AI or A-bye

Man, I’ve been thinking about how the dot-com bubble turned into the dot-bomb fall and I can’t shake the feeling we’re replaying that season. Back then, every company stapled “.com” to the end of their name and boom, valuation. Today it’s AI everywhere, AI in your email, AI in your fridge, AI writing AI to manage more AI. It’s fun. It’s cool. It’s also starting to look like the same movie with different outfits.

I’m not saying AI is fake. I love this stuff. I literally use it every day to learn and build. But a technology can be real while a market gets ahead of itself. That happened in the late 90s. The internet was real. Prices were not.

Here’s why I think we’re moving from AI to A-bye when you look at the stock market narrative and the actual physics of building this thing.

The limiting stuff that doesn’t care about vibes

Power. Training and serving big models is a power problem. You can’t wish new electrons into data centers. Utilities move slow, substations move slower, and you can’t YOLO a grid upgrade because your Q4 deck needs a hockey stick.

Silicon. Chips don’t compound like SaaS logos. Foundries and packaging lines take time, money, and boring permits. Everyone wants more compute, but the pipe that makes compute is not infinite. Scarcity keeps costs high and starves the long tail.

Data. The web has a ceiling. We scraped a lot of it already. The next batch is gated, paywalled, copyrighted, or private. Synthetic data is cool, but it also loops on the same patterns. Garbage in, confident garbage out.

Quality. We went from bad to decent to wow really fast. Going from wow to reliable everywhere is a different sport. Diminishing returns kick in. You can spend 10x more compute and get a smaller, pickier improvement that most users won’t notice at 2 a.m. on a flaky Wi-Fi.

Inference bills. Demos are cheap. Daily active users who press Run 100 times are not. You either pass the cost to customers, who flinch, or you eat it and call it growth until finance taps you on the shoulder.

Distribution. Every giant platform already ships their own model or wrapper. If your product is a thin layer around the same models, your margins are their setting, not yours. Platforms don’t like to be your cost of goods sold forever.

Enterprise time. Pilots are fast. Security reviews, data residency, SOC 2, procurement, and “circle back next budget cycle” are not. Sales cycles went from weeks to quarters. That alone kills the quarterly “up and to the right” story.

Regulation. Not “maybe someday.” Real sectors already need explainability, audit, and provenance. That means slower releases, more headcount that doesn’t ship features, and fewer magical one-click upsells.

Talent. There aren’t enough people who can run giant training jobs, ML infra, and safety evals while keeping costs sane. Tools will help, but right now payroll lines are heavy and onboarding is not a weekend.

User trust. People love the magic until it hallucinates in front of their boss. Then they turn it off or fence it in. That changes usage curves from exponential to staircase.

Why this smells like dot-com again

Back then it was “eyeballs now, profits later.” Today it’s “tokens now, margins later.” Same mismatch, different metric. Valuations priced in a world where every task becomes AI-native, costs fall to near zero, and everyone pays premium for it. Reality is messier. Costs fall, then plateau. Adoption climbs, then stalls on trust or budget. The S-curve is real, but the steep part is shorter than the slide decks promised.

Also, in bubbles, narratives outrun accounting. You get second-order plays on top of first-order plays. ISPs, routers, portals, sock puppets. Now it’s model makers, model renters, fine-tuners, prompt platforms, AI-in-a-box for X. A lot of these are just wrappers. Wrappers compress when the platform decides to care.

“But AI is different”

Sure. The internet was different too. Being different doesn’t prevent repricing. The infrastructure that matters will stick around, and a handful of apps will become verbs. The rest will discover gravity. That is not anti-AI. That is how every platform wave matured. We overbuild, we crash, we consolidate, we go again.

What would throttle the hype into a fall

• growth in usage that shifts from “new” to “replacement,” so net adds flatten
• capex that stays high while revenue per token, per seat, or per API call slips
• power and chip queues that delay roadmaps by quarters, not weeks
• tighter copyright and data rules that raise the cost of training and retraining
• customers moving to smaller, local models for cost and privacy, which compresses top-line for the biggest providers
• platforms bundling “good enough” AI into existing products, which eats the indie wedge

Put all that into a market that already priced perfection and you get the same ending we saw with dot-com. Not a tech winter, more like a reset. From AI to A-bye, not because AI dies, but because the multiples do.

I’m still building with this stuff. I’m still going to learn it, teach it, and use it at work. I just don’t think a straight line on a chart is a law of nature. The internet survived the dot-bomb and then ate the world anyway. AI will also survive a drawdown and then live everywhere quietly, inside everything, where the slides stop and the work begins.

Not financial advice, just a young adult who is seeing history potentially repeat itself…