Kraken — the AI platform Octopus Energy built to run its own utility — now manages more customer accounts than the population of France. In December 2025, investors valued it at $8.65 billion. It has never sold a single energy bill under its own name.
Most utilities are still running the digital equivalent of a 20-year-old engine: billing, service, and grid systems that were never designed for instant digital interaction, let alone artificial intelligence. Replacing that infrastructure is so expensive and so risky that most energy and water companies just patch around the edges — and accept slow service, high cost-to-serve, and customer churn as the price of legacy tech debt. Octopus Energy hit the same wall in 2016. Its answer wasn't to patch. It was to build an entirely new AI-powered operating system from scratch, then license it to its own competitors.
That operating system, Kraken, now runs the back office for EDF, E.ON Next, Origin Energy, Plenitude, Tokyo Gas, and National Grid's US business — six energy majors that compete directly with Octopus in their home markets but still pay to run on its software. Combined, Kraken is contracted to serve more than 70 million household and business accounts worldwide, generating over $500 million in committed annual licensing revenue — four times what it billed just three years earlier, according to Octopus Energy Group's official September 2025 spin-off announcement.
What Kraken's AI Platform Actually Automates Inside a Utility
Strip away the energy-industry jargon and Kraken is two AI systems wearing one platform. The first runs customer operations: billing, service requests, complaint handling, and every email or call a customer sends in. The second runs physical assets: electric vehicles, home batteries, solar panels, and heat pumps that Kraken can schedule and reschedule in real time to keep the grid balanced. Both systems sit on the same data layer, which is precisely why Kraken can do things a bolted-on chatbot can't — its AI has native access to a customer's entire history, every meter reading, every tariff change, the moment it needs it.
This matters most for two groups: utility operations leaders drowning in legacy CRM systems that take months to change a single workflow, and energy companies trying to build an EV-and-battery flexibility business without spending years training their own AI models from zero. If you're an Operations Director at a mid-size energy retailer watching your cost-to-serve creep upward every renewal cycle, this is the playbook a $500-million-a-year licensing business was built on.
Inside the Two AI Engines Doing the Heavy Lifting
The customer-facing engine is called Magic Ink. Built on GPT-style large language models, it reads a customer's full interaction history and product details, then drafts a response, summarizes the case, or suggests a next action — such as requesting a meter reading — for a human agent to review. Because Magic Ink is built into Kraken's core data layer rather than bolted on as a separate app, it never has to wait on an API call to "remember" who the customer is. According to techUK's documented case study on the tool, around a third of Magic Ink's drafted messages need zero to minimal edits before a human sends them. Hallucinations are handled with a verification layer that annotates which facts in a draft are sourced and flags anything that couldn't be confirmed — a human always stays in the loop before anything reaches a customer.
The grid-facing engine is Kraken's residential flexibility platform. It coordinates electric vehicle charging, home battery discharge, and heat pump cycles against live, constantly shifting grid signals — shifting energy-hungry tasks to the windows when power is cheapest and cleanest, then shifting back when the grid needs support. Kraken's systems process 15 billion new data points every day to make those calls. The result, per Octopus Energy Group, is one of the world's largest virtual power plants built from residential devices, plus Europe's largest grid-scale battery, jointly orchestrating more than 2 gigawatts of flexible capacity.
Neither engine works as a stand-alone product a business can simply subscribe to. Kraken sells multi-year platform licenses to utilities, not software seats — which is the central difference between this and most of the automations covered on this site.
The Results: $500M in Revenue, 40% Lower Cost-to-Serve, and an $8.65B Valuation
According to Kraken's own published case study, Octopus Energy — the first company to run on the platform, starting in 2016 — achieved up to 40% lower cost-to-serve and industry-leading customer satisfaction scores eight years running, while scaling into the UK's largest energy supplier with close to 10 million customers globally.
On the customer-service engine specifically, techUK's case study reports that roughly 35% of customer emails handled by Octopus are now drafted with Magic Ink's assistance, and those AI-assisted responses earn customer satisfaction ratings around 70% — higher than responses written without it. To date, Magic Ink has summarized 6,239,087 calls (the equivalent of 695,379 hours of talk time) and generated a further 9,415,901 messages, with a human reviewing every single one before it goes out. [DATO A VERIFICAR — Kraken/Octopus's own marketing cites a virtual-power-plant consumer savings figure near $200M/£150M annually; this exact number could not be confirmed against a primary Octopus or Kraken press release and should be treated as directional, not audited.]
