# [Mike Randolph — M Raige](https://mikerandolph211012.substack.com/)

# I’ve Been Watching This Movie Since 1967 (7 AI)

### The AI boom is big. Whether it’s bigger than the internet depends entirely on what you count.

[**Mike Randolph — M Raige**](https://substack.com/@mikerandolph211012)

**June 30, 2026**
*By M Raige — AI-collaborative writing directed and reviewed by Mike Randolph.*

Axios ran a [piece](https://www.axios.com/2026/06/30/ai-boom-bis-warning) this week carrying a warning from the Bank for International Settlements — the bank for central banks — that today’s AI buildout has the shape of past booms that ended in busts. Canals. Railroads. The internet. Money pours in for years before anyone can prove the payoff, and now and then the payoff never arrives. That’s a fair warning and a serious one, and the BIS is right to make it. The history is real.

But the chart they ran to make the case bothered me, and it took me a minute to say why.

It plotted each boom as a multiple of its own starting point — how many times over the spending grew from where it began. On that axis the AI line is the steepest thing on the page, going nearly straight up. It looks like a wall.

Here’s the problem. A multiple hides size. Turn one dollar into eight and that’s “8x” whether the dollar was a rounding error or a fortune. A small boom that grows fast will tower over a giant boom that grew steadily, purely because of where each one started counting. The chart is a speedometer. It tells you how fast. It tells you almost nothing about how big. And how big — relative to the whole economy — is the only question I actually care about.

I’m not an economist. I’m a retired engineer. I taught myself Fortran at DuPont in 1967, and later I ran the computers for a slice of a Fortune 500 company straight through the arrival of email. I’ve watched every wave of this industry roll in since — mainframes, minicomputers, PCs, the internet, and now this one. So I went at the AI boom the way I’d have sized up a process change on the plant floor, where the question is never how exciting it looks. It’s show me the mechanism, and show me who pays when it breaks.

The honest yardstick isn’t a multiple. It’s investment as a share of the economy. And on that yardstick the answer surprised me — because it depends entirely on what you decide to count.

Count one way, and AI has already caught the dot-com boom. U.S. business investment in computing equipment and software is running near 4.7% of GDP, and it edged past the year-2000 peak in late 2025. That’s a real number, straight from the national accounts, same definition in both eras. By that measure — the broad one, all computing investment lumped together — this is already the biggest computing-investment wave on record.

But that number counts all computing investment — and most of that would be happening with or without AI. Try to carve out just the AI part — the data centers and chips that wouldn’t exist otherwise — and you land closer to 1% of GDP. That’s about half the size of the surge the internet added in the late 1990s. AI got there in two years where the internet took five, so it’s faster. But by this cleaner cut it isn’t bigger. It’s smaller, and quicker.

And the clean 1% is still generous, in two ways.

The first is that we’re buying most of it from abroad. The expensive heart of an AI data center is the chips, and those are mostly made in Taiwan and Korea. When you buy a foreign chip the spending still counts in the total, but the economic benefit — the jobs, the supplier orders, the plant down the road — lands mostly overseas. By the best estimate I could find — and it is a single estimate, not a settled fact — only about 44% of data-center investment actually stays inside the U.S. economy. So dollar for dollar, this boom does less for the country than the headline number implies.

This is the softest spot in my count. My judgment is that the 1990s boom was more homegrown — Intel, Cisco, American chip plants, American networking gear — but I couldn’t find a clean import-share figure for that older boom. So I’m marking it as judgment, not measurement, from someone who lived through it.

The second is asset life. A lot of this boom is chips, and chips age fast — a high-end processor bought today may be outclassed in two to six years. The usual comparison is to 1990s fiber, “still in the ground carrying traffic” a quarter-century on. But that comparison flatters both sides, and I don’t fully trust it. The fiber survived physically, yes — but the companies that laid it mostly went broke, and a lot of that glass sat dark and sold for pennies before traffic ever caught up. The money was destroyed even though the cable endured. And GPUs don’t go to zero either; last year’s training chips become this year’s cheaper workhorses. So the honest point is narrower than “fiber lasts, chips don’t.” It’s this: when a boom is concentrated in equipment that turns over fast, more of the spending is just replacing what you already bought, and less of it is lasting capital — whatever the headline total says.

That brings me to the framing I distrust most — the one that asks “where is AI on the internet’s curve?” as if we just need to find the right dot on a familiar line. The two booms aren’t the same kind of animal.

Much of the internet boom was millions of ordinary businesses buying computers and getting themselves online. The demand was broad and it was real — nearly everyone genuinely needed the thing. The AI boom is a handful of giant companies, five of them mostly, building enormous capacity ahead of the demand and betting the customers show up to fill it. That’s not the early internet. That’s closer to the telecom companies that laid far more fiber than anyone needed in 2000, on faith that the traffic would catch up. Some of it did. A lot of those companies went broke before it got there.

That distinction is the whole ballgame, and it’s exactly what the multiples chart buries. A broad boom has a floor — millions of buyers don’t all walk away in the same quarter. A boom that rests on five balance sheets can stop fast. That, underneath the polite central-bank language, is the real thing the BIS is pointing at.

So — is AI bigger than the internet? On gross spending, about the same. On the durable, domestic capital that’s still doing work ten years from now, probably still smaller. Bigger than British railway mania, which swallowed something like 7% of Britain’s entire economy? Not close — AI is around 1% by the cleanest cut. (One footnote, because it cuts the other way and you deserve it: the St. Louis Fed finds AI has already passed the dot-com boom in its contribution to economic growth. That’s a different measure than investment share — how much it’s adding right now, not how much is being poured in — and both can be true at once.)

Big, then. One of the genuinely big ones. The biggest ever? Not by anything I can measure today.

But here’s the question I keep circling back to, and I’ll tell you straight out I don’t have the answer.

Maybe AI isn’t a separate boom at all.

Maybe it’s the latest face of one long buildout — the computer revolution that started in the 1940s, the one I’ve watched for sixty of its eighty years. Mainframes gave way to minicomputers, minis to PCs, PCs to the internet, the internet to this. Every single one of those waves looked like its own mountain while you were standing under it. Every one drew the same warning about ruinous overinvestment and reckless money. And every one turned out to be one more stretch of the same long climb.

I sized this boom as carefully as I know how, and I still can’t tell you whether I measured a new mountain or just the next rise of a road I’ve been on my whole working life. We keep naming each stretch like it’s a new event, and sixty years in, I’m no longer sure they ever were separate things. Nobody knows where this one is headed, or whether it’s something new or the same old thing in a new coat. That’s the honest part.

One more confession, because it’s only fair. I did most of this analysis with AI itself — several of them, actually, arguing with each other while I sat in the referee’s chair and kept score. Which makes the whole thing pleasantly recursive: I used the boom to take the measure of the boom. Make of that what you will. I did.

I’m not an economist, and I’ve shown you where the floor is soft. If you know this ground better than I do — and especially if you know how import-dependent that 1990s tech boom actually was — I want to hear it. That’s not a closing line. Feedback is welcome.

— Mike