Wages Draw a C, But Assets Still Draw a K

The US Treasury Secretary says he’s sick of the letter K

In early August, US Treasury Secretary Scott Bessent said in a CNBC interview that he was tired of hearing about a “K-shaped economy”1 and declared that phase over. The US, he said, is now heading toward a “C-shaped economy” — shorthand for low-wage workers’ earnings finally starting to catch up with those at the top.

On September 26, Taiwan’s Commercial Times (工商時報) ran an editorial that took Bessent’s comment as its starting point. The headline: “In the AI era, is society heading toward K, C, or I?” The argument goes like this. If AI ends up replacing not just low-skill jobs but high-paying white-collar work too, the wage gap could actually narrow. But AI assets — models, compute, data centers, chips, platforms, equity stakes in AI companies — are concentrated in the hands of a small number of large corporations and capital holders. So you could get an “asset-K” scenario where wages flatten out even as asset ownership keeps diverging. And if that trend deepens, you could end up with an “I-shape,” where an extremely small group occupies the single point at the top and the middle thins out entirely. Conversely, if ordinary people don’t just use AI but also come to share in AI assets and corporate profits, you get a “C-shape” where the gap narrows instead.

It’s a tidy framework, three letters doing a lot of work. But hold that framework up against the actual numbers, and one thing jumps out right away. The wage chart can look like a C or a K depending on which data you pick. The asset chart looks like a K no matter which data you pick.

---

Same Wage, Different Letter

Bessent’s evidence was data from the U.S. Bureau of Labor Statistics (BLS). Wages for the bottom 25% of full-time workers rose 5.5% over the past year—three times the rate for the top 25%. That number isn’t wrong. In the BLS statistics that look only at full-time wage earners, that’s indeed what you get.

But switch to the Federal Reserve Bank of Atlanta’s wage tracker, and the picture changes. That index includes part-time workers and looks at the 12-month moving average of hourly wages. As of June, the bottom 25% rose 3.6%, while the top 25% rose 3.9%. Fortune pointed out that by this measure, the bottom quartile’s wage growth has not once overtaken the top quartile’s since the start of 2026.

Same country, same labor market—yet the letter changes because the denominator does. Count only full-time workers, and you see a C, with the bottom catching up. Add in part-time workers, and it’s still a K. Even NPR, which had been skeptical of the C-shape claim, conceded that on an inflation-adjusted weekly-earnings basis, there’s at least one measure where the bottom rises and the top falls—and on that measure, you could call it a C. Wages are, by nature, numbers that change shape depending on whom you count. Bessent’s figure isn’t false. He simply picked the denominator, among several available, that shows a C.

The Asset Graph Isn’t Confusing at All

Assets are a different story. The US Federal Reserve’s Distributional Financial Accounts (DFA)2 break down household assets by wealth rank every quarter. Here’s what the Q2 2026 numbers look like.

Wealth Percentile	Share of Stock & Mutual Fund Holdings	Share of Net Worth
Top 1%	50.9%	32.5%
Top 1–10%	37.2%	36.4%
Top 10–50%	11.4%	28.8%
Bottom 50%	0.6%	2.3%

The top 10% hold 88% of all stock directly owned by US households, while the bottom half’s share is 0.6%. The top 1%‘s share of net worth has climbed from 22.8% in Q3 1989 to 32.5% today.

This table doesn’t isolate AI-related stocks specifically. Still, when AI companies’ valuations rise, which tier of households the resulting paper gains accrue to first has to follow this same distribution.

One caveat is worth flagging. Stocks held inside retirement accounts show up in this table under a separate line item called “pension entitlements,” so they’re not included in that 0.6% figure. Would the picture change if we counted indirect holdings too? In that category, the bottom 50%‘s share stood at 3.2% as of Q1 2022, the last period for which this figure was published. In other words, opening up one more channel doesn’t change the picture much.

