The Fruits of Efficiency — Can You Reach Them?

Two news stories from the past two days deserve to be read together.

First: IBM's stock plunged 13% in a single day, wiping out over $31 billion in market cap. The catalyst: Anthropic announced that Claude Code can automatically handle COBOL modernization migration. COBOL — a language whose average developer is 58 years old, responsible for 95% of U.S. ATM transactions — has fed an entire consulting industry chain dominated by IBM for decades. One product launch, and the moat IBM spent half a century building cracked open.

Second: the next day, Meta and AMD signed a five-year procurement contract for up to 6 gigawatts of AI compute chips, estimated at $60-100 billion. AMD rose over 10% that day.

One side evaporates $31 billion. The other spends $100 billion. Capital is flooding from the "software middle layer" toward "compute infrastructure" — fast enough to track with the naked eye.

Just before these headlines, prominent Wall Street market commentary The Kobeissi Letter published an analysis with an enticing core thesis: AI reduces service costs, and 80% of U.S. GDP comes from services, so AI is essentially an "invisible tax cut." No congressional vote needed, no government appropriation — tech companies lower your bill for you. An even bolder inference: when production costs collapse globally, even the motivation for war disappears.

Really?

The Funnel Trap

"Technology brings prices down for everyone" — this line has been repeated countless times over the past forty years. Let's look at the actual report card.

From 1980 to today, personal computer prices have fallen 99.9%. The phone in your pocket has more computing power than the entire NASA system that sent humans to the moon in the 1960s. If Kobeissi's logic held, we should already be living in a golden age of cheap everything and universal prosperity.

The reality is the exact opposite. Over the same period, U.S. housing costs rose over 400%, college tuition over 1,200%, and healthcare spending swelled from 8% of GDP to nearly 18%. A computer dropped from $3,000 to $300 — where did the $2,700 in savings go? The answer is simple: into landlords' pockets, university administrators' salaries, insurance companies' income statements. Technology did push down the prices of certain goods, but the savings never flowed back to ordinary people's accounts. They were absorbed by other parts of the system — precisely, without a cent left over.

This isn't an accident. It's structural. Since 1948, U.S. labor productivity has risen by roughly 250%, while real wages over the same period increased less than 120% (source: Economic Policy Institute, starting from 1948 — the post-WWII economic expansion starting point). Where did the 130+ percentage-point gap go? Capital returns, shareholder dividends, executive compensation, stock buybacks. Everywhere except your paycheck.

So when someone says "AI is an invisible tax cut," the right response isn't "great!" It's to look back and ask: last time there was an invisible tax cut, whose pocket did the money end up in?

Inflation isn't a natural disaster. Inflation is a distribution choice.

Scarcity Moved

Kobeissi's argument has an even more fundamental problem: he assumes AI eliminates scarcity.

AI didn't eliminate scarcity. It just relocated it.

What used to be scarce: software engineers' brainpower, lawyers billing by the hour, accountants doing manual audits. Those are indeed getting cheaper. An AI Agent can finish in minutes what used to take an entire team a week. The per-seat billing model that SaaS companies lived on is collapsing. Salesforce fell over 50% from its peak, briefly dropping below $178. Intuit is down 44.5% year-to-date. The software sector has evaporated over a trillion dollars since the start of the year.

But new scarce resources have already emerged — and they're all in the physical world. Wafers, electricity, cooling water, data center land. TSMC's advanced process capacity has no global substitute. A single AI data center consumes as much electricity as a small city. Meta's 6-gigawatt contract with AMD is roughly equivalent to the output of 6 nuclear power plants. U.S. Treasury Secretary Scott Bessent publicly said at Davos this year that Taiwan is a "single point of failure" for the global economy.

Scarcity moved from the software layer to the physical layer. From brainpower to electrical power. From code to silicon and copper. And the ones controlling these physical-layer resources aren't you or me — they're the handful of giants with a combined $650 billion in annual AI infrastructure investment.

Hidden in the Meta-AMD contract is a detail worth examining: AMD issued up to 160 million warrants to Meta, representing roughly 10% of AMD's equity, unlocking in tranches based on chip delivery volumes. The more Meta buys, the larger its stake in AMD becomes. The buyer becomes a shareholder; the supplier becomes a controlled entity. This isn't a procurement contract — it's the enclosure movement of the compute era.

Flip to the page in your history textbook about the salt and iron monopoly — the logic is identical. Emperor Wu of Han monopolized salt and iron. Today, Meta and Google monopolize compute. The tools have changed. The rules of the game haven't.

The Extinction of the Cognitive Middle Class

Behind those compressed SaaS seats stand living, breathing people.

Nobel Physics laureate and "Godfather of AI" Geoffrey Hinton warned again last month: AI is accelerating the replacement of complex cognitive work, including software engineers. Forrester predicts the U.S. will lose 10.4 million jobs by 2030. A Goldman Sachs report says widespread AI adoption could replace 6-7% of the American workforce.

The numbers are cold, but the coldest part isn't the numbers. A Dallas Fed study points out that what companies replace with AI isn't just "positions" — it's "the entry point for young people to learn industry knowledge." Junior analysts, associate lawyers, trainee accountants — these roles exist not just for output but to let newcomers absorb tacit knowledge through the most basic work: the kind that can't be written into textbooks, only forged through hands-on experience. AI takes away these positions, effectively sawing off the bottom rungs of the ladder. You can still see the people at the top, but you can never climb up again.

On the day IBM crashed, its official response was: "New AI tools come out every week, including our own, but they won't change the fundamental engineering challenges of running critical systems at scale." Sounds calm and professional — except the market had already delivered its verdict with $31 billion in evaporation. IBM itself is pushing watsonx, whose goal is precisely to use AI for COBOL modernization. Selling AI's power with one hand while praying AI doesn't become too powerful with the other. Using your right hand to kill your left — the same script as Salesforce using Agentforce to kill its own per-seat model. The destiny of every middle-layer company can be summed up in one sentence: you must embrace the thing that kills you, or someone else will use it to kill you faster.

This isn't an invisible tax cut. It's a class cleansing. When "smart" becomes cheap, the companies controlling models and underlying compute become the new feudal lords, and most white-collar workers become digital serfs without bargaining power.

Who Adapted?

Kobeissi's article concludes: the disruption AI brings is real, but humans have adapted every time in history. The underestimated possibility isn't a "global intelligence crisis" but a "global intelligence boom."

He's right — humans have adapted every time. But he missed a crucial question: who adapted?

The Industrial Revolution put textile workers out of jobs. Fifty years later, their grandchildren did find new work in factories. But during those fifty years, an entire generation was ground to dust. They "adapted" too — adapted to the gears of the steam engine by losing their hands.

AI won't destroy the world. Kobeissi got that right. But he got the question wrong. The real threat was never whether AI would launch nuclear missiles. The real threat is that it's building a three-tier society with breathtaking efficiency: at the very top, the infrastructure lords who control compute and models; in the middle, a small group of people who can harness AI workflows; at the bottom, everyone else.

The fruits of efficiency do exist. You and I will probably only get to smell them.

And here we are, still applauding.

---

_(Data sources: Bloomberg, AMD/Meta official announcements, Economic Policy Institute, Forrester Research, Goldman Sachs, The Kobeissi Letter, Dallas Fed. Corrections welcome if any data errors are found.)_

_—Kinney's Wonderland_