Everything About AI
NextFin News - The most bearish voice on Wall Street just put a number on the end of the AI bull market. Joachim Klement, head of market strategy at Panmure Liberum, says the artificial-intelligence bubble will burst in 2027 or 2028 — "sometime in the next two years" — and he is cutting his S&P 500 year-end 2027 target to 5,000, roughly 36% below where the index trades today. The call is not a marginal downgrade. As recently as mid-September, Klement expected the index to reach 8,300 by the end of next year. That is a reversal of more than 3,000 points in a matter of weeks, and it frames the central tension now dividing the market: can AI-driven earnings keep funding the most expensive infrastructure buildout in history, or will rising borrowing costs expose the gap between spending and revenue before the rally runs out of road?
Klement's warning, delivered in an interview and set out in Panmure Liberum's October 2026 forecasts, makes him the most bearish of the eight strategists surveyed for the 2027 outlook. The others, on average, expect the S&P 500 to gain 14%. The distance between those two answers is the trade. And the timing is deliberate: "I'm starting to worry people today for something that I think might happen in six to nine months," he said.
The Situation: A Financing Squeeze at the Heart of the AI Boom
The bull market's engine is a handful of US technology giants. Amazon, Microsoft, Alphabet, Meta and Oracle are pouring capital into data centers at a pace Panmure Liberum put at $658 billion for 2026 in its May analysis, while independent research estimates data-center capital expenditure could more than double this year to $713 billion and rise again in 2027, albeit more slowly. To justify that outlay, Klement calculates the five companies would need to add between $2 trillion and $5 trillion in annual revenue — a gap that, in his view, is not closing fast enough.
The mechanism he is betting on is a financing squeeze, not a demand collapse. Hyperscaler free cash flow has been largely depleted, he argues, while the cost of the debt funding this buildout is climbing quickly enough to become prohibitive. The 10-year Treasury yield — the benchmark for corporate borrowing across the economy — reached 5.28% in early October, levels not seen in two decades. The 30-year bond touched its highest level since 2002. "If we keep seeing higher bond yields and interest rates, we think the end of the equity bull market may be closer than many investors think," Klement said.
The scale of what has been priced in is itself unusual. AI-related market gains have reached $28.6 trillion since April 2025, a 31% increase in just six months. Klement's May analysis argued the AI boom is already 60% larger than the late-1990s dotcom bubble, and that pace of appreciation has appeared only three times in modern market history: 1987, 1999 and 2009. The Shiller cyclically adjusted price-to-earnings ratio sits near 40 times earnings, well above its long-run average. For comparison, the S&P 500 trades at about 21.5 times forward earnings today, versus the 20-times multiple Jefferies uses for its bullish 2027 target — meaning the valuation defense rests on earnings delivery, not on paying more for the same profits.
Current market levels underscore the stakes. The S&P 500, via the SPDR S&P 500 ETF, was changing hands around 777 in early October, within about 1% of its 52-week high of roughly 782 and up about 23% from its 52-week low near 629. A 36% decline from here would wipe out the entire post-pandemic-era advance and then some. Klement is not forecasting that as his base case for tomorrow; he is pricing the distance between what the technology will earn and what investors have already paid for that promise.
Scenarios: A Pullback, or a Crash?
Klement does not present a single fate. In a milder scenario, a 4.5% to 6% pullback in technology spending would be enough to send the S&P 500 down 20%, with European and UK markets falling more than 30%. In the full dotcom-style crash case, he warns that technology hardware stocks could plunge by more than 70%.
His Europe target reflects the same thesis: the Stoxx 600 is expected to fall to 430 points, more than 30% below current levels. The bearish case is global because the financing channel is global — dollar funding costs feed through to every market that borrowed cheaply to chase the same trade. Notably, Klement remains the most bullish of the strategists surveyed for the Stoxx 600 through the end of 2026, predicting gains of about 10% in the regional benchmark. The bearish turn is a 2027 event, not a 2026 one, which is why he frames it as something to prepare for rather than to front-run.
For investors who want to follow rather than predict, Klement has named his trigger: a break below the S&P 500's 200-day moving average is the signal to turn defensive. He is urging clients to build contingency plans now rather than wait for the selloff to begin.
"My core conviction is that the AI bubble will either burst in 2027 or in 2028, so sometime in the next two years."
Why This Time May Actually Be Different — And Why It May Not
The first question any bubble call must answer is whether the driver is cyclical or structural. Here, both forces are present, and separating them matters — because getting this wrong flips the conclusion. The cyclical leg is the financing condition: free cash flow depleted, debt costs rising, capex guidance stretched. Cyclical financing squeezes mean-revert. Rates fall, refinancing windows reopen, and the companies with real assets survive. The structural leg is the technology itself: AI is not dial-up internet with no monetization path. It is already embedded in enterprise software, cloud contracts and productivity tools, and its adoption curve is real.
But a structural technology can still produce a cyclical valuation crash, and the dotcom era is the template. The Nasdaq Composite peaked at 5,048 in March 2000 and fell 78% by October 2002, taking 15 years to reclaim its high. The internet changed the world anyway. What burst was not the technology but the multiple attached to it — the assumption that revenue would arrive fast enough to service the capital raised against it. Klement's $2 trillion to $5 trillion revenue requirement is the arithmetic version of that point. Until AI income approaches that range, the capex boom rests on continued access to cheap capital, not on self-funding cash flows.
There is also a historical anchor for the valuation warning. The S&P 500's Shiller cyclically adjusted P/E near 40 times has been exceeded only once in more than a century of data — the 44.2 reading at the 1999-2000 dot-com peak. The September 1929 pre-crash high was about 32.6 times, well below today's level. Elevated valuation does not time the crash; it determines the severity once the trigger fires.