This is where the core argument of the Gongsang Times (a fictional financial newspaper) editorial gains traction. The biggest winner from AI might not be the people who use AI, but the people who own it. The editorial points to economies of scale as the reason. Only companies that can spend tens of billions of dollars on data centers can compete on the same terms, and once platforms with large user bases start a flywheel where more data makes better models, a handful of giant firms tend to end up holding the field. Wages, meanwhile, could actually flatten out as AI takes over white-collar work. Yet productivity gains get captured as shareholders’ paper profits well before they ever show up in wages. That’s why the two graphs tracing different letters isn’t a contradiction at all.

Type C Wasn’t a Gift from Technology

So is Type C just wishful thinking? The editorial argues there’s precedent: divergence first, convergence later, in the Industrial Revolution. And this claim holds up against the numbers.

Economic historian Robert Allen found that in Britain between 1780 and 1840, output per worker rose 46%, while real wages rose only 12%. Because this is the exact period Engels described in The Condition of the Working Class in England, Allen named this stretch the “Engels’ pause”3. The following 60 years told a different story: between 1840 and 1900, output per capita rose 90%, while real wages rose 123%.

What’s worth noting here is the timeframe. The divergence phase alone lasted 60 years — long enough to span an entire working life. And economic historians still debate what actually drove the eventual convergence. There’s no consensus on how much credit belongs to capital accumulation, demographic shifts, or institutional change, respectively.

So the mere fact that “it converged eventually” offers no real comfort. To borrow the editorial’s framing, Type C isn’t a gift technology hands to society — it’s the outcome of institutional choices. [Editorial suggestion — author to confirm] And the cost of delaying that choice is usually paid, in full, by a single generation.

Korea Already Has a Pipe Like This

The editorial names the pathway to Type C as ordinary people becoming shareholders in AI assets and corporate profits. So what’s the route by which an ordinary person actually touches AI assets? In Korea this year, two such routes were put to the test almost simultaneously.

One is direct investment. As I covered in Issue 209 last August, 92% of the net buying in single-stock leveraged ETFs — permitted starting in late May — came from individual investors through June 22. Citi Securities’ late-July report estimated that individuals had lost about ₩56 trillion (~$40.3B) on this product. That’s not realized loss but mark-to-market loss from the peak. When ordinary people tried to directly own AI assets, the door they walked through most often was the one with debt attached.

The other route is the National Pension Service. The National Pension fund stood at ₩1,866 trillion (~$1.34T) as of end-June, with a first-half return of 27.22%. Looking only at domestic equities, the return was 107.37%. Since domestic equities make up 29.1% of the fund, a simple calculation puts over ₩540 trillion (~$388.5B) in domestic stocks centered on large-cap semiconductor names. The National Pension Service is also a shareholder in SK Hynix, holding a 7.50% stake as of the end of the first quarter. Without any individual subscriber picking a stock, and without anyone taking on debt, a share of the AI boom has flowed into the subscribers’ collective basket.

The reforms the Gongsang Shibo editorial names as the road to Type C — education, social security, institutional reform — sound like pipes that still need to be built. But Korea already has a pipe laid down, one that most income-earning citizens are already connected to. The problem is where that pipe lets out. Pension benefit amounts aren’t linked to investment returns — they’re set by statutory formula. The returns instead work to delay the point at which the reserve fund runs dry. That’s a long way from what subscribers actually feel in their bank accounts. What’s more, this pipe moves in lockstep with the ups and downs of the domestic stock market — and the fact that Korean equities took a sharp correction starting in July, right after the first-half results came in, shows up in that same shared basket.

Oswarld’s Lens

In issue 112 last June, which covered the gap in Samsung Electronics’ performance bonuses, I wrote that the hardest tension to resolve is the one between where value is created and where value is captured. I also argued that the design challenge isn’t picking one correct answer among a national dividend, a sovereign wealth fund, or a solidarity wage system—it’s figuring out the right mix of all three.

