BC + AI
← All news

Infrastructure·

Both Hands on the Grid

By Kris Krüg

What the data-centre protests get right, where governments are getting it wrong, and a framework for building AI infrastructure with communities, not on top of them.

Both Hands on the Grid

The intelligence explosion needs a power source. That power source needs a conscience. Here’s where I stand on the machines we’re about to pour into our backyards: critique in one hand, curiosity in the other. I’m not putting either one down to make anybody comfortable.

Why I'm writing this

Twice this summer I sat down with doctoral researchers who study, of all things, how communities feel about data centres. It's patient, unglamorous work: knocking on doors and asking ordinary people what goes through their heads when a windowless server hall the size of a shopping mall gets proposed down the road. Their study crosses borders, with a site in Wisconsin, another spread across Washington and Oregon, and a researcher doing fieldwork right here in British Columbia. I told her she'd have trouble shutting me up. I care about this more than is strictly healthy.

Part of that is personal. I was born in the American Midwest. One of the fights they pointed me toward is in Menomonie, Wisconsin, a town of about 16,700 where a Delaware shell company called Balloonist LLC, fronting an undisclosed "tech giant," proposed a $1.6-billion data centre on 300 acres of farmland. Residents organized, asked the questions the developer wouldn't answer, and got their city council to rewrite the zoning and kill the project in January 2026. Then they did the part I really admire: they built a toolkit so the next town wouldn't have to start from zero.

The other part is professional. I run the BC + AI Ecosystem Association, Canada's first community-powered AI network. In our first year we're closing in on 300 dues-paying members, with more than a dozen regional chapters and a standing seat in rooms where this technology's future gets decided. I spend my days talking artists, founders, students, and public servants out of fleeing this stuff and into engaging with it. I'm not a Luddite. I'm not against compute. But I've read enough industrial history, in a province that knows what it feels like to be the place other people's futures get extracted from, to know that "the future's coming, get on board" is not an argument. It's a sales pitch, and I've been pitched by better.

The researcher and I kept circling one thing: the strongest position in her interviews isn't no. Most people aren't protesting AI itself. They're contesting the way it's being rolled out, the giant absentee-owned box that lands in the backyard with a benefit story and a non-disclosure agreement stapled to it. That distinction is the whole ballgame. Almost nobody in the public fight is drawing it cleanly.

So here's my position before the numbers. I'm not Team Fossil, and I'm not Team Fantasy. I'm Team Outcome. The AI build-out is real. Wishing will not un-build it. The question worth arguing about is whether it happens to our communities or with them. That's the paper. Everything below is me trying to earn that sentence.

The scale is real: curiosity's ledger

Let me give the boosters their due, because a critique that won't admit what's genuinely happening isn't worth the paper.

AI stopped being software a while ago. It's heavy industry now. It eats copper and rare earths. It eats land and water. Above all, it eats electricity in quantities that should make you sit up. Meta is building a single data centre called Hyperion that will draw five gigawatts, roughly a small province's worth of power running flat-out around the clock. That's not a metaphor for scale. That is the scale.

The independent numbers hold up. The Lawrence Berkeley National Laboratory found U.S. data centres already burned about 4.4% of the country's electricity in 2023 and could hit 6.7% to 12% by 2028. That's a tripling in five years. Globally, the International Energy Agency's Energy and AI analysis has data-centre electricity demand roughly doubling to about 945 terawatt-hours by 2030, more than all of Japan uses today, with the AI-specific slice tripling. In 2025 alone that demand grew 17%, five times faster than electricity demand overall, riding more than $400 billion in AI infrastructure spending.

Now the other hand: some of this could be worth it. Compute is genuinely useful. It moves drug discovery, climate modelling, materials science, translation, accessibility, wildfire detection, and yes, the creative work that pays a lot of my community's rent. These tools have made me more productive and more creative than at any point in my life. I'm not going to pretend otherwise to win an argument. No apology.

