Energy / Grid Management (AI) · Deep dive

Gridsight

Six-year-old Wollongong-founded AI capacity-management platform whose digital twin of a distribution network — fed by smart-meter data across 700,000 monitoring points on Endeavour Energy alone — let that DNSP double household solar export limits from 5 kW to 10 kW for 95% of the year, unlocked ~600 MW of additional rooftop solar and roughly A$100M of customer value, and now, after a US$26M Series B on September 1 2026 led by Insight Partners with Galvanize participating, is trying to sell the same math to Xcel Energy, Avangrid's United Illuminating and the rest of a US IOU sector planning ~US$1.3T of grid CapEx through 2030.

emerging

The question that decides it: Does an AI-based DSO capacity layer survive when EPRI's OpenDSS, Siemens' PSS SINCAL and Bentley's already-installed OpenUtilities suite each bundle equivalent DER-aware hosting-capacity analysis into the planning tools US IOU planners already own — and does that answer even matter if the regulated rate-case economics that reward CapEx (poles, wires, substations) over software mean the same utilities keep choosing to build the upgrade rather than route around it?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Gridsight is a six-year-old Australian software company that turns the distribution grid into a live digital twin — a physics-plus-ML model of every low-voltage feeder, transformer and smart meter — and sells the resulting hosting-capacity numbers to the distribution utility as an alternative to building poles and wires. Its Endeavour Energy reference in NSW uses the platform across ~700,000 monitoring points to double household solar export limits from 5 kW to 10 kW for 95% of the year, unlocking about 600 MW of extra rooftop solar and A$100M of customer value with no substation upgrade (Utility Magazine, 2025). On September 1 2026 it closed a US$26M / A$36M Series B led by Insight Partners with Galvanize new and Airtree, Energy Transition Ventures and Aera VC following on (PR Newswire, September 1 2026). Total capital: ~US$33.5M. The round funds a US push — Xcel Energy and Avangrid’s United Illuminating are named — before the incumbents already inside US planning departments bolt equivalent AI onto their own tools.

Founding story

Gridsight is a University of Wollongong spinout dressed as a Sydney AI company. CEO Brendan Banfield holds a PhD from UOW in distribution-network integration of solar and batteries; he worked at UOW’s Sustainable Buildings Research Centre, where the Desert Rose Solar Decathlon project crystallised the founding question — you could put arbitrary solar, storage and EV load behind the meter, but the transformer on the pole outside had no idea it was there.

Banfield incorporated Gridsight in June 2020 with childhood friends Kurt Walkom and Hugh Chan; first paid pilot signed Christmas Eve 2020; full-time in 2021 through iAccelerate. Australia was the right laboratory: by 2024 more than 3.2 million homes had rooftop solar, roughly 41% of premises hosted PV, and rooftop generation had hit ~20 GW of NEM capacity — a saturation no other country had reached (PV Tech, 2025). South Australia now periodically meets 99.9% of demand from rooftop solar (PV Magazine Australia, September 2 2026). Australian DNSPs had to solve DER-driven voltage and thermal constraints five to ten years before US utilities will; Gridsight built the tool while that problem was on fire.

How it works

The platform is an electrical digital twin of a distribution network, kept live with metered data. Four layers: (1) ingest of GIS topology, SCADA telemetry and — critically — smart-meter interval data from the DNSP; (2) a physics engine running unbalanced three-phase load-flow and probabilistic power-flow simulation across the LV network, same math family as OpenDSS but continuously calibrated against measured voltage and current; (3) ML forecasts of rooftop-solar generation, EV charging, battery cycling and heat-pump load by node and by hour; (4) an optimisation layer that computes, feeder by feeder and hour by hour, how much additional load or export the network can carry within voltage and thermal constraints. That number is a Dynamic Operating Envelope (Gridsight Blog, Q3 2023) — the real-time cap on what an inverter or EV charger can safely do at a given time and location.

The Endeavour deployment is the fullest expression: static planning sets a conservative 5 kW export cap fleet-wide; Gridsight’s live model shows the network has 10 kW of headroom 95% of the year, hands that back, and only clamps down when a specific feeder is constrained (Utility Magazine 2025).

Product and business overview

Gridsight sells one platform in three layers: Insight, the digital-twin and hosting-capacity map (where is capacity, when, for how long); Flow, the workflow tool that lets utility engineers approve or reroute DER, EV and data-centre interconnection requests against live capacity; and Flex, the DOE service that pushes real-time export/import limits to inverters, chargers and CER aggregators. Positioned as a DERMS-adjacent capacity layer between GIS/SCADA and any downstream DERMS or ADMS — deliberately complementary, because ripping and replacing an ADMS is a decade-long procurement no buyer will authorise.

Business model and pricing

Pricing unpublished. The shape is a multi-year enterprise SaaS contract per DNSP scaled by network size (monitoring points, connected customers or feeders) — the same shape Bentley and Siemens sell into the same buyer. Endeavour’s ~700,000-monitoring-point deployment implies six- to low-seven-figure ARR per large network; ~50% of Australia’s DNSPs at Series A implied ~4.2 GW of DER under management (Business News Australia, April 2025). The Series B narrative is that the same shape ports to US IOUs, where a single Duke or Xcel service territory is an order of magnitude larger by connected customers than any Australian DNSP.

