securing affordable domestic gas supply (through reservation or price agreements) or rapidly replacing gas/coal with cheaper renewables and storage is necessary to avoid repeat crises. Western Australia’s example shows that a stable domestic energy supply policy can directly translate to more stable (and lower) electricity prices. The east coast is now considering a permanent gas reservation mechanism, but implementation and timing remain uncertain.

  • Slow Transition & Investment Uncertainty: Australia, especially the NEM states, is in the midst of an energy transition—closing old coal plants, building renewables, adding transmission and storage. This is a delicate dance. If policy is unclear or changes frequently, investors get skittish about funding new generation capacity. In recent years, there’s been policy whiplash (e.g. shifting renewable targets, debates over capacity markets, state vs federal energy plans). Uncertainty can delay the very investments (wind farms, batteries, pumped hydro, etc.) that would increase supply and lower prices in the long term. The “coal cliff” is a real concern: if multiple big coal stations retire before replacements are ready, supply scarcity drives up prices. Stable, long-horizon policy—with incentives for new dispatchable capacity—is needed to ensure the transition doesn’t create price spikes. The periodic price volatility in NSW/QLD is partly a symptom of an aging fleet struggling to meet demand; clear policy on what fills that gap (gas? pumped storage? solar + big batteries? maybe even lifting the ban on nuclear down the track) will affect price stability.

  • Network and Regulatory Costs: Australia’s grid covers long distances and a relatively small population, meaning transmission and distribution costs are significant in the end price. Critics of Australian energy policy such as Tony Wood from the Grattan Institute have long pointed out that regulatory overhead and gold-plated network investments have kept Aussie power prices high historically. Policies that encourage smarter, more efficient network spending (and integration of distributed resources) could ease some pressure. Moreover, lengthy approval processes for new transmission lines (needed to connect renewable-rich regions to load centers) can slow down getting cheap renewable power to market. Streamlining these processes is more policy opportunity than a “harm” per se, but it’s crucial for enabling lower-cost energy in the future. Recent analysis by the Institute of Energy and Economic Financial Analysis, however, suggest that little has changed to address long-standing issues in energy infrastructure policy.

  • Environmental Schemes and Taxes: Ironically, while renewables are driving down wholesale prices, the way Australia structures some of its green schemes can add costs elsewhere. Programs like the Small-scale Renewable Energy Scheme (solar credits) and various state renewable targets do add a few percentage points to bills. Carbon policy is another wildcard—currently Australia has no nationwide carbon tax on electricity, but the Safeguard Mechanism requires big emitters to cut or offset emissions. If power generators pass on the cost of carbon offsets, that could nudge prices up. At the same time, global investors increasingly demand clean energy, so the greener grid may be a competitiveness necessity as much as an upfront cost. There’s a balance to strike between short-term price relief and long-term sustainability commitments.

In summary, Australia’s high electricity prices have been partly self-inflicted—through policy choices like no east-coast gas reservation, under-investment in new generation until recently, and some regulatory missteps—and partly a factor of geography and timing. The encouraging news is that many of these factors are recognised and being addressed (slowly). The federal and state governments are now investing heavily in renewables and storage, and talking up measures to ensure reliability (e.g. capacity incentives). But energy infrastructure changes can take years, and until Australia decisively lowers its cost of electricity, it risks being a less attractive location for energy-intensive industries like AI computing.

Which Australian State is Best for an AI Supercomputer?

If a prospective investor asked today where in Australia they should put a large-scale AI deployment—potentially dozens of liquid-cooled GB300 NVL72 racks consuming tens of megawatts—the answer would likely be “wherever the power is cheapest, most reliable, and ready for these next-generation power demands.” Based on current trends and policies, a strong case can be made for Western Australia as the top choice. WA enjoys lower wholesale prices than any NEM state at the moment, has a government actively courting data centers, and is expanding its renewable energy capacity fast. It also has improved data connectivity internationally (new subsea fiber links to Asia/Middle East), addressing the historical knock on Perth being “far away” digitally. From a policy perspective, WA’s relatively insulated market and pro-industry stance (e.g. funding the Pawsey supercomputing center, highlighting its “clean energy” potential) make it attractive.

Crucially, WA also has a home-grown advantage in cooling technology: DUG Technology’s immersion cooling solutions could provide significant operational advantages for AI infrastructure, potentially reducing the cooling power overhead that makes these deployments so energy-intensive. An AI data center powered by WA’s planned 50GW Western Green Energy Hub, combined with advanced immersion cooling technology, could offer both low cost and 100% renewable credentials while minimizing the cooling power penalty.

That said, Victoria shouldn’t be counted out. Vic has the second-largest electricity grid and is aggressively pushing toward renewables (targeting 95% renewables by 2035). The revival of the State Electricity Commission to build renewables is aimed at permanently lowering prices. If the Q4 2024 pattern holds, Victoria could sustain very low wholesale prices in spring and beyond. For an AI facility, Victoria also offers proximity to Australia’s tech hub (Melbourne) and major research universities—and it straddles the balance of decent power cost with good infrastructure. The main risk is ensuring supply stays reliable as coal plants retire; however, Vic is investing in transmission to bring in offshore wind and new storage which should help.

