NextFin News - Artificial intelligence is keeping investors willing to take risk, but the rally it fuels is narrowing rather than broadening, Shaun Cochran, head of research at CITIC CLSA, said in a television interview this week. His warning lands as the Nasdaq Composite closed at a record high on Monday, up 2.26% to 27,122.09, carried by a surge in AI-related chip stocks even as fewer than half of all US shares trade above their 50-day moving averages.
Cochran's message is a study in two speeds. On one side, artificial intelligence remains the single strongest engine of risk appetite in global markets: money keeps flowing into anything tied to AI infrastructure, and that flow is powerful enough to push the Nasdaq to its highest close since early June. On the other, the gains are concentrated in a thin slice of the market, and the question of how the companies at the center of the buildout will ultimately convert that spending into profits remains unanswered.
The setup: the Nasdaq Composite jumped 2.26% to 27,122.09 on September 21, its best single-day performance since an almost 2.6% advance on August 4 and its first record close since June 2. The S&P 500 climbed 1.49% to 7,764.70, ending within 0.4% of its all-time high of 7,798.99 set on August 13. The Dow Jones Industrial Average added 366.19 points, or 0.71%, to 52,048.83. Chip stocks led the advance: Intel rose 12%, Advanced Micro Devices gained about 10% and pushed past a $1 trillion market capitalization at an all-time high above $610 a share, and Qualcomm advanced more than 9%. The Philadelphia Semiconductor Index jumped 4.3%.
The move came after a difficult week. The Dow slid 1.7% over the prior five sessions, its worst performance since March, while the S&P 500 was roughly flat and only the tech-heavy Nasdaq managed a gain. The reversal was helped by falling Treasury yields and cooling oil prices, which eased pressure on rate-sensitive growth stocks. Brent crude and WTI futures both slipped below $100 a barrel on Monday on hopes that the United States and Iran would resume diplomatic talks, after Brent traded around $102 earlier in the month.
Against that backdrop, Cochran's comment cuts both ways. In his view, the rally's breadth is the tell.
"As long as markets remain focused on IT-oriented, AI-driven growth, the K-shaped phenomenon will continue. Markets are looking ahead to AI's potential, how companies ultimately make money from it remains unclear."
The K-shape is the crux: the upper arm of the K is the AI complex, rising on the promise of future earnings; the lower arm is everything else, lagging because the payoff has not yet shown up in cash flow.
The macro setting is not friendly to risk assets, which makes the AI bid all the more striking. The Federal Reserve raised its benchmark rate by 25 basis points at its September 16 meeting, a unanimous decision that reflected inflation running above the central bank's 2% target: the core personal consumption expenditures price index rose 3.3% in the 12 months through July, while the headline PCE index advanced 3.7%. The 10-year Treasury yield briefly topped 5% on September 15, its highest level since October 2023, touching 5.04%. In that environment, a record Nasdaq close is not a broad vote of confidence; it is a targeted bet on one theme.
What the K-Shaped Rally Actually Looks Like
The clearest evidence that this rally is narrow sits in the breadth data. As of the September 8 close, only 43.3% of US stocks traded above their 50-day moving average, while 51.4% were above their 200-day average, leaving a 50-day-to-200-day breadth ratio of 0.84 — near parity, a configuration that typically appears during a flush inside an uptrend rather than a broad advance. By September 18, the combined percentage of stocks above their 50-, 150- and 200-day averages had fallen to 39.86%, down 4.16 percentage points on the week and below the 40% level for the first time since early April. A separate gauge of S&P 500 stocks above their 50-day average stood at 46.91% on September 4, down 7.82% in a week.
Yet the Nasdaq hit a record. That divergence is the K-shape in one picture: index levels rising while participation shrinks. The beneficiaries are concentrated in the AI supply chain — semiconductors, data-center infrastructure, power and cooling — while the rest of the market trades below its short-term trend.
This is not the first time a dominant theme has produced a two-speed market. During the late-1990s technology boom, the Nasdaq roughly tripled between the start of 1998 and its March 2000 peak while the equal-weighted S&P 500 lagged badly; the spread between mega-cap growth and the median stock reached extremes that only resolved through a painful mean reversion. The difference today is that the leaders are selling shovels, not stories: the semiconductor and hyperscaler complex is posting real revenue from AI infrastructure, not dot-com-era page views. That gives the current cycle more fundamental ballast — but it does not answer Cochran's monetization question for the companies buying the equipment.
The concentration is visible in the leaders' own valuations. AMD's advance past $1 trillion in market capitalization on Monday put it in a club that, as of mid-2026, included only a handful of US companies — and it did so while trading on expectations of AI-driven data-center demand that has yet to be fully reflected in earnings. Intel's 12% jump, by contrast, was a recovery trade in a company still rebuilding its manufacturing roadmap. Both moves were AI-driven; neither rests on the same foundation. That is the K-shape within the K-shape.
Why AI Keeps Risk Appetite Alive Even With Rates at Restrictive Levels
The puzzle is not that AI stocks are rising; it is that they are rising while the cost of capital is climbing. The Fed just hiked into an inflation problem, the 10-year yield is at a 16-year high, and oil is back above $100. By conventional logic, higher real rates should compress the valuations of long-duration growth assets, because the present value of earnings far in the future falls as the discount rate rises.
The transmission mechanism runs through capital expenditure, not near-term earnings. Global AI-related investment is forecast to exceed $1 trillion in 2026, according to Goldman Sachs Research, with roughly $581 billion of that in the United States. A consensus of analyst estimates puts US hyperscaler capital expenditure at about $800 billion for the year, and Citigroup expects Amazon, Microsoft, Alphabet and Meta to spend more than $630 billion in 2026 alone. That spending is visible, contractible and front-loaded: chip orders, data-center construction and power infrastructure show up in revenue today, while the returns on that investment arrive years later.
