Beyond BESS: Local Energy Domains, Not Bigger Storage

Act I — The pain

Energy is becoming a constraint on what a territory can do

Energy is usually discussed as supply: there is enough capacity, or there is not. Increasingly it behaves as something else — a constraint on what a territory is able to do at all.

The constraint shows up on three levels at once.

First, electrification raises local peaks. Electric vehicles, heat pumps, electric heating, local commercial load. Peaks grow not on a national average but at specific points on specific feeders.

Second, critical load requires resilience when individual elements fail. Telecommunications, water, medical facilities, cold storage, emergency services. For such loads the failure of a single supply element can mean the loss of the critical function, which makes redundancy and restoration an engineering requirement in its own right.

Third, and this one is new in scale, dense new loads arrive faster than the network can be reinforced. Charging hubs, edge and computing facilities, local manufacturing. Connection timelines in congested markets can run into years [2], and major network build-out can take many years [1].

In many cases the chain looks like this: load grows, connection is constrained, reinforcement is expensive and slow, backup becomes necessary, storage is added as a buffer — and the topology of dependency stays largely as it was.

The village that wants a data centre

A village is designed around today’s peak. Houses, a few commercial sites, water supply.

Five years later its energy problem looks different: hundreds of electric vehicles, electric heating, small manufacturing. Then a computing facility appears nearby, or the village wants to host one.

In the conventional architecture this becomes, first of all, a request for network capacity: more kilowatts at the connection point, possibly a new transformer, feeder reinforcement [4], a place in the connection queue [3].

The question worth asking instead is a different one.

A computing facility here is not simply one more consumer. It is a concentrated, growing and critical load, for which predictable local power and redundancy are particularly valuable [5]. It works as a load-side stress test for the whole energy architecture of a territory.

From there a reverse logic appears, and for a municipality or an investor it matters more than the defensive one. Not “a data centre needs energy, so let us build a source next to it”, but rather: the territory already holds distributed energy infrastructure — what new kinds of economic activity is it now able to accommodate?

Two values, and the second is usually underrated. Resilience is what a territory can survive. The ability to accommodate new load is what it can afford to connect without proportional reinforcement of the upstream infrastructure.

The same logic applies to charging hubs, heat pumps, water supply and treatment, cold storage, small manufacturing, telecommunications, medical and emergency facilities.

Act II — What is already known

What physically happens when one house adds load

Take a street: grid → transformer → feeder → houses A, B, C, D.

House C switches on an electric sauna, puts a car on charge and starts a heat pump. In our example — roughly another twenty kilowatts at a single point.

What happened physically? Not that a power station “sent extra electrons to house C”. The electromagnetic state of the coupled network changed: currents and power flows redistributed across the shared topology, and the increased demand is met through the existing electrical paths.

So the additional load at house C changes the distribution of flows across the shared electrical topology. How far that change appears at the feeder, the transformer and the upstream boundary depends on the state of the other loads, generation and storage at the same moment. There is no addressed channel “from someone to C”; there is a change in the state of a shared coupled system.

Why local generation is not addressed help to a neighbour

The reverse example. House A generates more than it consumes.

Intuition suggests the surplus “helps neighbour B”. Physically it does not. The surplus changes flows in the shared electrical network, and the consequences are well known and quite specific: reverse power flow, local voltage rise, hosting-capacity limits, and the need to curtail generation [4] [6].

Increasing local generation without a corresponding ability of the network to accept it or to use it on site may therefore not remove the constraint but change its form: instead of a capacity shortfall there appear reverse flow, voltage rise, hosting-capacity limits and curtailment. And that is why the question “where to put the surplus kilowatt-hour” only makes sense inside the existing topology of dependency.

What existing approaches already do

An honest picture: none of the instruments below is disputed here. What matters is which limitation each of them confirms.

Storage. It answers, above all, the question of when previously stored energy should be spent. In a suitable architecture an energy storage system can take part in voltage support, peak shaving, reverse-flow management and microgrid operation [6] [11]. The limitation it confirms: a time buffer solves the problem of moving energy in time, not the problem of distributing responsibility among several local nodes.

Microgrids, distributed energy resource management, energy management systems, virtual power plants. This has to be said plainly, because it is decisive for the position of this article: the idea of coordinating local sources, storage and load is not new, is widely researched, and is implemented in existing microgrid and distributed-resource architectures [7] [8] [10]. A microgrid is defined normatively as a group of interconnected loads and distributed energy resources with clearly defined electrical boundaries, acting as a single controllable entity and able to operate in grid-connected or island mode [9]. Modern microgrid controllers handle dispatch, resource coordination, power flows, reserve, economic optimisation and peak management, including while working with the grid [7] [9].

From which follows something important for us: that literature and practice are not an opponent of the hypothesis stated here but its scientific foundation. This article does not claim that nobody coordinated local resources before it, and such a position would be indefensible.

Hosting capacity, reliability and restoration indicators. The apparatus is developed, the metrics exist and are applied [11] [12]. It describes how much the existing topology withstands and how an area behaves during events.

Act III — The gap

The question that stays open

Existing solutions can store, generate, aggregate and island. One architectural question stays open.