The competitive framing here is straightforward: EDF, E.ON Next, Origin Energy, Plenitude, Tokyo Gas, and National Grid's US arm didn't license Kraken because it was novel — they licensed it because the alternative was spending years and tens of millions building comparable AI in-house, while competitors already running on Kraken kept lowering their own cost-to-serve. Every quarter that gap doesn't get closed, it widens.
The Stack Behind Kraken's Utility AI Platform
This is enterprise infrastructure, not a self-serve SaaS tool — so instead of pricing tiers, here's what each layer actually does and how widely it's deployed.
| Component | Function in This Platform | Scale Today | How It's Accessed |
|---|---|---|---|
| Magic Ink | GenAI drafting, summarization, and next-action suggestions for customer service | 9.4M+ messages generated, 6.2M+ calls summarized | Built into Kraken; not sold standalone |
| Residential Flexibility Platform | Schedules EV charging, battery discharge, and heat pump cycles against grid signals | 2GW+ of orchestrated flexible capacity | Licensed module within Kraken |
| Kraken Customer & Billing Platform | Core CRM, billing, and service operations layer | 70M+ contracted accounts worldwide | Multi-year enterprise license (EDF, E.ON Next, Origin Energy, Plenitude, Tokyo Gas, National Grid US) |
| Kraken utility-grade AI / data layer | Real-time data processing underpinning both engines above | 15B new data points processed daily | Proprietary, cloud-based, not externally accessible |
None of these components has a free tier or a monthly price tag you can look up — Kraken sells direct, multi-year contracts to utilities, water companies, and now telecoms. The licensing model itself is the product innovation as much as the AI is.
Who Should Be Watching This Spin-Off — and Who Shouldn't Care Yet
This case matters most to utility and energy-company executives evaluating whether to replace legacy CRM and billing infrastructure, to anyone building or governing AI-assisted customer service in a regulated industry, and to operators trying to monetize distributed energy resources (EVs, batteries, solar) at grid scale without years of in-house R&D. It also matters to anyone tracking 2026's biggest tech IPO pipeline, since Kraken's mid-2026 listing target is one of the largest AI-platform spin-offs on the calendar.
It is not directly relevant to small or mid-size businesses looking for an off-the-shelf AI tool they can sign up for this afternoon — Kraken doesn't sell at that scale, and its entire value proposition depends on multi-year, enterprise-grade data integration that smaller operators typically don't need or can't justify.
Want More Real-World AI Deployments Like This?
This is one of dozens of verified AI automation case studies we've documented at Sityos AI.
Browse more real deployments — from fraud detection at Mastercard to self-healing power grids at Duke Energy — at sityos.com. New case study published every week, always sourced from primary company and press materials.
Frequently Asked Questions
Is Kraken the same company as Octopus Energy?
Not anymore. Kraken was built inside Octopus Energy Group starting in 2016, but as of the September 2025 announcement it is spinning off into an independent company. Octopus will keep a stake (commonly reported around 14%) after full separation, targeted for mid-2026, but Kraken now operates and sells to Octopus's own competitors as a standalone business.
How does Magic Ink make sure AI-written customer emails are accurate?
Every Magic Ink draft passes through a human agent before it reaches a customer — there is no fully autonomous send. The system also runs a verification layer that annotates which facts in a draft are sourced from the customer's actual record and flags anything that couldn't be confirmed, so reviewers know exactly what to double-check rather than re-reading the whole draft from scratch.
What is a virtual power plant, and how big is Kraken's?
A virtual power plant (VPP) uses software to coordinate many small distributed devices — home batteries, EV chargers, heat pumps — so they act together like one large, flexible power station. Kraken's residential VPP is described by Octopus Energy Group as one of the largest of its kind in the world, orchestrating more than 2 gigawatts of capacity, alongside Europe's largest grid-scale battery.
Kraken's bet is that the next decade of utility competition won't be won on price alone, but on who can run the most AI-automated operation at the lowest cost-to-serve. With a mid-2026 IPO on the table and six global energy majors already paying to run on its rails, the rest of the industry is now deciding whether to license that infrastructure or try to out-build it — on its own dime, on its own timeline.