Civilizational infrastructure is exactly what you build deliberately, with your eyes open and on your own timeline. You don't build it with a gun to your head and an NDA on the table.

The bill comes due in the community: critique's ledger

The problem was never that the build-out uses resources. Everything uses resources. The problem is who pays, who profits, and who decides. Right now those three answers point in three different directions.

Who pays for the power. In the PJM grid, the biggest in North America, serving 67 million people from Virginia to Chicago, the 2025 capacity auction cleared at a record $269.92 per megawatt-day, nearly nine times the prior year. The grid's own market monitor pinned roughly 38% of that on data centres, about $6.3 billion of $16.4 billion. Ordinary households are eating it in their monthly bills. Virginia's legislative watchdog, JLARC, warned in December 2024 that satisfying unconstrained data-centre demand would require so much new generation and transmission that the cost gets shifted onto all ratepayers, including people who never asked for a server farm. Northern Virginia now hosts about 13% of the planet's data-centre capacity. The externalities are socialized; the profits are not. That's not a bug in the model. That's the model.

Who pays in water. Berkeley Lab estimates U.S. data centres used about 17 billion gallons directly for cooling in 2023, plus another 211 billion gallons indirectly through the power plants feeding them. Direct use could double or quadruple by 2028. Companies are improving, but the industry barely discloses facility-level numbers. That tells you something on its own.

Our side has an honesty problem too. "One query costs two bottles of water" is a killer protest line and a lousy unit of measurement. We don't price a table, a car, or a haircut in gallons. That doesn't make the worry wrong. A watershed is finite, cooling is real, and a drought doesn't care about your quarterly ESG report. But if we want to be trusted, we have to tell the science straight, not serve up a scary edit. Liz Marshall's question still cuts: is AI draining this world for another? Ask it honestly and it hits harder, not softer.

Who pays in the jobs that never show up. This one makes me angry because it's the promise every council meeting hinges on. A finished data centre is one of the least labour-intensive big facilities you can build. Two of them in Cedar Rapids, Iowa, with more than $1.3 billion invested, were on the hook to create sixty-one permanent jobs. Georgia shovelled hundreds of millions in tax breaks at the industry and, on audit, ended up paying about $289,000 per permanent job it could actually count. A typical facility runs on a few dozen full-timers, half of them contractors. The efficiency that makes these things profitable is the same efficiency that makes them garbage anchors for a local economy. Nobody says that part out loud in the ribbon-cutting photo.

None of this adds up to "don't build." It adds up to this: an honest ledger does not balance itself. Somebody has to make it balance. That's a political choice we are mostly choosing not to make.

The consent problem: what everyone is actually fighting about

Listen to the people protesting data centres. I did, at more than one Vancouver demonstration this past June, where the chants ran from "No data centres" to "Fund schools, not AI." Under the noise and the specific fights about water, power, and farmland, nearly every complaint comes back to one thing: consent. Decisions get made before the community is even in the room.

And the industry has earned every ounce of the suspicion, because it has built an actual machine for dodging consent:

  • Secrecy by design. Developers routinely hide behind shell companies, project code names, and NDAs signed by the public officials who are supposed to work for us. The watchdog Good Jobs First found that of at least 36 U.S. states subsidizing data centres, only 11 will even tell you which companies got the money. Public Citizen documented NDAs in 80% of Virginia localities that host them. In Tucson, residents learned "Project Blue" was Amazon only after the vote, when a document got mailed to a journalist by mistake. In Menomonie it was a Delaware shell. You cannot give informed consent to a counterparty you're not allowed to name. That's a signature on a blank cheque.

  • Environmental injustice, out loud. In South Memphis, Elon Musk's xAI installed as many as 35 methane gas turbines while holding a permit for 15, then ran 59 unpermitted turbines at its second Colossus site in Boxtown, a mostly Black neighbourhood where cancer risk already runs about four times the national average. It took the NAACP and the Southern Environmental Law Center threatening Clean Air Act litigation to force the issue. That is not a rounding error. That is the build-out telling you whose lungs it considers expendable.

