There is an assumption embedded in most commercial energy investment analyses that deserves more scrutiny than it typically receives. It goes something like this: the network is infrastructure. You connect to it, you pay for it through your tariff, and it delivers or absorbs power as required. Your job is to optimise what sits behind the meter.
That assumption was reasonable for most of the past decade. It is becoming progressively less reasonable, and in parts of the National Electricity Market it has already broken down entirely.
Understanding why matters practically for any organisation making onsite energy decisions in 2025 and 2026. And understanding how artificial intelligence is now governing real-time network operation, not only informing long-term planning, opens a second question about where the energy transition actually gets stuck.
____________________________________________________________
THE GRID YOUR SOLAR SYSTEM CONNECTS TO IS NOT THE ONE IT WAS DESIGNED FOR
When my PhD research at the University of Melbourne began in 2017, rooftop solar contributed a modest fraction of total electricity supply across the National Electricity Market. Reverse power flows, electricity travelling back up the network from consumer to substation, were already an emerging concern, but still confined to specific locations and periods. The distribution network, built and configured over decades to carry power in one direction, was coping.
What has happened since is not incremental. In South Australia, rooftop solar has supplied up to 80% of total electricity demand during certain daytime periods. Australia now has over 25 gigawatts of rooftop solar installed across more than four million sites. The distribution network was not designed for this. Its conductors, transformers, protection devices, and voltage management systems were sized and calibrated for a world in which electricity flowed one way and the behaviour of the network was broadly predictable.
The consequences are real and localised. When solar generation at the feeder level exceeds local consumption, voltages rise, particularly at feeder extremities. Protection devices calibrated for unidirectional current behave unexpectedly under reverse flows. Thermal limits on infrastructure that was sized for load-serving become binding constraints on export. Distribution Network Service Providers are now responding with dynamic operating envelopes, constraining export at the inverter level in real time based on actual network conditions at each connection point.
For a commercial organisation evaluating onsite solar today, this is not a background condition. It is a direct input to the investment case. A system sized to export regularly generates returns only to the extent that its connection point allows export. Many commercial sites are discovering post-installation that the export assumptions in their financial model did not reflect the actual conditions at their specific feeder. The shortfall is not recoverable by adjusting the technology or the tariff. It was baked in at the design stage.
The localisation problem makes this harder than it looks. Two sites in the same suburb, connected to different feeders, can face materially different export conditions. Where a site sits on its feeder, the existing concentration of solar on that feeder, the infrastructure that was installed when the street was developed: these are feeder-level facts that postcode data cannot reveal. This was a central finding of the research I conducted for my PhD: that the aggregated impact of DER penetration is highly localised and node-specific, not uniform across a network, and that assessing it accurately requires analytical tools that work at that level of granularity.
The practical implication for organisations making investment decisions is straightforward. Engaging with the network operator early, understanding the connection conditions at the specific site, and designing systems around those conditions rather than assuming unconstrained access consistently produces better outcomes. It is not a regulatory formality. It is an investment prerequisite.
____________________________________________________________
AI IN NETWORK MANAGEMENT: FROM PREDICTIVE PLANNING TO REAL-TIME OPERATION
The second angle is less obvious but arguably more consequential for the pace of the energy transition overall.
When I developed the machine learning frameworks in my PhD research, the ambition was predictive planning: giving network operators tools to assess, before problems materialise, how voltage and reverse power flows would behave across thousands of DER penetration scenarios. The idea was to move from reactive to anticipatory, from conservative blanket limits to evidence-based risk prioritisation.
That work was research. What has happened in the years since is that AI and algorithm-based tools have moved into live network operation.
The clearest Australian example is Project EDGE, an ARENA-funded programme led by AEMO and AusNet Services, with the University of Melbourne as research partner. Project EDGE demonstrated real-time dynamic operating envelopes at meter level across a trial network of over 320 premises in Victoria. The system calculates time-varying import and export limits at every connection point, updated every 5 to 30 minutes, based on live network conditions. When the network approaches its hosting capacity on a given feeder, those limits tighten automatically. When capacity is available, they open. The inverter at each customer premise responds in real time to the signal it receives.
This is AI governing network operation, not just informing it. The export limit your commercial solar system receives is no longer a static number agreed at connection. It is a dynamic output of an algorithmic system that is continuously assessing network state and allocating available capacity across all connected DER on that feeder.
The implications run in two directions. For commercial operators, it means the performance of an onsite energy system is increasingly a function of its position in a real-time network management system, not just its technical specification. A battery that can respond intelligently to dynamic envelope signals extracts more value than one that cannot. A solar system connected to a feeder with sophisticated operating envelope management has a different risk profile than one on a feeder still operating on static limits.
For the energy transition more broadly, it reveals the full scope of what AI now needs to do in the grid. The planning challenge and the operational challenge are distinct but connected:
• Predictive planning tools: assessing the probability and severity of voltage excursions and reverse power flows across thousands of future DER penetration scenarios, identifying which feeders face the highest risk, and directing augmentation investment before problems emerge. This is what my PhD research addressed.
• Real-time operational governance: calculating and communicating time-varying connection limits at every node, optimising the allocation of available hosting capacity across all DER assets on a feeder, and making those decisions at the speed and granularity that a network with millions of distributed assets requires. This is what Project EDGE demonstrated.
Neither is sufficient without the other. Real-time envelope management without good predictive planning produces a system that reacts intelligently to current conditions but invests poorly in future capacity. Predictive planning without operational intelligence produces well-targeted augmentation decisions that are then undermined by connection constraints that were not anticipated at the operating level.
The bottleneck in the energy transition is not only technology cost or consumer willingness. It is increasingly the pace at which the network can intelligently accommodate the DER that consumers and businesses are ready to deploy. Connection delays and restrictive export limits are among the most significant friction points for commercial DER deployment across the NEM right now. Better planning tools and smarter operational systems both directly affect the economics and timelines of commercial solar, battery, and fleet electrification programmes.
The tools needed to manage this transition are no longer theoretical. They are in deployment. The question now is how quickly the planning and operational layers become fully integrated, and how commercial organisations position their investments in anticipation of a network that is being actively reshaped by the intelligence governing it.
____________________________________________________________
GG ADVISORY PERSPECTIVE
These two observations connect in a practical way. The network environment that commercial energy investments operate in is not a given. It is feeder-specific, rapidly evolving, and increasingly governed by AI systems operating at timescales and granularities that were not part of the commercial energy conversation even five years ago. Organisations that understand this context make better investment decisions. Those that treat network access as a standard assumption often find the investment underperforms expectations in ways that could have been anticipated.
GG Advisory brings this analytical perspective to the strategy and business case work we support for commercial organisations evaluating onsite solar, batteries, and fleet electrification. Our understanding of how distribution networks respond to DER penetration is informed by direct research experience in machine learning based network planning, combined with practical energy transition strategy grounded in project work with Shell, ARENA, and Powerlink. If your organisation is working through an onsite energy strategy and wants a perspective that includes the network context, we welcome a conversation.
