Record Markets, Depression-Era Sentiment: AI Divide

Markets Hit Records While Consumers Sink to Depression-Era Sentiment: The AI Rally's Growing Disconnect With the Real Economy

I've been tracking this market-sentiment divergence in real time, and the data keeps getting more extreme. the Dow Jones Industrial Average closed at 50,579.7 on Friday, another all-time high. The S&P 500 has now risen for eight consecutive weeks — a streak not seen since 2004. The KOSPI is trading above 7,800 with Nomura Securities publishing a 10,000-11,000 target range. By almost any measure of equity market performance, mid-2026 is a spectacular time to own stocks.

Except that it isn't, if you look at the people living in the economy rather than the institutions trading it. The University of Michigan Consumer Sentiment Index fell to 44.8 in May — a level that in every prior cycle has been associated with either a recession in progress or one imminent. For context, the index bottomed at 55.3 during the 2008 financial crisis. It hit 51.7 in the depths of the 1980 recession. The current reading of 44.8 is worse than both. US consumers are more pessimistic now than they were when unemployment hit 10% in 2009.

I think this is the single most important chart in global markets right now, and here's why. this gap between asset prices and economic sentiment is not merely unusual. It is, by the available data, the widest divergence since the University of Michigan began its survey in 1952. Something has to give. The question is what.

The AI Semiconductor Engine That Is Powering Everything

The bull case, articulated most clearly by Goldman Sachs' US equity strategy team, is that this time the divergence is rational. "The S&P 500 is not the economy. It is increasingly a collection of AI-exposed technology companies whose revenue growth is decoupled from US GDP," wrote David Kostin, Goldman's chief US equity strategist, in a May 19 note. "Nvidia's revenue does not depend on whether a Michigan household feels optimistic. It depends on whether Microsoft, Amazon, Google, and Meta continue to build out AI infrastructure. And they are — capex guidance for the hyperscalers in 2026 totals $320 billion, up 42% year-over-year."

The numbers support this argument — for now. Nvidia's market capitalization has crossed $4 trillion. The company reported Q1 fiscal 2027 revenue of $48.2 billion, up 52% year-over-year, with data center revenue alone accounting for $41.1 billion. Gross margins are 76%. TSMC, which manufactures the chips Nvidia designs, posted Q1 revenue of $24.1 billion and guided for 35% full-year growth. Samsung Electronics and SK Hynix, the twin pillars of Korea's AI semiconductor exposure, are both running near full capacity on their high-bandwidth memory lines.

The AI capex cycle has not peaked. If anything, it is accelerating. But there is an uncomfortable pattern here: every major technological investment cycle in modern history — railroads in the 1870s, fiber optics in the late 1990s, shale oil in the early 2010s — has followed the same trajectory. Massive investment. Overbuilding. Consolidation. The companies that sell the picks and shovels outperform until suddenly they don't.

My view on the semiconductor cycle aligns with Kim's assessment, but I'd add one caveat. "Semiconductor cycles don't die of old age. They die of overcapacity," said Kim Dong-won, a veteran semiconductor analyst at KB Securities, in a May 21 interview. "Right now demand is exceeding supply for HBM and advanced logic chips. But Samsung, SK Hynix, Micron, and TSMC are collectively investing over $200 billion in new capacity. In 18 to 24 months, that capacity comes online. The question every semiconductor investor should be asking is not whether AI demand will be strong in 2026 — it will be — but whether it will be strong enough in 2028 to absorb all the capacity being built right now."

The Real Economy: What 44.8 Consumer Sentiment Actually Means

Consumer sentiment at 44.8 is not just a statistic. It translates into behavior. When sentiment is this low, households delay major purchases. They draw down savings rather than take on new debt. They trade down in quality and brand. The US personal savings rate, which fell to a historic low of 2.1% in mid-2025 as households spent down pandemic-era savings, has rebounded to 4.7% — a sign that precautionary saving is returning.

Retail sales data for April showed a 0.4% month-over-month decline, with discretionary categories — furniture, electronics, clothing — falling more sharply than staples. Credit card delinquency rates have risen to 3.6% at major US banks, the highest since 2011, according to Federal Reserve data. Auto loan delinquencies are at their highest in 13 years. These are not recession indicators yet, but they are deterioration indicators, and the slope of deterioration has steepened in 2026.

In Europe, the picture is worse. The Eurozone manufacturing PMI has been in contraction territory (below 50) for 14 consecutive months at 46.7. Germany, the bloc's industrial engine, is grappling with energy costs that remain 40% above pre-2022 levels and competition from Chinese electric vehicle exports. In China, GDP grew 4.6% year-over-year in Q1 2026, but property sector investment fell 11% and consumer confidence indices remain deeply negative.

Korea sits somewhere between the two narratives. Exports rose 8.3% year-over-year in April, driven almost entirely by semiconductors — chip exports were up 42% while non-chip exports were flat. The domestic economy is sluggish: retail sales, construction investment, and small-business sentiment all registered declines in the latest Bank of Korea data. The export engine is running hot. The rest of the car is idling.

