GreenPoint's 800MWh Grid-Forming Battery: How AI Data Centers Became the Ultimate Energy Storage Offtaker — Analysis
Clean energy developer GreenPoint Energy (formerly Equis Development) has achieved financial close on the Koolunga grid-forming battery energy storage system in South Australia — a 200MW/800MWh installation that marks a pivotal moment in the convergence of energy storage and artificial intelligence infrastructure. The project, backed by Australia's federal Capacity Investment Scheme and underpinned by a long-term offtake agreement with global energy trader Gunvor Group, will supply clean, dispatchable power to AI data centers operated by Firmus Technologies at Tailem Bend and Stirling North in South Australia. This tripartite structure — developer, commodity trader, and data center operator — represents what both parties call a "replicable commercial framework" for powering the AI revolution with 100% renewable energy. But the true innovation lies deeper: this is the first major project where a grid-forming BESS is explicitly designed to serve a hyperscale computing load that demands power quality standards approaching those of semiconductor fabrication facilities.
Overview of the Koolunga Grid-Forming BESS and CIS Framework
The Koolunga project, located near Brinkworth in South Australia's mid-north region, is a 200MW/800MWh battery storage facility with a 4-hour duration — the configuration increasingly recognized as the minimum viable specification for providing genuine system security services while capturing energy arbitrage value. The 200MW grid-forming capability means the BESS can establish voltage and frequency reference without requiring a synchronous generator to be online — a critical attribute for South Australia's grid, which already operates at approximately 75% renewable penetration and is projected to reach 100% net renewable generation by 2027.
The project received a contract under the Australian federal government's Capacity Investment Scheme, a policy mechanism designed to underwrite new dispatchable renewable capacity through revenue underwriting agreements that reduce financing risk for private developers. The CIS effectively guarantees a minimum revenue floor, enabling GreenPoint to secure project finance at investment-grade terms for an asset class — large-scale BESS — that lenders still view as having unproven long-term revenue profiles. Construction is expected to begin in 2026 with commissioning targeted for early 2028, positioning Koolunga as one of the first wave of CIS-supported BESS projects to reach commercial operation. Explore AGAIC POWER's grid-scale battery energy storage systems for utility and data center applications.
Why AI Data Centers Are the Perfect BESS Offtaker
The offtake structure between Gunvor and Firmus Technologies is elegantly designed. Gunvor, a global energy and commodity trading company with over US$160 billion in annual revenue, acts as the intermediary: it purchases the BESS output from GreenPoint under a long-term contract and sells electricity to Firmus under a separate power purchase agreement. This structure allows Firmus — a technology infrastructure company, not an energy trader — to focus on its core business while Gunvor absorbs the market risk, dispatch optimization, and regulatory compliance burden.
The commercial terms include a notable flexibility provision: Firmus has agreed to reduce its AI factory load by up to 220 hours per year when wholesale electricity prices exceed a contractually defined threshold. This is not a hardship provision — it is a commercially rational arrangement that reflects the operational reality of AI training workloads. Unlike traditional data centers running latency-sensitive cloud services, AI training clusters can often be paused or slowed during periods of extreme electricity prices without degrading the end product. A large language model training run that takes 45 days at full power may take 46 days if paused for 24 hours of high-price intervals — a 2.2% timeline extension that is insignificantly small compared to the millions of dollars saved by avoiding peak electricity costs. This load flexibility is the operational attribute that makes AI data centers fundamentally different from traditional data center loads — and fundamentally more compatible with variable renewable generation and battery storage. Gunvor has separately committed to developing 1.2GW of new renewable generation and 1.5GWh of additional energy storage by 2032, with Firmus as the anchor offtaker — suggesting that the Koolunga project is the first of many such configurations.
Technical Deep Dive: Grid-Forming Inverters and AI-Grade Power Quality
The grid-forming capability specified for Koolunga represents a significant technical escalation from conventional grid-following BESS installations. A grid-following inverter synchronizes to an existing grid voltage waveform and injects current in phase with that waveform — it can provide power but cannot establish the voltage reference itself. A grid-forming inverter operates as a voltage source rather than a current source: it actively synthesizes the grid voltage waveform, maintaining stable frequency and voltage even when no synchronous generators are online. This capability is essential for a BESS that must operate in South Australia — a grid that occasionally achieves periods of 100% inverter-based resource operation, where the last remaining synchronous generators may be offline.