This is the part the fight keeps tripping over: consent isn't real when the power gap is total. We spent the last decade learning that "they agreed to it" means nothing when one side holds every card. Now scale that up to the civic level. A rural community of a few thousand people, with no standing organizing network, no legal budget, and a two-week comment window slipped in over a holiday, is not a free and equal party across the table from a $15-billion counterparty with a federal partner and a wall of lobbyists.

The researcher said it plainly from her fieldwork. The places already fighting effectively are cities because cities come with organizing muscle and political networks already built. Small towns, where the biggest facilities get sited, get the announcement first and the meeting second. That asymmetry isn't an accident. It is the site-selection strategy.

I've been saying versions of this for years, so let me put it flat: governance is set by whoever shows up. Shell companies, code names, and comment windows timed for the week between Christmas and New Year's all serve the same purpose. They keep the community from showing up until the concrete is poured. Menomonie's real win wasn't stopping one building. It was refusing to be kept out of the room.

The movement, read honestly

Now the harder thing, because loving a movement means being straight with it.

The anti-data-centre movement is mostly right about process, water, ratepayers, consent, the theft of creative work that fed the training sets, and the jobs that never materialize. Its critics need to hear this: you can grant every complaint about the movement and it still works.

Yes, some signs are unsophisticated. "Use your brain, not AI" collapses under about four seconds of scrutiny. Nobody marches with use your brain, not your calculator. The messaging slides from a precise, winnable demand ("no data centre on this block without real consultation") to a fuzzy one ("what about the water") to the frankly absurd. Grant all of it. The movement is still doing the single most valuable thing on offer: dragging power to the table. Cities are pausing builds to do the consultation they skipped. Ministers who wouldn't return an email are suddenly hungry to "dialogue." The value was never the veto. It's the summons.

Where the movement genuinely runs out of road is the blanket "No data centres" itself. That slogan is the mirror image of the boosters' "get on board or get left behind," and it fails for the exact same reason: it pretends the choice is binary when it isn't. The compute is coming; it's going somewhere. If it doesn't land in a jurisdiction with clean power, real labour standards, Indigenous rights law, and a functioning civil society, it lands in one with a gas turbine, a tax holiday, and an NDA. "No, not here, not ever" doesn't stop the machine. It just aims it at the community with the least power to say no. That's not a victory. That's a hot potato with a human cost.

I hold both hands here too. The doomer saying refuse and resist is being just as lazy as the booster saying adapt or die. Menomonie found the useful stance: not like this. Not in secret. Not without consent or our fair share. Not until you answer our questions in public. That's not obstruction. That's citizenship with its sleeves rolled up.

There's also a quiet hypocrisy threaded through this moment, and the movement is not exempt. Look closely at who's holding the megaphone and you'll find people who use these tools every day. Think of the teacher running lesson plans through AI while accusing students of cheating with the same tool. That's human. I'm not here to scold anyone for it. But a movement gets stronger when it can say the honest version out loud: I use this. I've chosen not to use it for some things. Here's the messy middle we have to govern together. That speech is harder than "no." It's also the only one that survives contact with reality.

How our politicians are handling it: the Canadian mirror

Here's where I get parochial. Canada, and British Columbia in particular, is lining up to fail this test in a very Canadian way. I would very much like us not to.

We have an extractive reflex wired deep. Vancouver's stock exchange is basically a global casino for junior mining companies. We see a resource and the lizard brain asks how do we flip it, not how do we build with it. Watch that reflex hit AI and you get Alberta. Kevin O'Leary's proposed "Wonder Valley" near Grande Prairie would be a $70-billion, up-to-9-gigawatt, gas-powered data-centre complex. The Sturgeon Lake Cree Nation says it got "zero consultation" about the use of its traditional territory and fired a cease-and-desist at the premier. That's the old colonial script with server racks where the pipeline used to be. Dig it up, ship it out, apologize at the ribbon-cutting.