Traction over time

The 4.2 GW at Series A is an unusually deep footprint for that stage (Startup Daily, April 8 2025).

Market analysis

US electric utilities are forecast to spend ~US$1.3T of CapEx through 2030 (S&P Global, April 2026); Duke plans US$102B, Southern US$81B, AEP US$72B; PG&E announced a US$73B upgrade tied to 10 GW of new data-centre load (DCD, 2026). Interconnection queues have exploded — >80 GW of forecast US data-centre demand by 2030 sits in front of PJM and MISO queues that already take 4-5 years to clear. The Gridsight thesis is that ~75% of grid capacity goes unused on average and finding the latent headroom is cheaper than building new wires (ESG Today, September 1 2026). True as physics. The commercial question is whether US IOUs — remunerated on regulated return-on-CapEx — prefer to build the upgrade (earn on for 30 years) or pay a software vendor to route around it. Australia is the leading indicator only because AER-regulated DNSPs are rewarded for non-network alternatives; FERC-regulated IOUs are not.

Competitive intel

Three layers. The incumbent planning tools already inside utilities are the most dangerous: EPRI’s OpenDSS (free, open-sourced 2008, the workhorse for hosting-capacity studies in every large IOU) and Siemens PSS SINCAL (paid, GIS/SCADA-integrated, powering Bentley OpenUtilities Analysis). Bentley in 2024-2026 rolled out OpenUtilities DER Optioneering — cloud screening on PSS SINCAL — exactly the workflow Gridsight sells. Free-to-cheap, already bought, already integrated, and their roadmaps are absorbing the AI math Gridsight is charging for.

The US-native AI capacity startups: GridCARE (Palo Alto, US$77.5M total after a May 14 2026 Sutter-Hill-led US$64M Series A; runs quadrillions of grid-physics simulations for hyperscale data-centre siting) is the direct competitor for the data-centre wedge Gridsight named in Series B messaging (Bloomberg, May 14 2026). More capital, Nvidia-family investors.

The DERMS/orchestration neighbours: Camus Energy (DERMS/ODMS for co-ops/munis), Utilidata (Providence, ~US$110M raised, Nvidia Karman edge-AI in distribution meters) and Kevala Analytics (SF, cloud DER analytics). Same DER-integration budget from adjacent wedges.

History and evolution

• 2018-2019 — Banfield leads UOW’s Desert Rose Solar Decathlon project.
• June 2020 — Gridsight incorporated in Sydney by Banfield, Walkom and Chan.
• December 24 2020 — First paid pilot signed.
• 2021 — Founders full-time via iAccelerate at UOW.
• 2022-2024 — Multi-DNSP expansion; Endeavour flexible-exports work begins.
• Q3 2023 — Blog publishes ‘Dynamic Operating Envelopes’ framing that anchors the Flex product.
• April 8 2025 — US$4.5M / A$7.5M Series A led by Airtree with Energy Transition Ventures, Aera VC and Pat Cummins.
• 2025 — Endeavour launches AI-driven Flexible Exports; ~600 MW additional solar / ~A$100M value forecast.
• September 1 2026 — US$26M / A$36M Series B led by Insight Partners; Xcel Energy and Avangrid’s United Illuminating named.

What people say

The case for

Trade press has been consistently positive: Utility Magazine, ESD News and One Step Off The Grid covered the Endeavour flexible-exports launch as one of the first at-scale DOE implementations anywhere, not a pilot. Business News Australia and Startup Daily flagged the 4.2 GW of DERs under management at Series A as an unusually deep footprint for that stage. Glassdoor is thin (N=1, 4.0 stars, a senior Wollongong data engineer) but the review is specific — flexible hours, hard technical problems, an A$2,500 wellness benefit.

The complaints

No G2, Capterra or Gartner Peer Insights presence — no negative reviews, but also no third-party validation of the ROI numbers, which all come from Gridsight or Endeavour communications. Two structural complaints. First, the reference-case ROI is a DNSP-avoided-CapEx story in a country where the regulator explicitly rewards non-network solutions; the US IOU rate-case regime is the opposite, and no Duke or Xcel planner has yet convinced a state PUC to accept a Gridsight-optimised capacity number in place of a rate-based transformer upgrade. Second, the US expansion is starting late: GridCARE has US$77.5M and Sutter Hill/Doerr, Bentley/Siemens are already inside the same accounts, and the IOU sales cycle is measured in years.