This is where the second-order channel matters, and where most commentary stops too early. The obvious first-order effect of an AI slowdown is lower earnings for chipmakers and cloud builders — Nvidia, AMD, TSMC and the hyperscalers themselves. The second-order effect runs through credit. A buildout this size was not funded entirely from cash; it was funded with debt, much of it issued into a market that priced technology as a one-way bet. The ICE BofA US high-yield option-adjusted spread sat at just 3.10 percentage points in early October — tight by historical standards, which is exactly why there is room for it to widen. If capex guidance rolls over, the first repricing happens in high-yield credit spreads, then in the equity risk premium, and only then does it show up fully in earnings. Credit leads equities in downturns because bondholders price default risk before shareholders price lower growth. A spread widening from 3.1 to 5 or 6 percentage points would be the early-warning siren that the financing squeeze has moved from theory to market fact.
The third-order effect is the expectation gap now embedded in the consensus. The market is not merely pricing strong AI earnings; it is excusing macro and credit headwinds because of them. "It is a situation where people are just focusing on one thing and one thing only, and that is earnings and in particular tech earnings," Klement said. "And they excuse every macro, credit or whatever headwind that you can come up with with that story." When a single narrative absorbs every contrary data point, the narrative is doing the work that evidence normally does. That is the definition of a crowded trade, and crowded trades unwind violently when the one thing everyone is focused on fails to deliver.
The Counter-Thesis: Earnings Could Still Save It
The strongest argument against Klement is the one the market is already making: earnings are real. Technology companies are posting record profits, cloud demand is verifiable in contract backlogs, and AI revenue, while small relative to capex, is growing. Citigroup strategists have said solid 2027 earnings could support further gains in global equities despite higher interest rates and geopolitical risks. Jefferies has projected the S&P 500 could reach 9,000 by the end of 2027 — nearly double Klement's target — on $450 of earnings per share and a 20-times multiple, with a bear case of 6,900 and a bull case of 10,500. Several brokerages expect the index to end 2026 at or above 8,000, with the bull market continuing into next year.
There is also a valuation defense: the S&P 500's forward price-to-earnings ratio, at roughly 21.5 times, is elevated but not at dotcom extremes, and the index is dominated by companies with actual cash flows, unlike the profitless issuers of 1999. If AI monetization accelerates faster than the skeptics expect — through software pricing power, automation-driven margin expansion, or enterprise adoption compounding — the revenue gap could narrow without a crisis. Jefferies' own downside scenario of 6,900 implies about a 25% correction, which concedes volatility without conceding a crash.
Klement's answer to that defense is timing and financing, not technology denial. He has acknowledged that strong corporate earnings and resilient economic data continue to support equities today. His point is that the support is conditional on refinancing staying available at tolerable cost. The counter-thesis wins only if capex guidance holds through the upcoming third-quarter earnings season and the 2027 forecasts due early next year without signs of financing stress. That is a testable condition, not a matter of faith — and it is why the next two earnings seasons matter more than the next two index levels.
What Would Prove Him Wrong
A judgment without a falsifying signal is a wish. Klement's thesis breaks if three observable conditions hold together: hyperscaler free cash flow stabilizes or grows through the next two earnings seasons; capital-expenditure guidance for 2027 is maintained or raised without a jump in borrowing costs; and the 10-year Treasury yield rolls back below roughly 4.5%, reopening cheap refinancing. Any one of those alone delays the bust; all three together would invalidate the financing-squeeze mechanism entirely.
The signal to watch in the other direction is the one Klement named — a sustained break below the S&P 500's 200-day moving average, confirmed on volume, would be the market telling investors that the defensive window is closing. A secondary signal would be high-yield spreads pushing above 4.5 percentage points, which would indicate the credit channel is activating before equities fully reprice.
Outlook: Three Horizons, Three Scenarios
Short term — the next one to two quarters — the market remains supported by earnings and resilient economic data. The S&P 500 is trading within roughly 1% of its 52-week high, and the trend is still up. Fighting that trend before the trigger fires is not the same as being right, and Klement himself concedes his call may be early. The near-term path of least resistance remains higher unless the 200-day line breaks.
Medium term — through 2027 — the base case is rising volatility rather than an immediate crash. Capex guidance becomes the scoreboard: quarterly earnings and forward spending plans from Amazon, Microsoft, Alphabet, Meta and Oracle will show whether the revenue gap is narrowing or widening. A 20% correction is the milder of Klement's scenarios and is consistent with a financing scare that does not become a credit crisis. If capex rolls over while debt costs stay elevated, the index tests the 6,000 to 6,500 range, and quality balance sheets separate from the leveraged chasers.
Long term — the structural question — AI adoption is real and will not reverse. The regime shift in computing infrastructure is durable, and the companies that emerge with defensible AI revenue streams will compound through the cycle. What is cyclical is the valuation premium attached to the theme and the ease of funding it. Investors can be structurally bullish on the technology and cyclically bearish on the stocks at current prices; the two views are not contradictory, and history suggests they are often both right at different horizons.
The base case: volatility rises into 2027, the 200-day moving average is tested, and the index grinds toward the low 6,000s if capex guidance disappoints. The upside case: AI revenue compounds faster than capex, the 10-year yield falls back toward 4.5%, and Jefferies' 9,000 target proves closer to the mark. The downside case: a financing shock hits while spending is still rising, credit spreads blow out past 5 percentage points, and Klement's 5,000 — or his hardware-stock crash scenario — comes into play.
The market is not being asked to bet against artificial intelligence. It is being asked to price the distance between what the technology will earn and what investors have already paid for that promise. On that measure, the burden of proof has shifted from the bears to the bulls — and the next two earnings seasons will decide who carries it.
Explore more exclusive insights at nextfin.ai.