The Geopolitical Overlay: Iran and the Risk Premium

The US-Iran conflict that erupted in early 2026 has receded from front pages but remains the most significant geopolitical risk in global markets. Brent crude is trading at $82 per barrel, with analysts at RBC Capital Markets estimating a $5-8 geopolitical risk premium embedded in the price. If the conflict de-escalates — a ceasefire, formal negotiations — oil could drop to the mid-$70s, providing a disinflationary impulse that would strengthen the case for Fed rate cuts. If it escalates — a disruption to Strait of Hormuz shipping — oil could spike above $100, triggering a global supply shock that would make the Fed's rate path significantly more hawkish.

For Korea, which imports nearly all of its energy, the difference between $75 oil and $100 oil is roughly $35 billion in additional annual import costs, or about 2% of GDP. The won, already under pressure from the rate differential, would face another leg of depreciation if oil spikes. Korea's energy-intensive industrial structure — semiconductors, steel, petrochemicals, shipbuilding — makes it more sensitive to energy prices than most developed economies.

The KOSPI at 8,000: Cheap on Paper, Expensive in Context

Nomura's 10,000-11,000 target for the KOSPI is based on a simple argument: Korean stocks are cheap relative to global peers on every standard valuation metric, and the AI cycle will close a significant portion of the Korea Discount as the market recognizes Korea's essential role in the global semiconductor supply chain. The forward P/E of the KOSPI is 11.2x, compared to 22.5x for the S&P 500 and 16.8x for the Nikkei 225. The PBR of 1.05x is less than a quarter of the S&P 500's 4.8x.

"If Korea were an emerging market with emerging market governance standards, these discounts would be appropriate," wrote Nomura strategist C.W. Chung in his May 22 report. "But Samsung Electronics and SK Hynix are technology leaders with global brand recognition, state-of-the-art manufacturing, and improving shareholder return policies. The market is pricing them like low-growth industrial companies. We think that is wrong."

The counterargument is that the market has been pricing Korea this way for 20 years and has been right for most of that period. The Korea Discount has survived bull markets, governance reforms, dividend increases, and multiple government campaigns. It is not a pricing error. It is a structural feature of a market where controlling families prioritize control over returns, where conglomerates cross-subsidize weak units from strong ones, and where minority shareholder rights remain limited despite legislative progress.

Foreign ownership of the KOSPI has fallen from 36.8% in 2021 to 32.1% in May 2026. Global investors are not, in aggregate, buying the Korea convergence story. They are buying Samsung and SK Hynix as AI pure-plays and ignoring the rest. Until this changes, the KOSPI at 8,000 is as much a reflection of narrow AI enthusiasm as it is of a genuine re-rating of Korea's equity market.

My Take

I've been watching the AI rally versus consumer sentiment divergence since it first appeared in Q4 2025, and I think we're now at the point where one of these two data series is going to break. Either the S&P 500 corrects to reflect the economic reality of 44.8 consumer sentiment, or consumer sentiment improves as the AI-driven productivity gains begin flowing into wages and employment. My bet is on the former — at least in the short term.

Here's my reasoning: the AI capex supercycle is real, but the market is pricing it as if the revenue growth for Nvidia, TSMC, and the hyperscalers will continue at current rates indefinitely. History tells us that every technology investment cycle — railroads, telecoms, the 1990s internet — goes through an overcapacity phase where supply catches up with demand and the marginal return on capital collapses. We're not there yet for AI, but the Nvidia forward P/E of 35x assumes a very long runway of 50%+ growth. I think the probability of an AI capex disappointment before year-end is higher than the market is pricing.

My view on KOSPI in this context: the index at 8,000 with a 12x forward P/E looks cheap on the surface, but that's a Korea-specific discount reflecting the chaebol governance discount, the geopolitical risk premium, and the demographic headwind that no amount of semiconductor revenue can fix. I don't buy Nomura's 10,000–11,000 target for 2026. I think KOSPI ends the year closer to 8,500–9,000, driven by genuine earnings growth from Samsung and SK Hynix, but with a significant correction in between as global sentiment sours. The time to buy Korean equities will be when the consumer sentiment index bottoms and starts recovering — not when it's still falling. We're not at that point yet.

What This Means for Global Portfolios

The AI rally is real, the earnings are real, and the capex is real. But markets priced for perfection are fragile. The consumer sentiment reading of 44.8 is a warning that the prosperity visible in equity indices is not widely shared — and when prosperity is narrow, the political and policy responses to economic pain can be unpredictable. Tariffs, capital controls, windfall taxes, industrial policy interventions — these are the tools governments reach for when voters are hurting and markets are celebrating.