For equity investors, the near-term revenue visibility is what keeps risk appetite alive. Every dollar of hyperscaler capex becomes someone's sales: the semiconductor foundry, the equipment maker, the power utility, the cooling provider. That chain is why a rate hike that would normally sink growth stocks instead coincided with a Nasdaq record — the market is pricing the capex cycle, not the terminal return on invested capital.
But the second-order problem is building. Cochran's point that "how companies ultimately make money from it remains unclear" is a demand-side warning disguised as a valuation note. If the enterprises and consumers who are supposed to pay for AI services do not generate enough incremental revenue to justify the infrastructure buildout, the capex cycle will slow, and the revenue currently supporting the AI complex will prove cyclical rather than structural. The capex is real today; the cash flow that must service it is still hypothetical.
The scale of the mismatch is worth measuring. If global AI investment reaches $1 trillion in 2026, the industry must eventually generate a return on that annual spend that exceeds its cost of capital — or the spending itself becomes the ceiling on growth rather than the engine. Citigroup's forecast of $8.9 trillion in cumulative AI capital expenditure between 2026 and 2030 implies that the buildout is only in its early innings. That is the bull case if monetization follows; it is the overcapacity case if it does not. The same dollar of capex cannot be both a revenue guarantee and a future write-down.
The Dollar, the Yen and the Cross-Currents in Cochran's View
Cochran's interview ranged beyond AI, and the other two themes reinforce the same caution about consensus positioning. On the US dollar, he has been contrarian bullish even as bearish calls became fashionable. "As credit conditions worsen and debt loads rise relative to sustainable cash flow, demand for high-quality collateral increases, will support demand for the dollar over time," he said.
On the yen, the setup is equally counterintuitive. In a tightening credit environment, Japan would normally see capital repatriation and upward pressure on its currency. Instead, the yen has stayed weak even as the Bank of Japan has moved to tighten policy and inflationary pressures have built. Cochran's base case is that the yen can find a floor and rally from here.
Both calls share a common thread with the AI view: the market is positioned for one outcome while the underlying mechanics point somewhere else. Just as investors have crowded into AI winners while breadth deteriorates, they have crowded into dollar weakness while credit conditions point toward collateral demand, and into yen weakness while repatriation logic points the other way. The pattern is not a bullish one for the consensus trade — it is a warning that the easy money has already been made.
The collateral argument deserves attention because it connects back to AI. A buildout financed with debt raises leverage across the technology complex; if credit conditions tighten faster than cash flows grow, the demand for high-quality collateral rises and the dollar strengthens — the exact opposite of the weak-dollar trade that has been crowded for much of 2026. Cochran's three themes are not separate; they are three angles on the same risk: positioning that has run ahead of the fundamentals that must eventually support it.
Conclusion: The Gap Between Spending Today and Profits Tomorrow
The central judgment: AI-driven risk appetite is real but structurally narrow, and it is being sustained by a capex cycle whose terminal returns are unproven. That makes the current rally more fragile than the record index levels suggest.
By time horizon, the picture splits. In the short term, sentiment and liquidity dominate: as long as hyperscaler capex guidance stays strong and yields stabilize below the levels that historically break risk assets, the AI trade can extend higher. In the medium term, fundamentals take over: the companies buying AI infrastructure must show that the spending is generating incremental profit, not just revenue. In the long term, the question is structural: whether AI is a general-purpose technology that lifts economy-wide productivity, or a capital-intensive boom that ends in overcapacity once the easy use cases are built out.
Neuberger Berman's third-quarter asset allocation outlook captures the bull case: it maintains an overweight to global equities on the view that the AI capital expenditure cycle is broadening beyond hyperscalers into utilities, power, industrials, semiconductors, copper and manufacturing supply chains, with credit spreads tight but not widening and real yields still below the levels that historically break risk assets. iShares' Fall 2026 investment outlook goes further, calling AI its highest-conviction US equity theme and explicitly rejecting comparisons to the late-1990s dot-com bubble, pointing to evidence that the AI opportunity is becoming broader rather than more zero-sum.
The strongest counter-thesis is straightforward: breadth this weak has rarely preceded a durable advance, and a 5% 10-year yield with the Fed still hiking is a hostile backdrop for extended valuations. If the market is wrong about AI monetization, the correction will not be confined to chip stocks — the capex chain runs through utilities, industrials and credit, and a slowdown would hit all of them. Neuberger Berman itself acknowledges the risk, noting that a moderate equity drawdown could happen quickly amid lighter trading volumes, mega-cap supply and election-related volatility.
The falsifying signal is specific. If the Nasdaq Composite closes below its prior record of 27,093.90 and the percentage of US stocks above their 50-day moving average stays below 40% for two consecutive weeks while the 10-year Treasury yield holds above 5%, Cochran's view that AI keeps risk appetite alive would be wrong: the theme would no longer be strong enough to offset deteriorating breadth and restrictive financial conditions.
What to watch: hyperscaler capex guidance in the next earnings cycle, the employment and inflation prints that will shape the Federal Reserve's next move, and whether the Nasdaq can hold its record while breadth improves. The base case is for continued AI-led strength with high volatility; the upside case is a broadening rally as yields fall and earnings validate the buildout; the downside case is a capex disappointment that turns the K-shape into a reversal.
The market is pricing AI's capital expenditure as if the returns were guaranteed. They are not — and that gap between spending today and profits tomorrow is where the next leg of risk either gets made or lost.
Explore more exclusive insights at nextfin.ai.