Quebec shows the pivot in real time. It slammed the door on crypto mining and swung it wide open for AI, roughly doubling its large-data-centre power rate to about 13¢/kWh and drafting new rules to match. British Columbia, to its credit, has shown a spine. BC Hydro paused new crypto connections in 2022. In late 2025 it moved to a permanent ban on new crypto mining while carving out a deliberately modest competitive call for power, up to 400 megawatts over two years, split 300 for AI and 100 for other data centres. The message underneath is exactly right: not all compute is created equal, and we get to choose what kind of digital future we power.

I want to be fair to that, because it genuinely beats Alberta's open-door approach. But being the best actor in a race to the bottom is not the same as getting it right, and the BC story has a second half the press releases skate right past.

In May 2026, Ottawa's AI minister, Evan Solomon, stood up in Vancouver with TELUS to announce a "sovereign AI" cluster: an expansion of TELUS's Kamloops facility plus two new Vancouver data centres. One of them, the M3, sits in my own neighbourhood of Mount Pleasant. It starts at 85 megawatts of BC Hydro power, scaling toward 150 megawatts and more than 60,000 GPUs, under Ottawa's "Enabling large-scale sovereign AI data centres" program.

Read that against the cap. The province rationed AI power into a scarce public pool. Then the anchor tenancy went to the incumbent national telco. Ottawa says its strategy will grow the industry by "championing innovation leaders." Champion the champions. It's an honest description of the policy and, I'd argue, the whole problem. It makes life easy for incumbents and brutal for upstarts: frontier builders, co-ops, and Indigenous-led projects that could change who owns this future.

Notice what still isn't nailed down. At the time of the announcement, no federal funding had been committed or distributed. The terms of the public's stake were still being written. That is precisely when the public should be in the room. From where I sit in Mount Pleasant, the local land-use process for a data centre felt like a short online comment window dropped over the December holidays, advertised in the blandest language available, naming neither the company nor the federal partner. Whatever the legal minimum was, it was met. Whatever consent means, that wasn't it.

Zoom out and Canada's federal AI money seems to sort the country into roles. Ontario gets the compute-and-capital story. Quebec gets ethics and governance. Alberta gets to be the energy patch. British Columbia gets the pick-and-shovel job again, a resource province with a modest carve-out and a defence-adjacent angle off the coast. Most parents sending their kids to UBC probably don't realize we've been slotted back into the role of resource outpost, this time with GPUs in the pitch instead of gas. That's the gap I was trying to close in Community AI Pathways for BC's Look West Report: industrial strategy needs a community layer or it becomes another extraction plan.

The "we can't afford to do it differently" excuse dies on the balance sheet. At the same moment BC rationed AI down to 400 megawatts, it was committing billions to the North Coast Transmission Line to electrify LNG, ports, and mining in the northwest. We find power and capital when we decide something matters. What's scarce is the political will to spend either one on us.

The failure mode is the same across these governments. They bounce between boosterism (jobs! investment! sovereignty!) and reactive bans, then mistake the bouncing for policy. Neither reflex asks the question that matters: what do we get, verifiably, in exchange for our power, water, and land? Answering it demands disclosure, and these deals are engineered to prevent disclosure.

One fact should reorganize everyone's thinking north of the 49th. In December 2025 the BC Court of Appeal confirmed that our Declaration on the Rights of Indigenous Peoples Act gives UNDRIP real legal force in this province. In BC, free, prior, and informed consent is no longer just a value statement. A developer asking for a fast, quiet, consult-later process should expect more than a public-relations problem. Alberta is learning the political price of pretending otherwise. We should not have to learn it twice.

The institution we forgot to build

I have to sit with the heaviest thing in this file, because it's the reason none of this is abstract.

In February 2026, an eighteen-year-old walked into Tumbler Ridge Secondary School in northeastern BC and killed five children and an educator, after killing two members of their own family. It later emerged that OpenAI had internally flagged and banned an account linked to the shooter months earlier, and told no one who could have acted on it. British Columbia is now pursuing legal action against OpenAI; the families have filed their own suits.