And then the experimental question for which this article exists: can VENDOR.Max, after independent validation of a single node, become a physical element of such a domain?

Note what is absent from both formulations. There is no claim that coordination of distributed resources is impossible without us: it is possible and it is implemented. There is no word about replacing the electricity network. There is no promise to survive any failure. There is no claim that network oscillations are eliminated.

The question is narrowed to a class of node, and to whether resource distribution can become a base property of the energy architecture of a territory built from nodes of that class.

Act IV — The architectural hypothesis

Localising part of the balancing responsibility

In the baseline network architecture, local loads and resources interact through shared electrical infrastructure and its upstream boundary.

The architecture under study adds a coordinated local domain inside that boundary: grid ↔ local domain ↔ nodes ↔ loads. The domain gains its own ability to distribute generation, load, reserve and the consequences of failure.

Return to the evening example. 18:00 — the domain is balanced relative to some state. 18:01 — house C needs additional power of the same order.

The conventional structure. The additional demand changes the distribution of flows across the shared electrical topology; how far that change appears at the upstream boundary is determined by the state of the other loads and local resources.

The structure under study. The control layer first determines what resource exists inside the domain — available power of nodes A, B and D, storage, other local sources. The design objective of the architecture under study is to use the available internal resource so that what appears at the upstream boundary is a residual exchange after the local response of the domain. This is a design objective, not a derived property: the specific dispatch hierarchy is not yet defined.

A terminological note. “Responsibility” here does not mean balancing responsibility in the electricity-market sense of a market party. It means the engineering ability of local resources to change their operating mode in response to a change of demand inside the declared domain.

A physical note, without which the formulation would be wrong. While the local domain remains electrically coupled to the shared network, balancing is not literally moved: the coupled system continues to satisfy a single electromagnetic state. What is moved is part of the responsibility for covering a change in load, and what changes is the magnitude of the residual exchange across the boundary. Declaring a separate physical balancing where the electrical topology of coupling is not yet defined would be incorrect.

A necessary engineering note. The phrase “neighbouring nodes take part of the load” describes an assumed redistribution of available resource, not an ability that already exists. The specific electrical topology of coupling between nodes, the interconnection scheme and the routing of power belong to a separate engineering task of later implementation and testing, and are not defined in this article.

The fabric: nodes, control layer, reserve, degradation

The cellular-network analogy is useful here but strictly architectural: the physics of a radio network and of an electrical network are different, and the two must not be equated. The shared principle is another one — coverage is created by a topology of cooperating nodes, not by the perfection of one central node.

In the architecture under study a node must report:

• its own state;
• available power and reserve;
• current load;
• the ability to take additional load;
• any fault or active limitation;
• the state of its communication channel.

The power of a node is set by its hardware. Participation in a coordinated fabric additionally requires a control layer and the corresponding interfaces.

Assumed behaviour on node failure. Node C reports a fault or stops responding. The control layer recomputes the state of the area and the available reserve. The remaining nodes change mode within the resource available to them. The design logic assumes priority for preserving the declared critical load. Any residual shortfall may be presented to the upstream network within the chosen coupling architecture.

The objective of such an architecture is to change the form of failure from a binary interruption into a managed degradation. How complete that degradation and the preservation of critical load turn out to be depends on the reserve of the remaining nodes and on which share of load is declared critical. The architecture does not promise that the domain survives any failure, and until the corresponding tests are passed, degradation remains a design objective and not an established property.

Storage does not disappear from this picture: it may well be one of the nodes of the fabric.

Act V — The test vehicle

VENDOR.Max: what exists, what is hypothesis, what is not shown

Up to this point the article has spoken about an architectural class. VENDOR.Max appears here not as a proven solution to the whole task, but as a physical candidate for the role of node, with which the architectural hypothesis can be moved into experiment.

What exists at development level. A single node is designed to form its own regulated direct-current boundary condition for the connected load. The architecture provides for local control functions. The physical node exists as a development-stage device.

First hypothesis. Several such nodes can be combined into a local energy domain in which their available power, states and reserves are coordinated.

Second hypothesis, further out. On a change of load or on the loss of one node, the remaining available resources of the domain will be able to change mode so that part of the load is redistributed without loss of the whole domain.

Third, further still. The measured actions of nodes can be turned into a metrologically unambiguous accounting layer.

What is not shown and is not claimed here.

• There is no deployed multi-node installation with the coordination described above.
• Synchronous parallel operation with the grid is not claimed: standalone output and parallel operation are different product classes requiring separate testing and certification.
• The electrical scheme of coupling between nodes is not defined.
• The operating-state dependency of the product has not been measured.

State of work. The installation worked, results were obtained and recorded. Repeat engineering validation of the current implementation belongs to the next stage; historical internal results do not replace it.

Patent protection: WO2024209235A1 (Published), ES2950176B2 (Granted). Project stage: TRL 4 — Prototype Rebuild After Relocation.

Act VI — What we test

The test programme

A hypothesis without a programme of verification is a promise. Below is a sequence in which each step is meaningful only after the previous one. Numerical pass criteria are deliberately not set here: they belong to the protocol, not to the article.