Outlook: the open question

Gridsight compounds if three conditions hold — and stalls if any breaks. (1) At least one US IOU or PUC accepts a Gridsight-computed hosting-capacity number as a substitute for a planned distribution upgrade in a general rate case by end-2027 (not merely as an interconnection-screening convenience). (2) Neither Bentley OpenUtilities nor an OpenDSS-based EPRI reference workflow ships an equivalent ML-driven DOE / hosting-capacity module bundled inside the existing IOU planning-tool licence within 24 months. (3) GridCARE’s Sutter-Hill-funded push into hyperscale data-centre capacity does not absorb the standalone data-centre use case before Gridsight converts its first US IOUs to expansion contracts. The physics is right. But this is a regulated-utility software sale, not SaaS, and the incumbents already inside the buyer are the ones that decide. If Insight’s operating partners shorten the Duke/Xcel/Southern/AEP/Dominion sales cycle from the multi-year default to 12-18 months, this works. If not, Gridsight becomes a very good Australian company having a difficult US year.

How to attack it

The wedge: sell capacity-as-a-service to US electric co-ops and municipal utilities on a flat-fee-per-MW-unlocked basis, and skip the IOU sales cycle entirely. There are ~800 rural electric co-ops and ~2,000 municipal utilities in the US, most serving 5,000-200,000 meters, all currently unable to onboard the data-centre or crypto load knocking on their door because they lack the planning bench of a Duke. They buy software on utility-scale procurement, not rate cases, so a per-MW-unlocked contract is legible to their boards in a way it is not to a state PUC. Camus Energy already proved a co-op/muni motion works with Holy Cross Energy; nobody is doing it with a capacity-first product.

Weaknesses an attacker can exploit: (i) no US-native product yet — Gridsight still sells an Australian reference; (ii) no PUC precedent for the capacity numbers, so every IOU sale requires custom regulatory work an attacker can pre-empt with a formal PUC filing programme in Colorado, Texas, California and New York; (iii) heavy dependency on smart-meter interval data — MISO/PJM co-ops with AMI 1.0 rollouts don’t have Endeavour’s resolution, so a smart-meter-agnostic estimator (probabilistic power flow tuned to feeder-head SCADA only) is a genuine differentiator; (iv) no data-centre-siting narrative yet — GridCARE already owns ‘find idle capacity for a hyperscaler’ with more capital; (v) Australian innovation velocity does not survive inside a US IOU where staff turnover, procurement gates and integrator dependencies (Accenture, Deloitte) drag every 18-month roadmap to 36. Playbook: land on co-ops, price per-MW, file a model PUC case in a friendly state, ship an AMI-agnostic estimator, and win the data-centre wedge with a hyperscaler co-marketing partnership (AWS, Meta, Google) GridCARE will match but Gridsight will not.

Adjacent-segment play

The digital-twin plus DOE stack could be repackaged for three adjacent buyers. First: large industrial and commercial site owners — hyperscale data-centre developers (Microsoft, Amazon, Google, Meta, Oracle, CoreWeave, Crusoe), Prologis logistics campuses, mining companies electrifying haul fleets, EV-charging depot operators. These buyers need to know, at a given substation, whether the utility can actually deliver 200 MW in the next 18 months or whether they’re stuck in the queue. A Gridsight-lite sold to the load rather than the wires inverts the current pitch and lands opposite exactly the accounts GridCARE is chasing. Second: transmission ISO/RTO capacity intelligence — PJM, MISO, ERCOT and CAISO all publish some flavour of hosting-capacity map, all bad; a probabilistic capacity layer sold as a data product to developers, PPA originators and merchant traders is a real market Enverus and Wood Mackenzie only partially serve. Third: insurance for grid-tied assets — the same live model computes localised outage and voltage-excursion risk, an underwriting input Munich Re, Swiss Re and Aon are looking for on utility-scale battery and behind-the-meter DER fleets. The ISO product is a data business (lower margin, longer cycle); the insurance play needs actuarial partners Gridsight lacks. The site-owner adjacency is the highest-fit — and, awkwardly, where the well-funded competition is already thickest.

Sources and further reading

• Gridsight Raises $26M Series B to Unlock Electric Grid Capacity and Affordability (PR Newswire / Gridsight, September 1 2026)
• Gridsight Raises $26 Million to Help Utilities Unlock Grid Capacity with AI (ESG Today, September 1 2026)
• Energy network management platform Gridsight charges up on $36 million Series B (Startup Daily, September 2 2026)
• Childhood mates’ AI energy startup Gridsight raises A$36M led by Insight Partners (Dealroom News, September 2 2026)
• Gridsight raises $36m in Airtree-backed series B round (Capital Brief, September 1 2026)
• Fast bowler Pat Cummins backs AI-powered energy grid startup’s $7.5 million Series A (Startup Daily, April 8 2025)
• Endeavour Energy powers up AI flexible export service (Utility Magazine, 2025)
• GridCARE Raises $64 Million Series A from Leading AI and Energy Investors (BusinessWire, May 14 2026)
• Early Nvidia Investor Sutter Hill Ventures Backs GridCARE to Tap Idle Grid Power (Bloomberg, May 14 2026)
• Bentley Systems announces OpenUtilities DER Planning & Design Assessment Solutions (Automation.com)
• Surging energy demand puts US utility capex forecast near $1.3T in 2026-30 (S&P Global Market Intelligence, April 2026)