I raise this carefully because it makes a point no energy statistic can. When that flag went up inside a multinational, there was no one in our system built to receive it. No commission. No clear duty-to-report regime. No legal line for an AI-safety escalation that protects privacy and still lets someone act. We hadn't built the thing.

The lawsuits matter, but you cannot litigate your way into an institution that doesn't exist. If we mean what we say about caring, the work is to build the receptacle: the regulation, the standing body, and the clear legal line a company is obligated to pick up. If there's a settlement, the useful question isn't "who do we blame?" It's "what do we build with it?" That is the difference between a society that is governed and one that is merely marketed to.

The dialogue is broken, and the hype is cracking

The public conversation has collapsed into what I call the Binary Trap: boosters versus doomers, each performing total certainty, neither one useful. The ground is moving under both of them.

Look at the SFU Centre for Dialogue's two-year study. Between 2024 and 2026, British Columbians' awareness and understanding of AI went up. So did their concern. By 2026, 79% were "more concerned than excited", up from 72%, and 65% said the risks outweigh the benefits. You cannot pin that on people not knowing enough. They know more and they're more wary. The adoption "lag" isn't a knowledge deficit. It's a rational read by people who've clocked that the upside has an out-of-town address. This isn't Luddism. It's wisdom wearing work boots.

The hype really is cracking. MIT's NANDA project reported in 2025 that about 95% of enterprise generative-AI pilots in its study showed no measurable return. I'm not calling a crash. I'm pointing out that we're being asked to hand over gigawatts, watersheds, and farmland on the strength of a revenue story that hasn't materialized yet. When infrastructure is permanent and returns are speculative, the community deserves to be the senior creditor, not the last one paid when the music stops.

Two things poison this dialogue worse than any bad number. The first is the cult of inevitability: "it's happening, get on board, you can't stop it." We can accept the half-truth and still refuse to be silenced by it. "Inevitable" is doing an enormous amount of dishonest work. Can we pause a year to think? No. Six months? No. A single weekend for one national conversation? Somehow, still no. A future that can't survive one weekend of public deliberation isn't inevitable. It's just well-funded.

The second is worse. It's creeping out of the accelerationist fringe and into rooms full of reasonable people in suits: the argument that any pause is immoral. If we slow down, they say, we're responsible for future humans who won't be born, cured, or lifted out of poverty by the optimized world we were too timid to build. I have watched a consultant deliver a version of this from a stage. It's a con dressed as ethics. It invents a fantasy of the future, measures the gap between fantasy and reality, then bills the difference to anyone who raises a hand. That's not accounting. That's coercion with a spreadsheet.

The way out of the Binary Trap isn't to split the difference between hype and doom. Change the axis. Stop asking are you for it or against it? Ask what does it cost, who consents, and can you prove the benefit? Liz Marshall calls the stance CASK: Curiosity, Awareness, Skepticism, Caution. Hold all four at once.

That's the conversation I want researchers, protesters, premiers, and tech founders shoved into the same room to have. Bring the elders too. Our deliberative institutions, and the people we prize for judgment and long memory, are built to move slowly on purpose. AI is lapping them. Designing a politics that can move at this speed without throwing away the people who can't might be the whole problem.

What I actually want: outcomes, in writing

Critique without a proposal is just complaint, so here's mine. If AI wants our gigawatts, it owes us gigaton outcomes: real, measured, verified outcomes, not a marketing deck. Before the asks comes a vision, because you don't beat a bad story with a list of demands. You beat it with a better story.

Mine is built for this place: clean, green, just, Indigenous-owned, hydro-powered tokens as British Columbia's export to the future. We have the water, the hydro, a reconciliation mandate with legal force, and an AI research base most provinces would trade for. We also know how to organize an industry into existence. A smart, long-running interactive digital media tax credit, made permanent and bumped to 25% in 2025, helped British Columbia grow a creative-tech cluster of 250-plus companies and 16,000-plus jobs in games, VFX, and animation.

