Can Flow Batteries Lower AI Data Center Power Costs?

After years of shortages of computing hardware like GPUs and memory, and just when these shortages are easing, a new bottleneck is slowing down data center deployment: energy. This is because the new waves of data center construction are very compute-intensive and therefore also consume a lot of power.

This results in the inability of most power grids to accommodate this sudden multi-gigawatt surge in demand. In theory, renewables could provide an alternative, especially as solar and wind power have gotten a lot cheaper in the past years.

However, renewable energy generation tends to be intermittent, and obviously, solar power does not work at night. But AI data centers need to run continuously, with an uptime rate so high (“five or six nines” – 99.9999%) that any power shortage is unacceptable.

The alternative can be energy storage, both to bridge the gap during the nights and to compensate for lower production during less sunny days. If it works, independent microgrids can then power the data centers and avoid this bottleneck entirely.

A new study by two researchers at Sejong University analyzed which energy storage technologies are the best fit for AI data centers using an off-grid renewable microgrid.

They found that vanadium redox flow batteries could be the best fit for this application, even though it also seems that lithium-ion batteries become more competitive as data center load increases. The study also compares these options to hydrogen-based backup and flywheels.

They published their findings in the Journal of Engineering Research¹, under the title “Comparative techno-economic assessment of energy storage technologies in an off-grid renewable microgrid for AI data centers”.

AI Hunger For Power

Contrary to previous generations of data centers, AI data centers are more akin to heavy industries when it comes to power consumption. For example, hyperscale AI data centers can require power capacities exceeding 100 MW, equivalent to the annual electricity consumption of approximately 100,000 households. And many of the latest data centers are aiming for gigawatt-scale, equivalent to one million households’ consumption, or enough to consume the power generation of an entire large-scale nuclear reactor.

As a result, in order to preserve their ESG profile, hyperscalers and tech companies are looking to source that power from low-carbon sources.

“The International Energy Agency (IEA), about 50% of the additional electricity demand from data centers is expected to be supplied by renewable energy by 2035. Major technology companies such as Google and Microsoft have announced “24/7 carbon-free energy” targets for 2030, aiming to match electricity consumption with carbon-free electricity generation on a real-time basis”

Simulating Energy Storage Options

Building A Realistic Off-Grid AI Data Center

The study evaluates a hypothetical off-grid renewable microgrid for AI data centers under the regional conditions of Taebaek, South Korea. The mountainous location helps with an average summer temperature of approximately 26.0°C, while the average minimum winter temperature is about −9.6°C, providing favorable conditions for reducing cooling demand in data centers.

Another advantage of this site is that during lower solar production in winter, wind production peaks, letting the system balance its energy production seasonally.

Source: Journal of Engineering Research

It considers an integrated power supply system composed of photovoltaic (PV), wind turbines, hydrogen storage, and three options for energy storage systems (ESS) technologies: Li-ion batteries, vanadium redox flow batteries (VRFB), and flywheel.

The hydrogen energy storage subsystem consisting of an electrolyzer, H₂ tank, and fuel cell is included to provide long-term storage and to utilize surplus renewable electricity.

Source: Journal of Engineering Research

The HOMER Pro microgrid optimization software was used to optimize component sizing and to compare the economic performance of different ESS-based system configurations.

Designing An Accurate Simulation

Overall, data centers show a relatively stable base-load profile throughout the year because servers, cooling systems, and supporting electrical infrastructure operate continuously to maintain uninterrupted service. To account for limited variability under actual operating conditions, hourly and daily load fluctuations of ±5% and ±3%, respectively, were applied.

Source: Journal of Engineering Research

On the financial side, the project lifetime was set to 25 years, the expected inflation rate to 2.0%, and the nominal discount rate to 2.5%, so the real discount rate was calculated as 0.49%.

The capital, replacement, and O&M costs of each ESS were simultaneously varied to 80%, 100%, and 120% of their baseline values, and the system was reoptimized for each condition.

As a proxy for improved cooling performance, a simplified cooling-efficiency improvement scenario was evaluated by reducing the total facility electricity demand by 5% for the 1 MW data center.

For each potential design, the energy storage system chosen has different advantages:

• Li-ion is characterized by high efficiency and rapid response for short- to medium-duration balancing.
• VRFB is best for independently scalable power and energy capacities and suitability for long-duration storage and repeated deep-discharge operation.
• Flywheel displays rapid high-power response and short-duration storage capability.

The maximum annual capacity shortage was limited to 2%. The operating reserve requirements were set to 10% of the current load, 80% of the PV output, and 50% of the wind power output.

This is a planning-level comparison, not a demonstration of mission-critical reliability. Annual unmet electricity demand was 1.18% for lithium-ion, 0.684% for VRFB, and 0.588% for flywheels. These percentages measure unmet energy, not downtime, and cannot be directly converted into availability figures. The model also uses a simplified load profile rather than explicitly simulating dynamic GPU utilization and cooling demand.

Finding The Ideal Renewable Microgrid Combo

When analyzing the lithium-ion configuration, the setup with these batteries + full hydrogen system was the most economical.

“For the Li-ion case, the optimal system consisted of 4443 kW of PV, five WT units, a 500 kW electrolyzer, a 500 kg hydrogen storage tank, a 2000 kW FC, a 1284 kW converter, and 14,600 kWh of battery capacity.”

