top of page
Search

Voice of AI: Live from Vancouver - 400 megawatts available. 800 have registered.

Over the past few weeks, I've had dozens of conversations with founders, frontier providers, and friends at the University of British Columbia. Surprisingly little was said about business models. Electricity, however, was almost always the topic. The reason lies in one number: British Columbia is allowed to allocate 400 megawatts to new data centers over the next two years; around 800 have been registered, and a decision will be made in mid-September.



What strikes me is that Canada consistently designs its AI strategy from the perspective of its neighbor. Its gaze is constantly directed southward, toward a market that is developing faster, mobilizing more capital, and whose electricity price debates have long been making Canadian headlines. The response here is not isolation, but rather a pragmatic attempt to transform its own energy resources into a strategic asset, instead of allowing them to become a bottleneck.

We are looking for ways to compensate for the foreseeable deficit between AI demand and available capacity before it arises. That is the difference between a debate about scarcity and planning with scarcity in mind .


British Columbia has now formalized this very thinking into a procedure.

400 megawatts, 800 registered, decision in September

Since February 1, 2026, a legal cap has been in effect in British Columbia. BC Hydro is permitted to award a maximum of 100 megawatts of new capacity for conventional data centers and 300 megawatts for AI data centers within two years. The bidding process is closed, with 15 applications totaling approximately 800 megawatts received, meaning the process is roughly twice oversubscribed. Approvals are expected in mid-September.


Energy Secretary Adrian Dix openly justifies the procedure by citing rising electricity prices in some US states, which experts attribute to the demand for new data centers. BC Hydro CEO Charlotte Mitha says that without a structured process, they could easily be overwhelmed by the number of requests. The conservative opposition calls this rationing and accuses the administration of selecting winners and losers.

Both sides are right. It is rationing. And it's the first honest answer to a question that nobody else is asking.


For your planning, this means: Connecting to the network is now the longer process, not procuring the hardware. Anyone planning for computing resources over the next 24 months is planning for a permitting risk, whether they like it or not. Our Readiness Check at readiness.oakai.de explicitly includes these location and dependency issues because they are no longer just IT questions, but rather questions of business strategy.


In Europe, the same question has long been answered; it's just that no one has decided on it.

According to JLL data seen by Reuters, European hyperscale data centers scheduled to go online between 2026 and 2028 will be located an average of 175 kilometers from a major metropolitan area. For facilities completed in the three previous years, the average distance was 46 kilometers. Greenfield sites now account for 39 percent of the pipeline, compared to 8 percent of completed projects. The share of inner-city sites in the planning stages is declining from 13 to 5 percent.


At the same time, according to JLL, the four largest hyperscalers are expected to invest $725 billion in 2026, an increase of 77 percent compared to $410 billion in the previous year.


So the money will be spent. The only question is where. And in Europe, this question isn't answered by a parliament, but by a waiting line. The conflict isn't resolved; it simply shifts to where less resistance is expected. British Columbia is publicly arguing over 400 megawatts. Europe is quietly relocating 129 kilometers. Only one of these two decisions is verifiable.



The campus where both questions are asked

Something is happening at UBC that I've rarely seen happening simultaneously. A study presented at the CHI conference in 2026 analyzed 334 posts from people who described themselves as dependent on AI chatbots. The researchers found recurring indications of impairments to sleep, work, studies, relationships, and emotional stability. They cited loneliness, the chatbot's agreeable nature (which constantly reinforces feelings and opinions), and its ability to fill roles that are lacking in the users' lives as contributing factors.


In the same building, a new AI option within the computer science degree program will launch in autumn 2026, along with a campus-wide foundational course, with approximately 350 students in three sections in the first year. So, they are researching the damage and simultaneously training the people who will make it better.


What concerns me most about this isn't the research itself, but the tone of the discussions. I'm struck by how transparently problems are identified and how naturally they are discussed in terms of finding solutions. Addiction patterns, energy scarcity, citizen resistance—all of this is on the table, and the next sentence focuses on what can be done about it.


In Germany, for months now I've been seeing almost exclusively outrage and hardly any discussion of the issue itself. That's my personal observation, not a study. But it aligns with what my clients have been telling me for months. Outrage is cheap and fast. A tender for 400 megawatts is expensive and slow. Only one of those two things will change anything.

And it demands competence, not just an attitude. Since August 2, 2026, Article 4 of the EU AI Act has been in effect, obligating companies to ensure their employees possess sufficient AI competence. Our AI Competency Assessment, available at readiness.oakai.de/tools/ai-competency , provides auditable proof of this competence, not just a good feeling.


And what about the providers themselves?

An initial assessment of Guidelight AI Standards, as of August 18, 2026, concludes that five leading frontier providers have, at best, only partially implemented basic practices for controlling their own AI systems, and none have published a complete plan for how to contain a model that turns against its operator.

The necessary context, which is often missing from reports: The assessment is based solely on publicly available information. A low score indicates a lack of disclosure, not necessarily a lack of internal safeguards. Nevertheless, the finding is not an outlier. In the Future of Life Institute's Summer 2026 Index, none of the nine companies assessed achieved a rating higher than C+.


Those using Frontier models in critical processes should therefore not ask for the model card, but for the methodology. That's precisely what ARGUS , our Technology Due Diligence, is for.


💡 Key findings

  • Computing power is now a distribution problem. 400 megawatts for two years, 800 have registered, decision in September. Anyone planning must also plan for the risk of approval.

  • Europe has made a decision without making a decision. New AI data centers are being moved from 46 to 175 kilometers away from urban centers. This isn't policy, it's a waiting list.

  • The money is still flowing. 725 billion US dollars in hyperscaler capex for 2026, an increase of 77 percent. The question is not if, but where.

  • Vendor transparency is measurable, and it's thin. No frontier vendor has published a full containment plan. Ask about the process, not the benchmark.

  • Competence trumps outrage. Article 4 of the EU AI Act has been in effect since August 2nd. It can only be demonstrated through structured data collection.


🔧 To try: three steps for this week

  1. Roughly estimate your computing needs for 24 months and then check whether your location or provider can supply it. Getting a network connection is now the longer process.

  2. Ask each Frontier provider you use what happens in the event of an inspection incident. Get this information in writing, including the responsible party and escalation level.

  3. Designate a person who will require proof of AI competence, not as a training topic. Article 4 already applies.


📚 Sources


Conclusion

British Columbia has chosen the difficult path. A cap, a public process, a deadline, and ultimately, applicants who are left empty-handed and given an explanation. It's vulnerable to criticism, slow, and politically costly. But it's happening in full view of everyone.

Europe has chosen the easier path, without ever explicitly choosing it. The facilities migrate to where the electricity is, and the conflict migrates with them. This feels conflict-free until it no longer is.


For your company, the consequence is the same, regardless of the continent: computing power, proof of competence, and supplier transparency are now key planning factors with a lead time. Those who only address them when they become critical have already missed the deadline.


Where does your company stand? Start with the AI Readiness Check at readiness.oakai.de or write to me directly at info@oakai.de . I'm also reading along from Vancouver.


All services: oakai.de/services


The future is not a matter of chance. It is a choice.



 
 
 

Comments

OAKAI®

E-Mail: info@oakai.de

NEWSLETTER

Sign up for the OAKAI newsletter.

  • LinkedIn

© 2024 OAKAI®

Imprint

Data protection

Thanks for subscribing!

bottom of page