Do it again with compute. Keep the best of it home to power our creative industries and researchers. Carve out capacity for education and public innovation. Sell the surplus at a green premium to buyers willing to pay for clean provenance they can defend in an annual report. That's not charity. That's a moat.

There's more than one shape this can take. The default, a giant absentee hyperscaler, is not the only way to deliver AI. Two people sitting in a restaurant already have AI-capable chips in their pockets that cost two grand and spend the day doing nothing. Networked community compute, mesh models, co-ops, and Indigenous-led facilities on northern hydro are different animals from a gas-fired box in a Black neighbourhood's backyard. Policy should be able to tell them apart and prefer the good one on purpose.

Here are the concrete asks for BC and any jurisdiction serious about doing this right:

  1. A public Compute Registry. Every data centre over 10 MW publicly discloses power source and carbon intensity, water use, methane controls if it burns gas, waste-heat recovery, local employment, and any Indigenous-nation partnerships. No registry, no interconnection. Sunlight is the cheapest regulation we own. It's also the one these deals are engineered to avoid.

  2. Consent, not consultation. Equity, not benefit crumbs. Where a project touches Indigenous land, Nations get board seats and ownership stakes from day one, not a late-stage benefit agreement slid across the table after the permits are signed. In BC, DRIPA already points here. Leadership means going past the legal floor into real partnership. Indigenous data sovereignty, including the FNIGC's OCAP principles, is a working prototype for how every community should relate to this infrastructure.

  3. A community stake, backed by ownership. If public power and public land make a facility possible, the public should own a piece of what it produces. That could mean capacity or revenue routed to host First Nations, local universities, community operations, and a per-resident dividend through a sovereign compute-and-wealth fund. Put what we keep into the announcement and the whole fight changes overnight.

  4. Make the load pay for the load. Ratepayer households should not quietly subsidize a hyperscaler's grid connection. Ireland now requires new data centres to bring their own generation or storage sufficient for their demand. Ring-fence the costs and require demand-response. When the grid is stressed, compute flexes. No exceptions or carve-outs for the biggest players.

  5. Kill the NDAs. No public official signs a non-disclosure agreement with a data-centre developer, and no council votes on a project whose true owner it cannot name. Menomonie proves this is winnable at the municipal level tomorrow morning.

  6. Seed the upstarts. Pay the people whose work built this. "Champion the champions" leaves frontier builders, co-ops, and Indigenous-led projects fighting over scraps. Rebalance it. The creators whose copyrighted work trained these models are also the people being told to retrain. Fund them to lead the upskilling, and pay for it with the plain recognition that they were the raw material all along.

None of this is anti-business. It's business with the bill paid. Tech giants are already restarting nuclear plants and signing gigawatt power deals. They know the energy math doesn't close without enormous clean power. The open question is whether that expansion happens to communities or with them.

Coda: both hands, still full

I keep coming back to those researchers patiently gathering what people actually feel, and to Menomonie, and to the strange comfort of a cross-border fight that turns out to be one fight wearing different hats. A farmer in Dunn County, an artist in Mount Pleasant, and a Cree Nation north of Grande Prairie are being asked to trade land, water, and public trust for a revenue story and an NDA. In their own idiom, they're saying the same three words: not like this.

That's not fear of the future. It's a demand to co-author it. The province that offers real agency will run circles around the one selling inevitability.

I want the machines. I want them badly enough to insist we build them right: methane monitors on, registry public, consent real, outcomes measured instead of promised. Curiosity in one hand, critique in the other. I am not putting either one down.

If AI wants gigawatts, it owes gigaton outcomes. Measure that, not vibes.

Sources & notes

This is a position paper, not a peer-reviewed brief. Every number is sourced inline, and contested figures are flagged. Key references, consolidated:

Kris Krüg is Executive Director of the BC + AI Ecosystem Association. BC + AI runs practical AI programs, community labs, meetups, and Film Club screenings across British Columbia.