ai is consuming power like we have endless amounts of it. we don't.
every cloud ai query your business sends feeds a data centre that draws more power than some small towns. there's a simpler way to run the same workloads.
the scale of the problem
ai doesn't run on magic. it runs on electricity. a lot of it.
the large language models that power cloud ai services — the ones behind every chatbot, every copilot, every "ai-powered" saas tool — run on gpu clusters inside data centres that draw power at industrial scale. a single ai training run can consume as much electricity as 100 households use in a year. inference (the bit that happens every time you or your system sends a query) is smaller per request, but it adds up fast when millions of businesses are sending millions of queries every day.
the international energy agency estimates that data centre electricity consumption could double by 2026. ai workloads are a significant driver of that growth. these aren't abstract numbers. they translate to real demand on power grids, real pressure on energy infrastructure, and real environmental cost.
and here's the thing most businesses don't think about: every time your booking assistant answers a guest question via a cloud api, every time your rostering tool processes a leave request through a hosted ai model, every time your dashboard pulls an ai-generated summary — that query travels to a data centre, gets processed on hardware that draws serious power, and travels back. you're paying for the compute. the grid is paying for the energy.
why this matters for operators
if you're running a tourism operation, an event company, or a transport business, you probably aren't thinking about the energy footprint of your software stack. fair enough — you've got guests to manage, staff to roster, and margins to protect.
but the subscription tools you rely on are increasingly ai-powered. and the infrastructure behind them is increasingly energy-hungry. the cost of that energy gets passed through in subscription pricing, api fees, and per-seat charges that climb every year.
you're funding the data centre — you just don't see the line item.
the local alternative
a local ai agent runs on a single device at your site. a mac mini. a small pc with a gpu. something that sits on a shelf in your office and draws about as much power as a gaming console.
that device runs the same class of language model that powers the cloud tools — quantised to fit the hardware, optimised for your specific workloads. roster queries, manifest generation, guest comms, knowledge lookups. all running locally. no round trip to a data centre. no contribution to the grid demand that comes with centralised ai infrastructure.
the power cost of running a local ai device is roughly $8–15 per month. that's it. no scaling costs. no per-query billing. no energy overhead beyond your own site.
it's not about being perfect
we're not claiming local ai solves the global energy problem. but we are saying that if your business can run its ai workloads on a device in your office instead of routing every query through a data centre on the other side of the world, that's a straightforward improvement. for your costs, for your data sovereignty, and for the energy footprint of your operation.
the technology exists to run capable language models locally. the hardware is affordable. the models are good enough for the structured, domain-specific tasks that most operations businesses actually need.
the question isn't whether local ai is ready. it's whether you need the cloud at all.
what we build
at blnk, every system we deliver runs on local hardware installed at the client's site. no cloud ai apis. no per-query costs. no data leaving the building.
we do this because it's better for the client — fixed costs, full ownership, no vendor dependency. but it's also better for the broader picture. fewer queries hitting centralised infrastructure. less energy consumed for the same outcome. a model that scales with hardware you own, not with data centre capacity someone else is building.
if you're running 4–6 saas subscriptions to manage your operation, and half of them are quietly routing your data through cloud ai infrastructure, it's worth asking: could this run locally instead?
for most operators we work with, the answer is yes.
blnk.nz · may 2026