Having read the Gongsang Sibo editorial, I think that question needs to be pushed back one step further. Before asking how to divide the pie, we first need to map how many pipes connect ordinary people to AI assets, and where each one leaks along the way. Direct investment looks like it has the lowest barrier to entry, but it’s been the most expensive door for whoever arrives last. The National Pension Service is the widest pipe, but its outlet opens onto a distant future. Any discussion of a sovereign wealth fund only becomes comparable once it’s placed on this same pipe map.

There’s one more thing. The list of AI assets the editorial names includes models. As I’ve said before, not every task requires a top-tier frontier model. There’s a model suited to each job and each environment. As more small and medium-sized businesses and public institutions start running open-weight models on their own servers, at least one category of AI asset—the model itself—stops being confined to the hands of a few companies. The path to a Type C outcome may not run through distribution policy alone.

Closing

Look only at wages, and there’s room to read this as a C. Look at assets, and K comes through clearly. Bessent’s C and Gongsang Sibo’s K were looking at two different axes of the same economy. When distribution debates over the AI era talk past each other, the conversation gets a lot shorter once you check whether the other person is looking at the wage graph or the asset graph.

💬 Reader, how do you currently hold a stake in AI companies’ gains — direct investment, a retirement pension, Korea’s National Pension Service, or none yet? Feel free to answer with whichever applies.

📨 If you have a colleague who’s been debating post-AI distribution, pass this piece along.

References & Further Reading

• Commercial Times (工商時報) editorial, “Will the AI Era Lead to a K-Shaped, C-Shaped, or I-Shaped Society?” (工商時報 社論), 2026.9.26. ··· This is where today’s piece starts. Both the K-shaped and I-shaped asset scenarios come from here.
• Fortune, “Scott Bessent is ‘sick of hearing about’ the K-shaped economy and declares it’s over”, 2026.8.5. ··· Bessent’s remarks and the Atlanta Fed’s wage-by-income-bracket figures.
• PBS NewsHour, “Bessent said the K-shaped economy ‘is over.’ Here’s what financial experts say”, 2026.8.12. ··· The 5.5% BLS figure Bessent cited, plus experts’ pushback.
• NPR Planet Money, “Has the economy gone C-shaped?”, 2026.9.8. ··· The counter-evidence: one indicator that looks C-shaped.
• Federal Reserve, “Distributional Financial Accounts”, as of Q2 2026. ··· The raw data behind the table. The detailed time series can be checked on FRED.
• Robert C. Allen, “Engels’ pause: Technical change, capital accumulation, and inequality in the British industrial revolution”, Explorations in Economic History, 2009. ··· The source of the 1780–1840 vs. 1840–1900 comparison figures.
• SBS Biz, “National Pension Service’s H1 fund hits ₩1,866 trillion (~$1.3 trillion), return rate 27.2%—an all-time high”, 2026.8.28. ··· The National Pension Service’s H1 return rate and asset allocation.
• Insight Korea, “The National Pension Service’s record profits—loading up on SK Hynix while trimming ‘Naver-Kakao’”, 2026.9. ··· The change in the National Pension Service’s SK Hynix stake.

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

1. K-shaped economy: A term likening the way incomes or assets split during the same economic cycle—upper brackets rising, lower brackets falling—to the two diverging strokes of the letter K. A C-shape describes the bottom catching up to the top and the gap narrowing; an I-shape describes an extreme concentration at the very top with a hollowed-out middle. ↩
2. Distributional Financial Accounts (DFA): A U.S. Federal Reserve statistic that combines national household balance-sheet data with the Survey of Consumer Finances (conducted every three years) to estimate, each quarter, how much of which assets each wealth bracket holds. The time series goes back to 1989. ↩
3. Engels’ pause: A term for the period in early industrial-revolution Britain when productivity rose rapidly while workers’ real wages stayed nearly flat for decades. The name was coined by economic historian Robert Allen. ↩