Three scenarios for the second half of 2026:

Base case (50% probability): The AI rally continues but narrows further — money concentrates in the five or six semiconductor and hyperscaler names that have genuine AI earnings. Broader indices (S&P equal-weight, Russell 2000, KOSPI ex-Samsung) underperform. Consumer sentiment stabilizes in the high 40s without triggering a recession. The Fed cuts once, providing a relief rally that is sold into. KOSPI trades 7,800-8,500.

Bull case (25% probability): The US-Iran conflict de-escalates. Oil drops to $72. The Fed cuts twice. Consumer sentiment rebounds above 60 as inflation eases and rate relief flows through to mortgages and auto loans. The rally broadens from semiconductors to cyclicals and small-caps. Korea benefits disproportionately as the risk-on trade returns to emerging markets. KOSPI hits Nomura's 10,000 target by Q1 2027.

Bear case (25% probability): Consumer sentiment below 45 triggers an actual consumer recession — spending contracts, corporate earnings miss, the AI capex cycle enters a "digestion phase" where hyperscalers slow investment to assess returns on the $320 billion already spent. The AI trade unwinds violently. Nvidia corrects 30%. Samsung and SK Hynix correct 25% on memory oversupply fears. KOSPI drops to 6,500 as the narrow foundation of the entire rally crumbles. The Korea Discount widens to crisis levels.

The AI Capex Supercycle: How Long Can It Run?

The numbers behind the AI infrastructure buildout are staggering in absolute terms and sobering in relative ones. The four major US hyperscalers — Microsoft, Amazon, Google, and Meta — have guided for combined capital expenditures of $320 billion in 2026, up from $225 billion in 2025 and $148 billion in 2024. Semiconductor equipment makers — ASML, Applied Materials, Lam Research, Tokyo Electron — are running at near-maximum output. TSMC's Arizona fab, Samsung's Taylor, Texas fab, and Intel's Ohio complex are all under construction simultaneously, representing a combined investment of over $200 billion.

Gartner projects global semiconductor revenue will reach $700 billion in 2026, with AI-related chips — GPUs, HBM, custom ASICs, networking silicon — accounting for roughly $230 billion, or 33% of the total. By 2028, the AI chip market could reach $400 billion, or nearly half of all semiconductor revenue. These are the projections that justify current valuations. But they depend on an assumption that AI demand — not just AI hype — compounds at 40-50% annually for multiple years.

The historical parallel that keeps semiconductor veterans awake at night is the 2000 telecom bubble. Between 1996 and 2000, global telecom companies spent roughly $4 trillion on fiber optic networks (in today's dollars), driven by the conviction that internet traffic would grow exponentially forever. The traffic did grow — it grew enormously. But the capacity that was built exceeded demand by a factor of roughly 20x. It took a decade to absorb the overbuild. Companies that manufactured telecom equipment — Nortel, Lucent, JDS Uniphase — saw their stock prices decline by 95% or more, not because the internet thesis was wrong, but because the investment cycle overshot.

"We are not at the 2000 telecom bubble stage yet," said Daniel Newman, CEO of Futurum Research, in a May 20 Bloomberg interview. "The difference is that AI demand today is supply-constrained. Nvidia cannot make enough GPUs. TSMC cannot make enough advanced packaging. The hyperscalers cannot build data centers fast enough. As long as demand exceeds supply, the capex is rational. The risk is that the supply response — all those fabs being built right now — creates the overcapacity that bridges the gap too quickly and too late."

Korean Households and the Stock Market: A Fractured Relationship

One of the more unusual features of the current rally is the participation — or lack thereof — from Korean retail investors. Historically, Korean retail has been an aggressively pro-cyclical force, piling into rallies and amplifying both the upside and the downside. In the 2020-2021 bull market, Korean retail investors were the single largest buyer of KOSPI stocks, accounting for nearly 30% of total trading volume. In 2026, their share has fallen to roughly 18%, according to Korea Exchange data.

The reason is straightforward: Korean households are stretched. The combination of high interest rates, elevated household debt, and stagnant real wages has depleted the financial cushion that retail investors used to deploy into stocks during previous rallies. A survey by the Korea Financial Investors Protection Foundation found that 62% of retail investors described their current financial situation as "worse than a year ago," and 48% said they had reduced or eliminated new stock investments in 2026. The retail crowd that powered the KOSPI to 3,300 in 2021 is largely sitting this one out.

This changes the character of the rally in important ways. Retail-driven rallies tend to be broader, reaching into small and mid-cap names that institutional investors ignore. Institution-driven rallies, like the current one, are narrower and more fragile — concentrated in the liquid, large-cap names that funds can enter and exit without moving the market. If institutional sentiment turns, there is no retail bid to cushion the fall.

🔍 Related Keywords

• AI semiconductor rally consumer sentiment divergence 2026
• KOSPI 8000 Nomura target 10000 valuation
• US consumer sentiment 44.8 recession signal
• Samsung SK Hynix HBM demand AI capex cycle
• Korea Discount foreign ownership structural reform