Separating H3 from H4 matters: the ability to exchange state and issue a coordinated assignment belongs to the control layer, whereas a reproducible change of flows belongs to the power layer. Merging the two questions into one test makes it impossible to establish which layer failed.

The chain reads: know → decide → act → survive → measure. Only after H8 is there a basis for unambiguous accounting of node interaction.

A metrological guard for H6 and H7

A short but obligatory discipline is needed here, otherwise the results of H6 and H7 will be described in words that promise more than was measured.

The word “dependency” conflates several distinct physical properties. The apparatus for them exists separately: the energy side is described by established self-sufficiency and autarky indicators [13] [14], the reliability and restoration side by its own indicators [12]. The difficulty is subtler: these properties are routinely called by one word and therefore substituted for one another.

For that reason this work will not call the system “independent from the grid”. Three things will be checked separately: energy import across the declared boundary; operating-state dependency on the external system while the coupling is preserved; preservation of function on loss of the declared boundary.

None of the three answers follows from the other two. The simplest counter-example is a good mains power supply: a disturbance at its input barely reaches the output, yet all the energy crosses the boundary, and on loss of that boundary the function is not preserved. Good line regulation is a regulation characteristic, not independence of operating state.

Act VII — What becomes possible

Infrastructure consequences

The consequences carry different evidential weight and must not be mixed.

Necessary — following from physics:

Why there are two separate rows for energy and for power. Energy and power require different conditions. Energy needs a declared interval; power needs a state, an instant or a profile. An increase in local coverage of energy over a day does not guarantee a reduction of peak power import in the evening hour. For this article the distinction is decisive: the ability of a territory to accommodate new load is determined first of all by a power and profile constraint, not by annual energy self-sufficiency.

Why it is phrased this way. Redistribution of load among local nodes does not by itself guarantee a reduction of flow across the boundary. If four nodes redistribute load while the total resource available inside the domain is unchanged, the import across the boundary may stay as it was. A reduction requires an additional physical fact: the domain must increase local coverage of demand — through generation, previously stored energy resource, controlled change of load or other resources located inside the boundary. Power conversion is not itself a covering resource: a converter changes form, voltage or mode, but does not create an energy resource; it only provides access to one. The first row is worded so that this condition is built into it by the definition of the boundary.

Conditional — subject to the stated condition:

System-specific:

The economic side is stated as a change of exposure to classes of cost, not as savings. What changes is not the bill but which cost classes the territory is exposed to at all: transfer capacity, connection capacity, congestion, network reinforcement, supply interruption, curtailment, export tariffs. Monetary figures are not given before an engineering model of a specific site exists.

Where the market is moving and why it matters

Two movements run in parallel and are almost never connected to each other.

Capital is moving behind the meter. According to Currence — formerly Sightline Climate — venture investment in climate technologies reached 26.1 billion dollars in the first half of 2026, up 55 per cent year on year, while low-carbon data-centre technologies made up 34 per cent of the total against 3 per cent a year earlier [17]. Goldman Sachs estimates that behind-the-meter systems will provide a quarter to a third of the incremental electricity demand from data centres anticipated through 2030 [18]. According to Rabobank, gas accounts for more than 80 per cent of the announced behind-the-meter capacity, and the selection is made not by fuel but by time to power [2].

One of the determining factors here becomes time to power — how long it takes a new load to actually gain access to supply. Alongside it, other properties are bought: independence from the connection queue, capacity adequacy, reliability.

The power architecture of computing is moving to direct current. NVIDIA, Google and Microsoft are developing an 800 V DC architecture through the Open Compute Project: a joint technical paper was published in March 2026, the low-voltage solid-state transformer specification version 0.3 in July 2026, and infrastructure compatible with 800 V DC is being developed by more than eighty equipment manufacturers and infrastructure companies [19] [20]. The point of that movement is to remove intermediate conversions and bring the regulated direct-current boundary closer to the load [19].

Telecommunications infrastructure has run on minus 48 volts direct current for decades [21] [22]. These are different voltage classes and must not be equated, but the architectural vector is the same.

Data centres are used here not as the single target market but as the most visible indicator of direction: high load density makes the constraints of time to power and of delivery architecture particularly apparent.

Accounting

When the physical architecture makes it possible to measure who generated how much, who consumed how much, who provided reserve, who took additional load and how much energy crossed the boundary of the domain, a basis appears for unambiguous accounting of node interaction.

The order matters: physics → node → network → control → validation → accounting. The reverse order produces a construction with nothing to count.

This article does not define the commercial, contractual or settlement model of such interaction.

What comes next

The industry asks how much storage to add to the grid. The question is reasonable, but it is the second one.

The first is where the responsibility for covering a local change in load should sit, so that growth of demand at one point does not automatically become a request to the upstream network, and so that the failure of one element does not mean the failure of the area.

The answer cannot be reduced to kilowatt-hours of installed storage. It has to be checked through the behaviour of the whole domain: exchange across the boundary, available reserve, response to a change of load, failure of individual nodes, and preservation of critical function.

Questions