Reading the Dashboard

My wife and I sat down the other night to talk about where to put some of our savings. A popular YouTube financial influencer was on in the background, walking through the sectors he was certain would dominate the rest of the decade. AI. Quantum computing. Drones. Robotics. Defence. At some point she turned to me and said: "You should be into all of this. It's tech. You're into tech."

She was right. I am into tech. I have been for a long time. And I don't want to put our money into any of it right now. The upside might be real, but I don't want to bet on a future I don't want to live in. We agreed pretty quickly: whatever we do with this money, we want it going to a future we'd want to be alive for.

"What about AI though?" she asked. We're using it every day, in both our personal and professional lives. Surely that's something we could invest in. I paused. I told her I don't trust the news enough to know what's even being sold. A lot of what gets reported about AI is hype, and rumours are either leaked or planted deliberately to move stock prices. The analysis downstream of that is shaped by distorted data. I've been watching this industry long enough to know I can't be confident in what I'm reading in my morning feed, and neither of us wants to put real money behind something we're not confident in.

So the conversation didn't end, it just rolled into the next one. We've been having a version of it for over a year now, and it keeps getting harder to get a clear sense of what's going on because the dashboards we're looking at are impossible to trust.

Some of what's making this hard to read clearly is structural. The boosters are inflating the numbers. The doomers are inflating the threat. Both sides are making the picture less legible than it needs to be.

Some AI safety groups have been paying social media influencers to convince the rest of us that superintelligent AI is going to kill us all by 2028. The Washington Post recently reported on the effort to seed doom messaging into ordinary lifestyle creators' content. When you ask the actual researchers, only about three percent rank existential risk as their primary concern. The messaging keeps coming anyway, partly because the loudest people stand to benefit regardless, and nothing generates engagement like controversy.

On the other side, the dashboard that news addicts like myself look at each morning, the one made of stock indices, AI capex announcements, and productivity headlines, is being inflated too. Token consumption gets reported as booming AI demand, but a lot of it is agentic AI doing more work behind the scenes for the same flat-rate subscription prices that were set two years ago when ChatGPT and Claude were glorified chatbots. Nowadays agentic tools might fire off hundreds or thousands of internal steps to complete a single user request, and the numbers make it look like an industry experiencing exponential growth. In reality it's a subsidized model that can't last. The underlying economics don't work without consumption growing tens of thousands of times its current rate by 2030.

Phantom GDP, a term Dylan Patel from SemiAnalysis uses, describes what happens when massive AI infrastructure spending shows up as economic growth before any of the productivity it's supposed to deliver has actually arrived. In a recent interview with Patrick O'Shaughnessy, he argues this is currently propping up a meaningful share of US growth. We're pouring the concrete now, on the bet that the value will catch up later. If it doesn't, a lot of what currently reads as a booming economy will turn out to be the cost of pouring concrete.

The physical world isn't keeping up either. Bloomberg has documented what's being called the AI memory crisis: memory chips, semiconductor packaging, and CPU supply are all constrained, with hyperscalers buying up production years in advance. Some of the biggest data centre build-outs are running ahead of the power grids needed to run them. The helium that cools the lithography that prints the chips that run the models is at the mercy of a war thousands of miles from any of it.

None of these are dashboard items. Yet these are the conditions running underneath everything. Meanwhile, people are becoming increasingly wary of how all of this is playing out, and some are starting to act.

The political body in Canada is finally starting to remember it has one. Manitoba just announced it will ban kids from accessing AI chatbots, the first province to draw that line. The federal government meanwhile has committed billions of dollars to a sovereign AI compute strategy. Cohere just announced a merger with the German lab Aleph Alpha on the explicit thesis that European and Canadian institutions must maintain control of their AI and not outsource it to foreign providers.

Greece, where democracy started, is going further: the prime minister has proposed amending the country's constitution to require that artificial intelligence "serve the freedom of the individual and the prosperity of society."

The reality of AI control is deeply asymmetrical, however, and likely to stay that way. A small number of corporations and states will continue to control the frontier. Everyone else will be working with what they can cobble together: smaller models, open-source weights, regional efforts, all of it running on older hardware. Both tiers will become deeply embedded in everything. The systems that shape your job, your loan, your kid's school day, your local hospital's triage, your border crossing, will all run on some combination of frontier and not-frontier AI, often invisibly.

The big question is who controls those frontier systems, and who decides how they get used. The answer, today, is that almost nobody has serious leverage outside a handful of San Francisco offices and a few state-backed labs in Beijing, Hangzhou, and Shenzhen.

The U.S. and China have both decided that AI is a strategic technology, both commercially and militarily. The major labs like OpenAI and Anthropic have promoted this framing since the beginning, and leveraged it to justify extreme levels of spending and expansion.

While these powerful corporations and governments are gung ho, some people are getting angry. Last month a city councilman in Indianapolis was woken up in the middle of the night by the sound of gunfire at his front door. His eight-year-old son was inside. Fortunately no one was hurt. On the doorstep was a handwritten note that read, "No Data Centers." Four days later, a twenty-year-old from Texas threw a Molotov cocktail at Sam Altman's home in San Francisco and was arrested an hour later trying to smash the glass doors at OpenAI's headquarters with a chair. He had been carrying a manifesto about AI-driven human extinction. He was eventually charged with two counts of attempted murder. 

These were two unconnected events with very different grievances behind them, in the same week. One driven by a proposed data centre in the community. The other driven by an AI extinction narrative. Both were reactions to the gap between what AI is being sold as and what it actually feels like, on the ground, in the lives of people who didn't get a vote on any of it.

That gap is likely going to keep producing reactions, and some of them could be violent. The small group of people who imagined themselves as architects of the future will find that other people do get a vote, and some of those votes might be cast in unsanctioned ways. None of this is going to resolve itself, and the institutions in charge appear unable to keep up.

What I would say to my wife, or anyone else trying to think clearly about this moment, is that there are three things worth investing in right now, and none of them are listed on any financial influencer's slides.

The first is in your understanding of how AI technology actually works. You don't need to become an expert, but be able to use it competently, recognize when it's being used against you, and talk about it honestly with the people in your life who feel uncertain about it. Most of the leverage in the next decade will go to people who understand what these systems can and can't do. The people who refuse to engage are more likely to be taken advantage of, outcompeted, or both.

The second is in the people around you. Think mutual aid, local groups, family and friends, professional networks, neighbourhood resilience. These exist to build redundancy beneath the larger system, so that when it glitches or seizes or starts falling apart there's a layer of human relationships that holds things together. It's what got communities through every previous upheaval.

The third is the political body you reside in. Whether it's at the municipal, provincial, state, or federal level, pay attention and participate however you can. The Cohere-Aleph Alpha deal happened because two governments decided sovereignty was worth pushing for. Manitoba acted because parents made it a political problem. The Greek amendment happened because a head of state decided it was worth writing into the constitution. The companies building this technology are not used to being told no, and the regulatory window is open right now in a way that may not stay open for long.

My wife and I still haven't decided where the money should go. Every week the landscape shifts. We're paying attention to the news, trying to make sense of the data, and asking each other how we can trust our instincts when the dashboard is giving us mixed signals.

So we're still deliberating. At least the conversation keeps getting more interesting the longer we sit with it.

Research and editing assistance provided by Claude. Images created with Midjourney.