What about New South Wales or Queensland? At present, their energy costs are less favorable—but they have other advantages like larger existing data center clusters (especially Sydney), and major cloud availability zones. If one’s priority is lowest electricity cost, NSW/QLD would rank lower today. But they shouldn’t be dismissed for the long term: both states are aiming for big renewable builds (Queensland’s plan includes 22 GW of renewables plus pumped hydro by 2035). Tasmania offers very green power and has been pitching “Battery of the Nation” projects to supply the mainland; a data center in Tasmania could leverage cheap hydro if you secure a dedicated contract, though connectivity and scale of grid are considerations.

In a nutshell: Western Australia currently leads on pure energy price and policy friendliness for AI centers, with Victoria a close second especially as its renewable fleet grows. States like South Australia and Tasmania offer clean power and could be competitive for smaller-scale or specialized deployments, while NSW/Queensland need to get their costs down to stay in the race for hosting AI mega-projects.

Recommendations: Powering Australia’s AI Future

Australia finds itself at an energy crossroads. To capitalize on the AI boom (and perhaps become an “AI compute hub” in the Asia-Pacific), it needs to get serious about delivering abundant, cheap, clean power. Based on the analysis above, here are some opinionated recommendations:

  • Double Down on Renewable Energy + Storage: The quickest path to lower energy costs in Australia is through more low-cost renewables, paired with storage to make them reliable. The states that have injected more wind, solar, and batteries (Victoria, SA) are already seeing wholesale price benefits. Australia has world-class solar and wind resources; turning that into a competitive advantage for industry means continuing to aggressively build and connect these resources. Policies should ensure there’s enough firming capacity (batteries, pumped hydro, fast-start gas or even future technologies like hydrogen turbines) to prevent volatility. The goal should be to drive wholesale prices down into the ~$40–50/MWh range consistently (which Victoria hit in late 2024) and keep them there. That level would make Australian power much more competitive with the US/China on a fuel cost basis.

  • Adopt Smart Energy Policy Reforms: Lessons from Western Australia’s gas reservation should inform east coast policy. Ensuring domestic energy security—whether via reserving some gas, or eventually producing zero-marginal-cost renewable power in excess—is key to shielding consumers from global shocks. Additionally, providing long-term price certainty to big energy users could attract AI and HPC investments. This might involve government-backed renewable energy supply contracts or special tariffs for hyperscale data centers that commit to using green power. Other countries (like some US states) offer discounted electricity rates or tax breaks to data centers; Australia could consider that, or at least streamline approvals, to entice investors. On the flip side, avoid ad-hoc interventions that scare off investment—energy policy should be consistent enough that companies feel confident in long-term electricity pricing when planning a 10+ year data center project.

  • Leverage Australia’s Strengths (Clean and Green): One thing Australia can offer is clean energy at scale, which will be increasingly important. Tech giants and AI research labs are under pressure to run on carbon-free energy. Australia’s rapid decarbonization of its grid (projected 82% renewables by 2030 nationally) is a selling point. Combine that with initiatives like the Pawsey Supercomputing Centre in WA—one of the world’s greenest supercomputers—and Australia can brand itself as a destination for sustainable AI computation. Policymakers should highlight and expand such programs. Perhaps establish “Green Data Center Zones” with guaranteed renewable power and modern grid infrastructure. This would play to Australia’s image while also potentially qualifying projects for ESG-focused investment funds.

  • Address Infrastructure and Connectivity: Lastly, ensure that supporting infrastructure (fiber connectivity, land, cooling water if needed) keeps up. The energy cost might be moot if a data center in Australia can’t get enough bandwidth to serve global users. The recent completion of a submarine cable linking Perth directly to international networks is a great example—more of this will integrate Australian data centers into the global fabric. Governments could assist with infrastructure grants or fast-tracking permits for data center builds in regions with surplus power.

Australia has a golden opportunity to become a major player in the AI revolution—but only if it can solve its energy cost challenges. The stark reality is that running a GB200 NVL72-based AI cluster in Texas, Guizhou, or Norway costs significantly less than operating one in New South Wales today. This competitive disadvantage threatens Australia’s potential as an AI hub. However, with decisive policy action to reduce energy costs, Australia could transform into a compelling destination for AI infrastructure. The race is on between states to deliver affordable power—and whichever state succeeds will attract the next generation of AI compute “factories.” Western Australia’s combination of cheap gas and massive solar resources makes it a strong contender, while Victoria’s ambitious renewable energy targets could help it take the lead. The question is whether state governments will move quickly and boldly enough to seize this opportunity before it slips away to other countries.

Western Australia’s model of leveraging local resources and pro-investment policy gives it an edge today. Yet, nationwide, Australia needs to hasten its transition to an affordable clean energy system to not miss out on the AI gold rush. A future where “Made in Australia” AI runs on low-cost, green Australian electrons is within reach—and that would be a win-win for both the tech industry and Australians at large, delivering innovation, jobs, and lower emissions.

Sources