Source: Journal of Engineering Research

When it comes to flywheels, their limited capacity poses a problem. It forces the design to shift the long-duration balancing to the renewable generation and hydrogen subsystem. So, compared to the lithium option, it requires a 7000 kW electrolyzer and a 7000 kg H₂ tank, a massive increase in the overall system scale by 12x for the hydrogen component.

This is a strong contrast to vanadium redox flow batteries. The ideal VRFB configuration combined a substantially larger energy-storage capacity (35,271 vs. 14,600 kWh) with a smaller fuel cell (FC) capacity (500 vs. 2000 kW).

This change is driven by the ability of redox flow batteries to hold power for a longer time than lithium-ion batteries, therefore taking over part of the role of hydrogen and reducing the need for massive electrolyzers and fuel cells.

As a result, VRFB systems have the lowest Levelized Cost of Energy (LCOE), 0.1427 $/kWh, compared to Li-ion (0.2106 $/kWh) and flywheel recording (0.3721 $/kWh).

This low overall cost was thanks to the VRFB configuration showing superior metrics in all aspects of the operation. It had the lowest initial capital cost, replacement cost, O&M cost, and annual operating cost.

Source: Journal of Engineering Research

What Does It Cost?

For the optimal VRFB-based system, the total Net Present Cost (NPC)

of the system was $29,135,914, of which the largest component was the wind turbines at $13,460,283, followed by the vanadium redox flow batteries ($7,396,824) and the solar panels ($5,613,025).

In this design, wind turbines accounted for 80.4% of total electricity generation and served as the primary energy source, while PV contributed 18.8% and played a complementary role.

The electricity generation of the hydrogen fuel cell was less than 1% and mostly acted as a backup.

“Despite the high dependence on WT, system resilience was supported by the seasonal complementarity between PV and WT and by the combined use of the ESS and hydrogen subsystem, which helped compensate for short-term fluctuations and prolonged renewable generation deficits”

Investors Takeaways

For a long time, the economies of scale in lithium-ion battery production have meant that this technology was the go-to option for rapidly deploying energy storage systems.

However, lithium-ion is fundamentally a technology designed for high energy density in a limited size and weight. Volume and weight hold no consideration for fixed energy storage, be it for grid stabilization or isolated microgrids.

Instead, holding charge for long durations, lower capital costs, and higher lifespan are all going to impact the total LCOE and economics, giving alternative battery chemistries a fighting chance.

In addition, these alternative battery designs are yet to be mass-produced at scale. This means that further reductions in LCOE are likely as they start being produced in greater volume.

So for applications other than homes and EVs, new chemistries like vanadium redox flow batteries could be the next technology to expand, with a trend of matching storage to the application, rather than assuming one battery chemistry will dominate.

This could make for a profitable investment if picking the right company with the right technology, especially as AI data center build-up is desperately needing better options for securing energy supplies.

Still, investors should be aware that there are many alternative chemistries to lithium-ion, and not just vanadium redox flow, as discussed in our report on this topic. And that the calculation in the study worked optimally with the renewable energy supply coming mostly from wind power. A solar-first mix might generate different results and favors other types of energy storage systems.

Investing In Flow Batteries

Invinity Energy Systems

Formed in 2020 by the merger of redT Energy and Avalon Battery, Invinity Energy Systems is a leading global manufacturer of utility-grade vanadium flow batteries (VFBs).

Thanks to their more sturdy design (30+ year operational lifespan), the company’s VFBs deliver 3.8x the lifetime energy throughput of lithium-ion batteries.

It also matches the asset life of solar and wind installations, making it a natural combination for renewable energy production.

Source: Invinity

This durability stems from the way the battery works, with a fully reversible liquid reaction. This delivers a system that has none of the failure points of lithium-ion batteries: no plating, dendrites, or phase change. In addition, safety is greatly increased as there is zero risk of thermal runaway, no chlorine off-gas, and no risk of thermal events caused by short circuits.

In total, the company has deployed 9 GW of battery capacity globally over 90+ projects, including Canada’s largest flow battery setup, the 8.4 MWh Chappice Lake Solar & Storage Project in Alberta, the 20.7 MWh VS3 vanadium flow battery system at the Copwood VFB Energy Hub in East Sussex, United Kingdom, and the upcoming 43 MWh sale of their Endurium vanadium flow batteries to Dairyland Power Cooperative.

In the future, the largest project so far for the company is the FlexBase Laufenburg project in Switzerland, with 1.5 GWh battery capacity, intended to expand up to 2.1 GWh in later phases.

Investors should, however, be aware that this is still a very young startup, in need of additional funding as it scales up.

This had notably resulted in a steep 38% drop in stock price at the announcement of £25 million fundraising designed to strengthen its working capital and accelerate product development ( £10 million from an institutional placing, £12.2 million from the National Wealth Fund and £2.8 million from a retail offer). The crash was because the new shares were priced at 10p, a 43% discount to the traded price at the time.

To improve profitability and reduce future need for capital raises, the company is aiming for a 66% reduction in cost by 2030, with most of these gains already materializing by the end of 2027.

Source: Invinity

So investors will want to properly assess the company valuation and judge if the October 2026 price crash is an opportunity to buy at a discount or a serious red flag.

(You can also read our article “Top 10 Battery Stocks To Invest In” for alternative investment options in this field)

Latest Invinity Energy Systems  (IES) Stock News and Developments

Study Referenced

^{1. Jimin Hong and Yosoon Choi. Comparative techno-economic assessment of energy storage technologies in an off-grid renewable microgrid for AI data centers. Journal of Engineering Research. 19 October 2026. 10.1016/j.jer.2026.10.038}