We took a close look at Zendo Energy, the startup selling an "Energy OS" — forecasting software fused with an energy brokerage — to data centres drowning in AI-driven power costs. Inside: why up to 25% savings is really a story about stranded capacity, who actually buys this, and why the whole procurement category is optimising for the wrong number.
Every data centre operator I talk to has the same spreadsheet, and the same fear buried in it. The spreadsheet models energy costs as a line item that grows roughly with rack count. The fear is inference. Training a model is a marathon: predictable, plannable, amortised. Inference is a stampede, and nobody can tell you when it arrives. The IEA's projection of roughly 945 TWh of global data centre electricity demand by 2030, nearly double today's level, is usually quoted as an infrastructure story. For the person running procurement at a colocation facility, it's a personal one. Their tariffs, their grid connections, and their ESG commitments were all designed for a world where load curves were boring.
That's the world Zendo Energy walked into with its £1.75M pre-seed round led by Fly Ventures, with Octopus Ventures and Pact VC participating. The company, founded by Jade Batstone (previously at Square and Product Innovation Lead at SWIFT) and Drew Barrett (former Head of Renewable Procurement at Octopus Energy), sells what it calls an Energy OS: software for forecasting AI-driven workloads and optimising consumption, bundled with an energy brokerage that writes bespoke, flexible tariffs. The claim that matters is a cost reduction of up to 25% on energy spend, without compromising uptime. Unpublished pricing, but B2B subscription SaaS plus brokerage services.
The wedge: procurement as software, not as a contract you sign every few years
Traditional energy procurement is episodic. An operator negotiates a tariff, signs it, and lives with it until the next renewal, occasionally flexing demand when the price spike is big enough to notice. Zendo's bet is that this cadence is now absurd. In a market thick with intermittent renewables and wild price volatility, the optimal move changes daily, sometimes hourly, and the value sits in continuous optimisation rather than periodic negotiation.
Two things make this hard to copy. First, the software and the brokerage are fused. A forecasting engine that knows when your AI workloads will surge is only worth something if there's a tariff structure flexible enough to exploit that knowledge, and writing those contracts requires actual market access. Pure software vendors don't have it; pure brokers don't have the telemetry. Second, the founders have lived on both sides. Barrett ran renewable procurement at Octopus Energy; Batstone built product at Square and SWIFT. The founding engineer, Jason McGeeney, came from Meta's infrastructure team, where he optimised capacity and rack placement across a global fleet. That's an unusual stack of scars for a pre-seed company.
The concept they've made their own is "stranded power": capacity that a facility is contracted for or connected to but can't use, because it's locked behind inflexible procurement or poor forecasting. Unlocking it is often cheaper and faster than waiting years for a new grid connection, which matters enormously when AI tenants are queueing at the door.
The ICP they actually win: mid-size operators who can't wait for the grid
This isn't aimed at hyperscalers, who employ their own energy traders and build their own forecasting. Zendo's natural customer is the small-to-mid-size colocation or HPC hosting provider in the UK and Europe, scaling into AI demand, with an energy procurement lead who is drowning. The proof point fits exactly: Deep Green, a UK operator, is using Zendo for its new AI-ready facility in Manchester, powered through a flexible clean energy contract with ENGIE. Caplight also notes Fortune as a customer added in March 2026.
For these operators, the pitch is survival arithmetic. If you can't get a grid connection for two years but you can sweat 15% more capacity out of what you already have, that's revenue this quarter. If energy is your largest operating cost and you can shave up to 25% off it, your margin story to tenants changes entirely. Energy stops being a cost center and becomes, in their words, a competitive edge, which is the sort of line that usually makes me wince except that here it's literally true: cheaper power is a pricing weapon in colocation.
What the category still gets wrong: optimising for price per kWh when volatility is the real cost
Most competing approaches, from conventional brokerage to the newer flexibility players, optimise for the wrong number. They chase the lowest headline unit price on a fixed contract. That's the procurement reflex of a stable-load era. But in a renewables-heavy market, the cheapest average price can hide brutal exposure to shape and timing, and AI workloads make your load shape the least stable thing about you. The real number is cost per unit of compute delivered under uncertainty, which folds in volatility, carbon exposure, stranded capacity, and the opportunity cost of turning away tenants because you didn't forecast your own headroom.
There's also a quieter failure in the category: treating decarbonisation as reporting. ESG dashboards that count emissions after the fact are theatre when the operator's actual question is which contract lets me shift load toward cheap green hours without breaking uptime guarantees. Deep Green's ENGIE deal is instructive because it's a flexible clean energy contract, not a certificate purchase. The emissions cut and the cost cut come from the same mechanism.
The takeaway
If you run a data centre or sell to people who do, watch what happens to energy procurement over the next two years. The operators winning AI workloads won't be the ones with the newest buildings; they'll be the ones who can forecast their own demand, contract flexibly against it, and monetise capacity everyone else wrote off as stranded. Zendo is early, pre-seed early, and unproven at scale. But the wedge, fusing load intelligence with market access, is aimed at a genuinely broken process, and the founding team has the resume to be credible about both halves of it. The category question worth stealing from them: not "what's our tariff?" but "what does our energy strategy earn us?"