At the engineering level, a grid-forming inverter requires a fundamentally different control architecture. Instead of a phase-locked loop that tracks grid voltage (the grid-following approach), a grid-forming inverter uses a virtual synchronous machine algorithm that emulates the electromechanical behavior of a physical rotating generator. The control system continuously solves the swing equation — the differential equation describing rotor angular dynamics — in software, producing an output waveform that exhibits inertia-like resistance to frequency changes. When a sudden load increase or generator trip occurs, the grid-forming inverter can inject additional current within 2-5 milliseconds to arrest frequency decline — approximately 100 times faster than a synchronous generator's governor response, which typically requires 500-2000 milliseconds to begin increasing mechanical power output.
For Firmus Technologies' AI data centers, this power quality attribute is not just a grid service — it is a direct operational requirement. AI training clusters draw power with a load factor exceeding 95% and exhibit almost no reactive power consumption, making them unusually "stiff" loads from a grid perspective. However, their computing hardware — particularly GPU clusters drawing multiple megawatts — is sensitive to voltage harmonics and frequency deviations that can cause computation errors or hardware damage. The grid-forming BESS at Koolunga effectively provides a power conditioning function between the transmission grid and the AI data center, absorbing voltage transients and frequency excursions that would otherwise propagate directly to Firmus' servers. Visit our store for commercial and industrial energy storage solutions with advanced grid-forming capabilities.
Real-World Applications: The "Replicable Framework" for AI Infrastructure
GreenPoint and Gunvor's characterization of the Koolunga structure as a "replicable commercial framework" is not marketing language — it reflects a genuine structural innovation that addresses the core bottleneck in AI infrastructure deployment: power availability. A hyperscale AI data center campus requires 500MW to 1,000MW of continuous electricity, but grid interconnection queues in most developed markets are 3-7 years long and transmission upgrade costs can exceed US$400 per kilowatt of new load. By pairing AI data centers with purpose-built BESS and renewable generation under a single offtake structure, developers can offer grid operators a package that includes not just new load but also new dispatchable capacity and grid services — transforming the interconnection application from a burden on the grid into an enhancement.
The Tailem Bend and Stirling North sites chosen by Firmus exemplify this strategy. Both are located in South Australia — a state with abundant renewable generation, some of the world's highest wholesale electricity price volatility, and a transmission network that is actively seeking new dispatchable resources to maintain system security as synchronous coal and gas generation retires. By siting AI data centers at locations where the grid operator needs the services that the co-located BESS can provide, GreenPoint and Firmus have effectively made the electricity system a stakeholder in their project's success — a fundamentally more powerful negotiating position than a standalone data center interconnection request.
Industry Impact: The Convergence of Energy Storage and AI Infrastructure Markets
The Koolunga transaction signals the emergence of a new asset class at the intersection of energy storage and digital infrastructure. Traditional BESS offtake agreements — with utilities, grid operators, or energy retailers — value the battery based on its projected market revenue from energy arbitrage and ancillary services. An AI data center offtake adds a third revenue dimension: the value of dispatchable, high-quality power to a customer whose business depends on uninterrupted electricity supply. This expansion of the revenue base fundamentally improves BESS project economics, potentially enabling projects that would not achieve financial close under a market-revenue-only structure.
The scale is significant. Global AI data center electricity demand is projected to grow from approximately 30TWh in 2024 to 200-350TWh by 2030 — equivalent to 3-5% of total global electricity consumption. If even 25% of this demand were served through the Koolunga-style structure (BESS + renewables PPA), it would require approximately 200-350GWh of operational battery storage — roughly equal to the current global utility-scale BESS pipeline. This convergence of AI infrastructure investment (projected to exceed US$1 trillion cumulatively by 2030) with energy storage deployment creates an investment flywheel where each sector's growth accelerates the other's.
Future Outlook: From Single Project to Standard Model
The Koolunga project's success will be measured not just in megawatts and megawatt-hours but in whether it catalyzes replication. Three conditions make replication likely. First, the AI industry's power demand growth rate — estimated at 25-35% annually — means that data center developers are actively searching for locations where power can be secured within commercially acceptable timelines, not just where it is cheapest. Second, grid operators globally are recognizing that AI data centers, if properly integrated with BESS and load flexibility, can provide demand-side services that enhance rather than strain grid reliability. Third, the financial community is developing increasing comfort with BESS revenue models, and the addition of a creditworthy offtaker like Firmus (backed by Gunvor's balance sheet) materially reduces perceived investment risk.
The next wave of convergence will likely involve direct co-location of data centers with renewable generation and storage at the same physical site — eliminating transmission constraints entirely and enabling truly 24/7 carbon-free energy matching at an hourly granularity. This configuration, sometimes called an "energy park," is already being explored by technology companies in Texas (ERCOT), the Nordics, and Western Australia. The Koolunga project, with its tripartite offtake structure and grid-forming technical specification, provides the commercial and engineering template that future energy parks will build upon.
