Writer Ed Zitron has been making a point that’s hard to shake: the amount of money AI needs to keep working at this scale isn’t something the industry can sustain in the medium term. Not “might struggle.” Not “will need to adjust.” Just quietly, structurally unsustainable.
I keep coming back to that because of what happened this week. By most accounts, consumer AI is having a moment again. Meta’s personal assistant Muse, along with its plush-looking mascot Jolly, turned into a surprise hit. OpenAI shipped Dots and people actually started using it. TechCrunch’s read is that you could reasonably call this a comeback.
So we’ve got a comeback and a warning sitting in the same news cycle. If you’re not technical, that combination is confusing. Let me explain what’s actually going on underneath it.
Popular and profitable are different things
When a consumer app takes off, we’re trained to read that as success. A million downloads means the thing worked. That instinct comes from the app era, where the cost of one more user was close to zero. Another person installing Instagram cost Instagram almost nothing.
AI assistants break that assumption. Every time you ask Muse something, or send a message through Dots, real computation runs on real hardware that someone is paying for. The cost doesn’t flatten out as you grow. It climbs right alongside your user count.
Which means a consumer AI product can be genuinely beloved and genuinely a money pit at the same time. Popularity is the expense, not just the reward.
Why everyone keeps talking about enterprise
This is the part that explains a lot of otherwise strange corporate behavior. TechCrunch’s framing is that consumer AI struggles to grow without enterprise revenue behind it. Businesses pay predictable contracts, negotiate at volume, and don’t flinch at a per-seat monthly fee the way an individual would.
Consumers, meanwhile, have been taught that software is free or close to it. Ask someone to pay twenty dollars a month for a chat assistant and a meaningful chunk of them will walk. Ask a company to pay for the same tool across four hundred employees and it’s a line item in a budget.
So the pattern you’re seeing is: build the charming consumer product, win attention and goodwill, and use that to sell to the companies that can actually cover the bill. The friendly mascot is the front of the house. The enterprise contract is the kitchen.
What that means for you as a user
- Free tiers are marketing, not generosity. They exist to build habit and reputation.
- Features can disappear. If a capability costs too much to serve for free, it migrates to a paid tier.
- Product direction follows the money. When enterprise pays the bills, enterprise priorities shape the roadmap, even in apps built for individuals.
- Stability isn’t guaranteed. Products that can’t find a funding path get folded or shut down, no matter how much users liked them.
The bigger question economists are asking
The worry isn’t only about individual companies. Oxford economist Carl-Benedikt Frey has raised a risk that he considers more serious than the usual list of AI drawbacks, one significant enough to overshadow them. At Stanford’s business school, Chad Jones has been working through the range of economic futures AI could produce, from abundance on one end to something much darker on the other. These aren’t skeptics shouting from the sidelines. They’re people whose job is modeling how technologies reshape economies, and the spread of outcomes they describe is wide.
Meanwhile the physical side is drawing scrutiny too. Data centers, the buildings where all this computation actually happens, have become a policy flashpoint. Colorado put a law into effect in February 2026 regulating high-risk AI in sensitive areas like hiring, housing, lending, and healthcare. Regulation is arriving at both ends: the buildings and the decisions.
How I’d read the next year
I don’t think the right reaction is cynicism about these products. Muse and Dots are useful. People aren’t being fooled into liking them.
What I’d suggest instead is a small adjustment in how you interpret AI news. When you see a launch announcement, ask who is expected to pay for this and when. When a free tool feels unusually capable, assume that generosity has an expiration date and don’t build anything critical on top of it without a backup plan. When a company pivots from consumer polish toward business customers, understand that as math rather than betrayal.
The gap between how much people love these tools and how much they cost to run is the real story this year. Mascots are easy to love. Infrastructure bills are harder to pay. Both of those things are true right now, and the next stretch of consumer AI depends on which one wins.
🕒 Published: